Pass `Bindspace` FD to `trio` children

Thread an optional live `Bindspace` through actor spawn APIs and
give each Trio exec child an inherited namespace descriptor before
runtime bootstrap.

Deats,
- duplicate the namespace FD without changing parent ownership
- preserve caller `pass_fds` and process options
- send the `(fd, inode)` pair through the `_child` CLI
- close the parent duplicate on success, failure and cancellation
- reject MP backends until descriptor reduction is implemented
- avoid cleanup before child publication
- exercise real E2E namespace relay through handshake and RPC
- verify parent FD-table cleanup after successful child spawn

Prompt-IO: ai/prompt-io/opencode/20260828T172943Z_2ca8c570_prompt_io.md

(this patch was generated in some part by `opencode` using `gpt-5.6-sol` (`openai`))

.claude/ai_notes/docs_todos.md

@@ -0,0 +1,38 @@
+# Docs TODOs
+
+## Auto-sync README code examples with source
+
+The `docs/README.rst` has inline code blocks that
+duplicate actual example files (e.g.
+`examples/infected_asyncio_echo_server.py`). Every time
+the public API changes we have to manually sync both.
+
+Sphinx's `literalinclude` directive can pull code directly
+from source files:
+
+```rst
+.. literalinclude:: ../examples/infected_asyncio_echo_server.py
+   :language: python
+   :caption: examples/infected_asyncio_echo_server.py
+```
+
+Or to include only a specific function/section:
+
+```rst
+.. literalinclude:: ../examples/infected_asyncio_echo_server.py
+   :language: python
+   :pyobject: aio_echo_server
+```
+
+This way the docs always reflect the actual code without
+manual syncing.
+
+### Considerations
+- `README.rst` is also rendered on GitHub/PyPI which do
+  NOT support `literalinclude` - so we'd need a build
+  step or a separate `_sphinx_readme.rst` (which already
+  exists at `docs/github_readme/_sphinx_readme.rst`).
+- Could use a pre-commit hook or CI step to extract code
+  from examples into the README for GitHub rendering.
+- Another option: `sphinx-autodoc` style approach where
+  docstrings from the actual module are pulled in.

.claude/notes/rt_vars_lift_plan.md

@@ -0,0 +1,125 @@
+# `RuntimeVars` env-var lift — design plan
+
+Status: **draft, awaiting user edits**
+
+## Goal
+
+Consolidate the sprawl of pytest CLI flags + ad-hoc env vars +
+hardcoded fixture defaults into a *single* env-var-encoded
+runtime-vars envelope, with a typed in-memory representation
+(`tractor.runtime._state.RuntimeVars`) as the sole source of
+truth.
+
+## Why now
+
+- `--tpt-proto`, `--spawn-backend`, `--diag-on-hang`,
+  `--diag-capture-delay` and (soon) `TRACTOR_REG_ADDR` etc. are
+  proliferating. Each adds a parsing seam.
+- `tests/devx/test_debugger.py` invokes example scripts as
+  separate subprocesses; they currently can't see the
+  fixture-allocated `reg_addr` at all (root cause of why
+  parametrizing devx scripts on `reg_addr` is on your TODO).
+- Concurrent pytest sessions on the same host collide on
+  shared defaults (the `registry@1616` race we just fixed is
+  one symptom; per-session unique addr is the structural
+  fix).
+- `tractor.runtime._state.RuntimeVars: Struct` is already
+  defined and **unused** — its docstring even says it
+  "should be utilized as possible for future calls."
+
+## Design
+
+### Module: `tractor/_testing/_rtvars.py`
+
+Lifted from `modden.runtime.env`, ~50 LOC, no new deps.
+
+```python
+_TRACTOR_RT_VARS_OSENV: str = '_TRACTOR_RT_VARS'
+
+def dump_rtvars(rtvars: RuntimeVars|dict) -> tuple[str, str]:
+    '''str-serialize via `str(dict)` — ast.literal_eval-able'''
+
+def load_rtvars(env: dict) -> RuntimeVars:
+    '''ast.literal_eval the env-var value, hydrate to struct'''
+
+def get_rtvars(proc: psutil.Process|None = None) -> RuntimeVars:
+    '''read the var from a target proc's env (or current)'''
+
+def update_rtvars(
+    rtvars: RuntimeVars|dict|None = None,
+    update_osenv: bool|dict = True,
+) -> tuple[str, str]:
+    '''mutate + re-encode + (optionally) write to os.environ'''
+```
+
+### Encoding choice: `str(dict)` + `ast.literal_eval`
+
+Pros:
+- stdlib only
+- handles all the types tractor's tests need: `str`, `int`,
+  `float`, `bool`, `None`, `list`, `tuple`, `dict`
+- human-readable in the env (greppable, inspectable via
+  `cat /proc/<pid>/environ | tr '\0' '\n'`)
+
+Cons:
+- non-stdlib types (msgspec Structs, `Path`, custom classes)
+  must be lowered first — fine for the test fixture set
+- not stable across Python versions for esoteric repr cases
+  (we don't hit any)
+
+Alternatives considered:
+- **msgpack**: adds a dep + binary form is ungreppable
+- **json**: doesn't preserve tuples (becomes lists), which is
+  a common type for `reg_addr`
+- **toml/yaml**: heavier deps, no real benefit
+
+### `RuntimeVars` becomes the single source of truth
+
+The legacy `_runtime_vars: dict[str, Any]` global in
+`runtime/_state.py` becomes a *cached view* of a
+`RuntimeVars` singleton instance:
+
+- `get_runtime_vars()` returns either the struct or a
+  `.to_dict()` view depending on caller's preference
+- `set_runtime_vars(...)` validates against the struct schema
+- spawn-time SpawnSpec sends the struct (already does
+  conceptually — just gets typed)
+- `__setattr__` `breakpoint()` debug instrumentation gets
+  removed (unrelated cleanup, mentioned in conversation)
+
+### Migration path
+
+**Phase 0** *(prep)*: strip the stray `breakpoint()` from
+`RuntimeVars.__setattr__`.
+
+**Phase 1**: land `_rtvars.py` as a leaf module, used only by
+test infra. Subprocess-spawned scripts in `tests/devx/`
+read `_TRACTOR_RT_VARS` on startup → reconstruct
+`RuntimeVars` → call `tractor.open_root_actor(**rtvars.as_kwargs())`.
+Concurrent runs become deterministic-isolated because each
+session writes a unique `_registry_addrs` into the env.
+
+**Phase 2**: migrate runtime callers (`_state.get_runtime_vars`,
+spawn `SpawnSpec`, `Actor.async_main`) to operate on the
+struct directly, with the dict as a compat view that gets
+deprecated.
+
+**Phase 3** *(structural)*: per-session bindspace subdir
+`/run/user/<uid>/tractor/<session_uuid>/` — encoded in the
+rt-vars envelope, picked up by every subactor automatically.
+Obsoletes the entire bindspace-leak warning class.
+
+## Open design questions (user input wanted)
+
+- (placeholder for your edits)
+- (placeholder)
+- (placeholder)
+
+## Out-of-scope for this lift
+
+- Anything in `modden.runtime.env` related to `Spawn`,
+  `WmCtl`, `Wks` — that's a workspace orchestration layer,
+  not an env-var helper. We only lift the four utility
+  functions + the var name constant.
+- Switching to msgpack/json — explicitly chosen against
+  above.

.claude/settings.local.json

@@ -0,0 +1,42 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(cp .claude/*)",
+      "Read(.claude/**)",
+      "Read(.claude/skills/run-tests/**)",
+      "Write(.claude/**/*commit_msg*)",
+      "Write(.claude/git_commit_msg_LATEST.md)",
+      "Skill(run-tests)",
+      "Skill(close-wkt)",
+      "Skill(open-wkt)",
+      "Skill(prompt-io)",
+      "Bash(date *)",
+      "Bash(git diff *)",
+      "Bash(git log *)",
+      "Bash(git status)",
+      "Bash(git remote:*)",
+      "Bash(git stash:*)",
+      "Bash(git mv:*)",
+      "Bash(git rev-parse:*)",
+      "Bash(test:*)",
+      "Bash(ls:*)",
+      "Bash(grep:*)",
+      "Bash(find:*)",
+      "Bash(ln:*)",
+      "Bash(cat:*)",
+      "Bash(mkdir:*)",
+      "Bash(gh pr:*)",
+      "Bash(gh api:*)",
+      "Bash(gh issue:*)",
+      "Bash(UV_PROJECT_ENVIRONMENT=py* uv sync:*)",
+      "Bash(UV_PROJECT_ENVIRONMENT=py* uv run:*)",
+      "Bash(echo EXIT:$?:*)",
+      "Bash(echo \"EXIT=$?\")",
+      "Read(/tmp/**)"
+    ],
+    "deny": [],
+    "ask": []
+  },
+  "prefersReducedMotion": false,
+  "outputStyle": "default"
+}

.claude/skills/commit-msg/style-guide-reference.md

@@ -0,0 +1,225 @@
+# Commit Message Style Guide for `tractor`
+
+Analysis based on 500 recent commits from the `tractor` repository.
+
+## Core Principles
+
+Write commit messages that are technically precise yet casual in
+tone. Use abbreviations and informal language while maintaining
+clarity about what changed and why.
+
+## Subject Line Format
+
+### Length and Structure
+- Target: ~50 chars with a hard-max of 67.
+- Use backticks around code elements (72.2% of commits)
+- Rarely use colons (5.2%), except for file prefixes
+- End with '?' for uncertain changes (rare: 0.8%)
+- End with '!' for important changes (rare: 2.0%)
+
+### Opening Verbs (Present Tense)
+
+Most common verbs from analysis:
+- `Add` (14.4%) - wholly new features/functionality
+- `Use` (4.4%) - adopt new approach/tool
+- `Drop` (3.6%) - remove code/feature
+- `Fix` (2.4%) - bug fixes
+- `Move`/`Mv` (3.6%) - relocate code
+- `Adjust` (2.0%) - minor tweaks
+- `Update` (1.6%) - enhance existing feature
+- `Bump` (1.2%) - dependency updates
+- `Rename` (1.2%) - identifier changes
+- `Set` (1.2%) - configuration changes
+- `Handle` (1.0%) - add handling logic
+- `Raise` (1.0%) - add error raising
+- `Pass` (0.8%) - pass parameters/values
+- `Support` (0.8%) - add support for something
+- `Hide` (1.4%) - make private/internal
+- `Always` (1.4%) - enforce consistent behavior
+- `Mk` (1.4%) - make/create (abbreviated)
+- `Start` (1.0%) - begin implementation
+
+Other frequent verbs: `More`, `Change`, `Extend`, `Disable`, `Log`,
+`Enable`, `Ensure`, `Expose`, `Allow`
+
+### Backtick Usage
+
+Always use backticks for:
+- Module names: `trio`, `asyncio`, `msgspec`, `greenback`, `stackscope`
+- Class names: `Context`, `Actor`, `Address`, `PldRx`, `SpawnSpec`
+- Method names: `.pause_from_sync()`, `._pause()`, `.cancel()`
+- Function names: `breakpoint()`, `collapse_eg()`, `open_root_actor()`
+- Decorators: `@acm`, `@context`
+- Exceptions: `Cancelled`, `TransportClosed`, `MsgTypeError`
+- Keywords: `finally`, `None`, `False`
+- Variable names: `tn`, `debug_mode`
+- Complex expressions: `trio.Cancelled`, `asyncio.Task`
+
+Most backticked terms in tractor:
+`trio`, `asyncio`, `Context`, `.pause_from_sync()`, `tn`,
+`._pause()`, `breakpoint()`, `collapse_eg()`, `Actor`, `@acm`,
+`.cancel()`, `Cancelled`, `open_root_actor()`, `greenback`
+
+### Examples
+
+Good subject lines:
+```
+Add `uds` to `._multiaddr`, tweak typing
+Drop `DebugStatus.shield` attr, add `.req_finished`
+Use `stackscope` for all actor-tree rendered "views"
+Fix `.to_asyncio` inter-task-cancellation!
+Bump `ruff.toml` to target py313
+Mv `load_module_from_path()` to new `._code_load` submod
+Always use `tuple`-cast for singleton parent addrs
+```
+
+## Body Format
+
+### General Structure
+- 43.2% of commits have no body (simple changes)
+- Use blank line after subject
+- Max line length: 67 chars
+- Use `-` bullets for lists (28.0% of commits)
+- Rarely use `*` bullets (2.4%)
+
+### Section Markers
+
+Use these markers to organize longer commit bodies:
+- `Also,` (most common: 26 occurrences)
+- `Other,` (13 occurrences)
+- `Deats,` (11 occurrences) - for implementation details
+- `Further,` (7 occurrences)
+- `TODO,` (3 occurrences)
+- `Impl details,` (2 occurrences)
+- `Notes,` (1 occurrence)
+
+### Common Abbreviations
+
+Use these freely (sorted by frequency):
+- `msg` (63) - message
+- `bg` (37) - background
+- `ctx` (30) - context
+- `impl` (27) - implementation
+- `mod` (26) - module
+- `obvi` (17) - obviously
+- `tn` (16) - task name
+- `fn` (15) - function
+- `vs` (15) - versus
+- `bc` (14) - because
+- `var` (14) - variable
+- `prolly` (9) - probably
+- `ep` (6) - entry point
+- `OW` (5) - otherwise
+- `rn` (4) - right now
+- `sig` (4) - signal/signature
+- `deps` (3) - dependencies
+- `iface` (2) - interface
+- `subproc` (2) - subprocess
+- `tho` (2) - though
+- `ofc` (2) - of course
+
+### Tone and Style
+
+- Casual but technical (use `XD` for humor: 23 times)
+- Use `..` for trailing thoughts (108 occurrences)
+- Use `Woops,` to acknowledge mistakes (4 subject lines)
+- Don't be afraid to show personality while being precise
+
+### Example Bodies
+
+Simple with bullets:
+```
+Add `multiaddr` and bump up some deps
+
+Since we're planning to use it for (discovery)
+addressing, allowing replacement of the hacky (pretend)
+attempt in `tractor._multiaddr` Bp
+
+Also pin some deps,
+- make us py312+
+- use `pdbp` with my frame indexing fix.
+- mv to latest `xonsh` for fancy cmd/suggestion injections.
+
+Bump lock file to match obvi!
+```
+
+With section markers:
+```
+Use `stackscope` for all actor-tree rendered "views"
+
+Instead of the (much more) limited and hacky `.devx._code`
+impls, move to using the new `.devx._stackscope` API which
+wraps the `stackscope` project.
+
+Deats,
+- make new `stackscope.extract_stack()` wrapper
+- port over frame-descing to `_stackscope.pformat_stack()`
+- move `PdbREPL` to use `stackscope` render approach
+- update tests for new stack output format
+
+Also,
+- tweak log formatting for consistency
+- add typing hints throughout
+```
+
+## Special Patterns
+
+### WIP Commits
+Rare (0.2%) - avoid committing WIP if possible
+
+### Merge Commits
+Auto-generated (4.4%), don't worry about style
+
+### File References
+- Use `module.py` or `.submodule` style
+- Rarely use `file.py:line` references (0 in analysis)
+
+### Links
+- GitHub links used sparingly (3 total)
+- Prefer code references over external links
+
+## Footer
+
+The default footer should credit `claude` (you) for helping generate
+the commit msg content:
+
+```
+(this commit msg was generated in some part by [`claude-code`][claude-code-gh])
+[claude-code-gh]: https://github.com/anthropics/claude-code
+```
+
+Further, if the patch was solely or in part written
+by `claude`, instead add:
+
+```
+(this patch was generated in some part by [`claude-code`][claude-code-gh])
+[claude-code-gh]: https://github.com/anthropics/claude-code
+```
+
+## Summary Checklist
+
+Before committing, verify:
+- [ ] Subject line uses present tense verb
+- [ ] Subject line ~50 chars (hard max 67)
+- [ ] Code elements wrapped in backticks
+- [ ] Body lines ≤67 chars
+- [ ] Abbreviations used where natural
+- [ ] Casual yet precise tone
+- [ ] Section markers if body >3 paragraphs
+- [ ] Technical accuracy maintained
+
+## Analysis Metadata
+
+```
+Source: tractor repository
+Commits analyzed: 500
+Date range: 2019-2025
+Analysis date: 2026-02-08
+```
+
+---
+
+(this style guide was generated by [`claude-code`][claude-code-gh]
+analyzing commit history)
+
+[claude-code-gh]: https://github.com/anthropics/claude-code

.claude/skills/conc-anal/SKILL.md

@@ -0,0 +1,297 @@
+---
+name: conc-anal
+description: >
+  Concurrency analysis for tractor's trio-based
+  async primitives. Trace task scheduling across
+  checkpoint boundaries, identify race windows in
+  shared mutable state, and verify synchronization
+  correctness. Invoke on code segments the user
+  points at, OR proactively when reviewing/writing
+  concurrent cache, lock, or multi-task acm code.
+argument-hint: "[file:line-range or function name]"
+allowed-tools:
+  - Read
+  - Grep
+  - Glob
+  - Task
+---
+
+Perform a structured concurrency analysis on the
+target code. This skill should be invoked:
+
+- **On demand**: user points at a code segment
+  (file:lines, function name, or pastes a snippet)
+- **Proactively**: when writing or reviewing code
+  that touches shared mutable state across trio
+  tasks — especially `_Cache`, locks, events, or
+  multi-task `@acm` lifecycle management
+
+## 0. Identify the target
+
+If the user provides a file:line-range or function
+name, read that code. If not explicitly provided,
+identify the relevant concurrent code from context
+(e.g. the current diff, a failing test, or the
+function under discussion).
+
+## 1. Inventory shared mutable state
+
+List every piece of state that is accessed by
+multiple tasks. For each, note:
+
+- **What**: the variable/dict/attr (e.g.
+  `_Cache.values`, `_Cache.resources`,
+  `_Cache.users`)
+- **Scope**: class-level, module-level, or
+  closure-captured
+- **Writers**: which tasks/code-paths mutate it
+- **Readers**: which tasks/code-paths read it
+- **Guarded by**: which lock/event/ordering
+  protects it (or "UNGUARDED" if none)
+
+Format as a table:
+
+```
+| State               | Writers         | Readers         | Guard          |
+|---------------------|-----------------|-----------------|----------------|
+| _Cache.values       | run_ctx, moc¹   | moc             | ctx_key lock   |
+| _Cache.resources    | run_ctx, moc    | moc, run_ctx    | UNGUARDED      |
+```
+
+¹ `moc` = `maybe_open_context`
+
+## 2. Map checkpoint boundaries
+
+For each code path through the target, mark every
+**checkpoint** — any `await` expression where trio
+can switch to another task. Use line numbers:
+
+```
+L325: await lock.acquire()        ← CHECKPOINT
+L395: await service_tn.start(...) ← CHECKPOINT
+L411: lock.release()              ← (not a checkpoint, but changes lock state)
+L414: yield (False, yielded)      ← SUSPEND (caller runs)
+L485: no_more_users.set()         ← (wakes run_ctx, no switch yet)
+```
+
+**Key trio scheduling rules to apply:**
+- `Event.set()` makes waiters *ready* but does NOT
+  switch immediately
+- `lock.release()` is not a checkpoint
+- `await sleep(0)` IS a checkpoint
+- Code in `finally` blocks CAN have checkpoints
+  (unlike asyncio)
+- `await` inside `except` blocks can be
+  `trio.Cancelled`-masked
+
+## 3. Trace concurrent task schedules
+
+Write out the **interleaved execution trace** for
+the problematic scenario. Number each step and tag
+which task executes it:
+
+```
+[Task A]  1. acquires lock
+[Task A]  2. cache miss → allocates resources
+[Task A]  3. releases lock
+[Task A]  4. yields to caller
+[Task A]  5. caller exits → finally runs
+[Task A]  6. users-- → 0, sets no_more_users
+[Task A]  7. pops lock from _Cache.locks
+[run_ctx] 8. wakes from no_more_users.wait()
+[run_ctx] 9. values.pop(ctx_key)
+[run_ctx] 10. acm __aexit__ → CHECKPOINT
+[Task B]  11. creates NEW lock (old one popped)
+[Task B]  12. acquires immediately
+[Task B]  13. values[ctx_key] → KeyError
+[Task B]  14. resources[ctx_key] → STILL EXISTS
+[Task B]  15. 💥 RuntimeError
+```
+
+Identify the **race window**: the range of steps
+where state is inconsistent. In the example above,
+steps 9–10 are the window (values gone, resources
+still alive).
+
+## 4. Classify the bug
+
+Categorize what kind of concurrency issue this is:
+
+- **TOCTOU** (time-of-check-to-time-of-use): state
+  changes between a check and the action based on it
+- **Stale reference**: a task holds a reference to
+  state that another task has invalidated
+- **Lifetime mismatch**: a synchronization primitive
+  (lock, event) has a shorter lifetime than the
+  state it's supposed to protect
+- **Missing guard**: shared state is accessed
+  without any synchronization
+- **Atomicity gap**: two operations that should be
+  atomic have a checkpoint between them
+
+## 5. Propose fixes
+
+For each proposed fix, provide:
+
+- **Sketch**: pseudocode or diff showing the change
+- **How it closes the window**: which step(s) from
+  the trace it eliminates or reorders
+- **Tradeoffs**: complexity, perf, new edge cases,
+  impact on other code paths
+- **Risk**: what could go wrong (deadlocks, new
+  races, cancellation issues)
+
+Rate each fix: `[simple|moderate|complex]` impl
+effort.
+
+## 6. Output format
+
+Structure the full analysis as:
+
+```markdown
+## Concurrency analysis: `<target>`
+
+### Shared state
+<table from step 1>
+
+### Checkpoints
+<list from step 2>
+
+### Race trace
+<interleaved trace from step 3>
+
+### Classification
+<bug type from step 4>
+
+### Fixes
+<proposals from step 5>
+```
+
+## Tractor-specific patterns to watch
+
+These are known problem areas in tractor's
+concurrency model. Flag them when encountered:
+
+### `_Cache` lock vs `run_ctx` lifetime
+
+The `_Cache.locks` entry is managed by
+`maybe_open_context` callers, but `run_ctx` runs
+in `service_tn` — a different task tree. Lock
+pop/release in the caller's `finally` does NOT
+wait for `run_ctx` to finish tearing down. Any
+state that `run_ctx` cleans up in its `finally`
+(e.g. `resources.pop()`) is vulnerable to
+re-entry races after the lock is popped.
+
+### `values.pop()` → acm `__aexit__` → `resources.pop()` gap
+
+In `_Cache.run_ctx`, the inner `finally` pops
+`values`, then the acm's `__aexit__` runs (which
+has checkpoints), then the outer `finally` pops
+`resources`. This creates a window where `values`
+is gone but `resources` still exists — a classic
+atomicity gap.
+
+### Global vs per-key counters
+
+`_Cache.users` as a single `int` (pre-fix) meant
+that users of different `ctx_key`s inflated each
+other's counts, preventing teardown when one key's
+users hit zero. Always verify that per-key state
+(`users`, `locks`) is actually keyed on `ctx_key`
+and not on `fid` or some broader key.
+
+### `Event.set()` wakes but doesn't switch
+
+`trio.Event.set()` makes waiting tasks *ready* but
+the current task continues executing until its next
+checkpoint. Code between `.set()` and the next
+`await` runs atomically from the scheduler's
+perspective. Use this to your advantage (or watch
+for bugs where code assumes the woken task runs
+immediately).
+
+### `except` block checkpoint masking
+
+`await` expressions inside `except` handlers can
+be masked by `trio.Cancelled`. If a `finally`
+block runs from an `except` and contains
+`lock.release()`, the release happens — but any
+`await` after it in the same `except` may be
+swallowed. This is why `maybe_open_context`'s
+cache-miss path does `lock.release()` in a
+`finally` inside the `except KeyError`.
+
+### Cancellation in `finally`
+
+Unlike asyncio, trio allows checkpoints in
+`finally` blocks. This means `finally` cleanup
+that does `await` can itself be cancelled (e.g.
+by nursery shutdown). Watch for cleanup code that
+assumes it will run to completion.
+
+### Unbounded waits in cleanup paths
+
+Any `await <event>.wait()` in a teardown path is
+a latent deadlock unless the event's setter is
+GUARANTEED to fire. If the setter depends on
+external state (peer disconnects, child process
+exit, subsequent task completion) that itself
+depends on the current task's progress, you have
+a mutual wait.
+
+Rule: **bound every `await X.wait()` in cleanup
+paths with `trio.move_on_after()`** unless you
+can prove the setter is unconditionally reachable
+from the state at the await site. Concrete recent
+example: `ipc_server.wait_for_no_more_peers()` in
+`async_main`'s finally (see
+`ai/conc-anal/subint_forkserver_test_cancellation_leak_issue.md`
+"probe iteration 3") — it was unbounded, and when
+one peer-handler was stuck the wait-for-no-more-
+peers event never fired, deadlocking the whole
+actor-tree teardown cascade.
+
+### The capture-pipe-fill hang pattern (grep this first)
+
+When investigating any hang in the test suite
+**especially under fork-based backends**, first
+check whether the hang reproduces under `pytest
+-s` (`--capture=no`). If `-s` makes it go away
+you're not looking at a trio concurrency bug —
+you're looking at a Linux pipe-buffer fill.
+
+Mechanism: pytest replaces fds 1,2 with pipe
+write-ends. Fork-child subactors inherit those
+fds. High-volume error-log tracebacks (cancel
+cascade spew) fill the 64KB pipe buffer. Child
+`write()` blocks. Child can't exit. Parent's
+`waitpid`/pidfd wait blocks. Deadlock cascades up
+the tree.
+
+Pre-existing guards in `tests/conftest.py` encode
+this knowledge — grep these BEFORE blaming
+concurrency:
+
+```python
+# tests/conftest.py:258
+if loglevel in ('trace', 'debug'):
+    # XXX: too much logging will lock up the subproc (smh)
+    loglevel: str = 'info'
+
+# tests/conftest.py:316
+# can lock up on the `_io.BufferedReader` and hang..
+stderr: str = proc.stderr.read().decode()
+```
+
+Full post-mortem +
+`ai/conc-anal/subint_forkserver_test_cancellation_leak_issue.md`
+for the canonical reproduction. Cost several
+investigation sessions before catching it —
+because the capture-pipe symptom was masked by
+deeper cascade-deadlocks. Once the cascades were
+fixed, the tree tore down enough to generate
+pipe-filling log volume → capture-pipe finally
+surfaced. Grep-note for future-self: **if a
+multi-subproc tractor test hangs, `pytest -s`
+first, conc-anal second.**

.claude/skills/pr-msg/format-reference.md

@@ -0,0 +1,241 @@
+# PR/Patch-Request Description Format Reference
+
+Canonical structure for `tractor` patch-request
+descriptions, designed to work across GitHub,
+Gitea, SourceHut, and GitLab markdown renderers.
+
+**Line length: wrap at 72 chars** for all prose
+content (Summary bullets, Motivation paragraphs,
+Scopes bullets, etc.). Fill lines *to* 72 — don't
+stop short at 50-65. Only raw URLs in
+reference-link definitions may exceed this.
+
+## Template
+
+```markdown
+<!-- pr-msg-meta
+branch: <branch-name>
+base: <base-branch>
+submitted:
+  github: ___
+  gitea: ___
+  srht: ___
+-->
+
+## <Title: present-tense verb + backticked code>
+
+### Summary
+- [<hash>][<hash>] Description of change ending
+  with period.
+- [<hash>][<hash>] Another change description
+  ending with period.
+- [<hash>][<hash>] [<hash>][<hash>] Multi-commit
+  change description.
+
+### Motivation
+<1-2 paragraphs: problem/limitation first,
+then solution. Hard-wrap at 72 chars.>
+
+### Scopes changed
+- [<hash>][<hash>] `pkg.mod.func()` — what
+  changed.
+  * [<hash>][<hash>] Also adjusts
+    `.related_thing()` in same module.
+- [<hash>][<hash>] `tests.test_mod` — new/changed
+  test coverage.
+
+<!--
+### Cross-references
+Also submitted as
+[github-pr][] | [gitea-pr][] | [srht-patch][].
+
+### Links
+- [relevant-issue-or-discussion](url)
+- [design-doc-or-screenshot](url)
+-->
+
+(this pr content was generated in some part by
+[`claude-code`][claude-code-gh])
+
+[<hash>]: https://<service>/<owner>/<repo>/commit/<hash>
+[claude-code-gh]: https://github.com/anthropics/claude-code
+
+<!-- cross-service pr refs (fill after submit):
+[github-pr]: https://github.com/<owner>/<repo>/pull/___
+[gitea-pr]: https://<host>/<owner>/<repo>/pulls/___
+[srht-patch]: https://git.sr.ht/~<owner>/<repo>/patches/___
+-->
+```
+
+## Markdown Reference-Link Strategy
+
+Use reference-style links for ALL commit hashes
+and cross-service PR refs to ensure cross-service
+compatibility:
+
+**Inline usage** (in bullets):
+```markdown
+- [f3726cf9][f3726cf9] Add `reg_err_types()`
+  for custom exc lookup.
+```
+
+**Definition** (bottom of document):
+```markdown
+[f3726cf9]: https://github.com/goodboy/tractor/commit/f3726cf9
+```
+
+### Why reference-style?
+- Keeps prose readable without long inline URLs.
+- All URLs in one place — trivially swappable
+  per-service.
+- Most git services auto-link bare SHAs anyway,
+  but explicit refs guarantee it works in *any*
+  md renderer.
+- The `[hash][hash]` form is self-documenting —
+  display text matches the ref ID.
+- Cross-service PR refs use the same mechanism:
+  `[github-pr][]` resolves via a ref-link def
+  at the bottom, trivially fillable post-submit.
+
+## Cross-Service PR Placeholder Mechanism
+
+The generated description includes three layers
+of cross-service support, all using native md
+reference-links:
+
+### 1. Metadata comment (top of file)
+
+```markdown
+<!-- pr-msg-meta
+branch: remote_exc_type_registry
+base: main
+submitted:
+  github: ___
+  gitea: ___
+  srht: ___
+-->
+```
+
+A YAML-ish HTML comment block. The `___`
+placeholders get filled with PR/patch numbers
+after submission. Machine-parseable for tooling
+(e.g. `gish`) but invisible in rendered md.
+
+### 2. Cross-references section (in body)
+
+```markdown
+<!--
+### Cross-references
+Also submitted as
+[github-pr][] | [gitea-pr][] | [srht-patch][].
+-->
+```
+
+Commented out at generation time. After submitting
+to multiple services, uncomment and the ref-links
+resolve via the stubs at the bottom.
+
+### 3. Ref-link stubs (bottom of file)
+
+```markdown
+<!-- cross-service pr refs (fill after submit):
+[github-pr]: https://github.com/goodboy/tractor/pull/___
+[gitea-pr]: https://pikers.dev/goodboy/tractor/pulls/___
+[srht-patch]: https://git.sr.ht/~goodboy/tractor/patches/___
+-->
+```
+
+Commented out with `___` number placeholders.
+After submission: uncomment, replace `___` with
+the actual number. Each service-specific copy
+fills in all services' numbers so any copy can
+cross-reference the others.
+
+### Post-submission file layout
+
+```
+pr_msg_LATEST.md                    # latest draft (skill root)
+msgs/
+  20260325T002027Z_mybranch_pr_msg.md  # timestamped
+  github/
+    42_pr_msg.md        # github PR #42
+  gitea/
+    17_pr_msg.md        # gitea PR #17
+  srht/
+    5_pr_msg.md         # srht patch #5
+```
+
+Each `<service>/<num>_pr_msg.md` is a copy with:
+- metadata `submitted:` fields filled in
+- cross-references section uncommented
+- ref-link stubs uncommented with real numbers
+- all services cross-linked in each copy
+
+This mirrors the `gish` skill's
+`<backend>/<num>.md` pattern.
+
+## Commit-Link URL Patterns by Service
+
+| Service   | Pattern                             |
+|-----------|-------------------------------------|
+| GitHub    | `https://github.com/<o>/<r>/commit/<h>` |
+| Gitea     | `https://<host>/<o>/<r>/commit/<h>` |
+| SourceHut | `https://git.sr.ht/~<o>/<r>/commit/<h>` |
+| GitLab    | `https://gitlab.com/<o>/<r>/-/commit/<h>` |
+
+## PR/Patch URL Patterns by Service
+
+| Service   | Pattern                             |
+|-----------|-------------------------------------|
+| GitHub    | `https://github.com/<o>/<r>/pull/<n>` |
+| Gitea     | `https://<host>/<o>/<r>/pulls/<n>`  |
+| SourceHut | `https://git.sr.ht/~<o>/<r>/patches/<n>` |
+| GitLab    | `https://gitlab.com/<o>/<r>/-/merge_requests/<n>` |
+
+## Scope Naming Convention
+
+Use Python namespace-resolution syntax for
+referencing changed code scopes:
+
+| File path                 | Scope reference               |
+|---------------------------|-------------------------------|
+| `tractor/_exceptions.py`  | `tractor._exceptions`         |
+| `tractor/_state.py`       | `tractor._state`              |
+| `tests/test_foo.py`       | `tests.test_foo`              |
+| Function in module        | `tractor._exceptions.func()`  |
+| Method on class           | `.RemoteActorError.src_type`  |
+| Class                     | `tractor._exceptions.RAE`     |
+
+Prefix with the package path for top-level refs;
+use leading-dot shorthand (`.ClassName.method()`)
+for sub-bullets where the parent module is already
+established.
+
+## Title Conventions
+
+Same verb vocabulary as commit messages:
+- `Add` — wholly new feature/API
+- `Fix` — bug fix
+- `Drop` — removal
+- `Use` — adopt new approach
+- `Move`/`Mv` — relocate code
+- `Adjust` — minor tweak
+- `Update` — enhance existing feature
+- `Support` — add support for something
+
+Target 50 chars, hard max 70. Always backtick
+code elements.
+
+## Tone
+
+Casual yet technically precise — matching the
+project's commit-msg style. Terse but every bullet
+carries signal. Use project abbreviations freely
+(msg, bg, ctx, impl, mod, obvi, fn, bc, var,
+prolly, ep, etc.).
+
+---
+
+(this format reference was generated by
+[`claude-code`][claude-code-gh])
+[claude-code-gh]: https://github.com/anthropics/claude-code

.claude/skills/run-tests/test-harness-reference.md

@@ -0,0 +1,255 @@
+# Tractor Test Harness Reference
+
+This repository-local file supplements the canonical [`/run-tests` skill][1]
+from the [`ai.skillz` repository][2]. Its deployer links the shared `SKILL.md`
+into both
+`.claude/skills/run-tests/` and `.opencode/skills/run-tests/` while preserving
+this project-owned reference:
+
+```text
+bash /path/to/ai.skillz/scripts/deploy.sh run-tests /path/to/tractor --provider all --method symlink
+```
+
+Keep shared environment permission, process-signal safety, target selection,
+failure inspection, and result reporting policy in the deployed `SKILL.md`.
+
+[1]: https://github.com/baudco/ai.skillz/blob/2d4896ca7e38fe2cb3090cdefc7245be4241a6d2/skills/run-tests/SKILL.md
+[2]: https://github.com/baudco/ai.skillz
+
+## Project And Environment
+
+- Project/import: `tractor`
+- Test root: `tests/`
+- Supported Python: `>=3.13,<3.15`
+- Runner: pytest `>=9.0.3`
+- Test dependencies: the `dev` group includes the `testing` group
+- CI uses uv's default `.venv`; the Nix flake uses `py313`.
+- Run from the repository root so pytest loads `pyproject.toml`.
+- Do not use `default.nix` as current test-environment authority; it still
+  selects unsupported Python 3.12.
+
+Environment directory naming is not a harness invariant. Use an already
+verified active project environment when available. Otherwise, use an
+existing uv environment without syncing it:
+
+```text
+uv run --frozen --no-sync python -c 'import pathlib, sys, tractor; root = pathlib.Path.cwd().resolve(); mod = pathlib.Path(tractor.__file__).resolve(); print(sys.executable); print(mod); assert mod.is_relative_to(root)'
+```
+
+After module moves or collection failures, check collection with:
+
+```text
+uv run --frozen --no-sync pytest --collect-only -q tests/
+```
+
+Collection is not a mandatory precursor to every narrow run. Ask before
+provisioning or changing an environment.
+
+Before trusting CLI-selected runtime settings, inspect
+`TRACTOR_SPAWN_METHOD` and `TRACTOR_LOGLEVEL`. They override the spawn method
+and runtime log level passed by callers, so report active values with test
+results rather than claiming the CLI flags alone selected the runtime.
+
+## Pytest Configuration And Commands
+
+`pyproject.toml` configures:
+
+- `testpaths = ["tests"]` and `--rootdir=./tests`;
+- importlib import mode;
+- the `tractor._testing.pytest` plugin;
+- xonsh plugin disablement;
+- `--show-capture=no` and `--capture=fd`.
+
+Do not silently add `-x`, `--tb=short`, or `--no-header`; those are not
+project defaults. In a verified active environment, replace `uv run
+--frozen --no-sync pytest` below with `python -m pytest`.
+
+```text
+# Full suite
+uv run --frozen --no-sync pytest tests/
+
+# Narrow file
+uv run --frozen --no-sync pytest tests/test_local.py
+
+# Exact node
+uv run --frozen --no-sync pytest tests/discovery/test_registrar.py::test_reg_then_unreg
+
+# Keyword selection
+uv run --frozen --no-sync pytest tests/ -k 'cancel and not slow'
+
+# Previous failures
+uv run --frozen --no-sync pytest --lf
+```
+
+After verifying that the no-sync environment is current, these pytest
+arguments match the Linux TCP CI row:
+
+```text
+CI=1 uv run --frozen --no-sync pytest tests/ -rsx --spawn-backend=trio --tpt-proto=tcp --capture=fd
+```
+
+## Plugin Options And Matrices
+
+Supported spawn backends:
+
+- `trio` (default)
+- `mp_spawn`
+- `mp_forkserver`
+
+Do not advertise `subint`, `subint_forkserver`, or
+`main_thread_forkserver` as runnable backends. Supported transports are
+`tcp` (default) and `uds`. Run one transport per pytest session.
+`mp_forkserver` and UDS are POSIX-only.
+
+Other Tractor plugin options include:
+
+- `--tpdb` / `--debug-mode`
+- `--ll` / `--loglevel`
+- `--tl` / `--tractor-loglevel`
+- `--enable-stackscope`
+
+Examples:
+
+```text
+uv run --frozen --no-sync pytest tests/ipc/ --tpt-proto=uds
+uv run --frozen --no-sync pytest tests/test_spawning.py --spawn-backend=mp_spawn
+uv run --frozen --no-sync pytest tests/test_spawning.py --spawn-backend=mp_forkserver --capture=sys
+```
+
+CI currently exercises Python 3.13 with the `trio` backend: TCP and UDS on
+Linux and macOS, plus an informational TCP row on Windows whose pytest step
+uses `continue-on-error`.
+
+## Registry And Transport Isolation
+
+Tests requesting the `reg_addr` fixture use addresses randomized per session:
+an unreserved unprivileged loopback port for TCP or a unique socket name under
+the platform runtime directory for UDS. A TCP collision remains possible.
+
+The runtime fallback remains `127.0.0.1:1616` or `registry@1616.sock`.
+Inspect that fallback only when the selected test intentionally uses runtime
+defaults or a failure identifies that address. Do not perform a mandatory
+`:1616` preflight or assume UDS sockets live under `/tmp`.
+
+## Capture And Hang Diagnosis
+
+Normal capture is `fd`. Use `--capture=sys` with `mp_forkserver`; some tests
+switch to `capsys`, but the harness does not enforce that suite-wide.
+
+For a suspected capture interaction, compare only the exact node:
+
+```text
+uv run --frozen --no-sync pytest <node> --capture=sys
+uv run --frozen --no-sync pytest <node> -s
+```
+
+Treat `-s` as a diagnostic comparison, not a pass-equivalent workaround. Do
+not use it to reinterpret an ordinary captured pass. Interactive `--tpdb` or
+`tractor.pause()` sessions are different: they require a real TTY and disabled
+capture, normally `-s`.
+
+Do not add a global pytest timeout. `fail_after_w_trace` is Trio-cooperative;
+`afk_alarm_w_trace` is a POSIX main-thread `SIGALRM` hard backstop and can
+raise asynchronously. Use the latter only as a last resort, not as a generally
+Trio-safe timeout replacement.
+
+For live task-tree diagnosis:
+
+```text
+uv run --frozen --no-sync python -c 'import stackscope'
+uv run --frozen --no-sync pytest <node> --enable-stackscope --capture=sys
+kill -USR1 <pytest-pid>
+```
+
+The import check and pytest command must use the same environment. Do not send
+SIGUSR1 if the import fails or setup warns that stackscope or SIGUSR1 is
+unavailable: without the installed handler, SIGUSR1 normally terminates the
+target process. Signal a subactor only after separately confirming that it
+installed the same handler.
+
+Stackscope appends dumps to `/tmp/tractor-stackscope-<pid>.log`, including when
+pytest capture hides terminal output. SIGUSR1 stackscope is unavailable on
+Windows and degrades to a no-op there.
+
+When a trace guard actually fires and snapshot capture succeeds, it writes
+under `$XDG_CACHE_HOME/tractor/hung-dumps/`, falling back beneath
+`~/.cache/tractor/hung-dumps/`, and prints an end-of-session index. A normal
+non-timeout run creates no snapshot.
+
+## Cleanup And `tractor-reap`
+
+On Linux, normal pytest teardown discovers surviving descendants through
+`/proc`, sends SIGINT, waits three seconds, then escalates survivors to
+SIGKILL. It does not sweep shared memory and cannot run if pytest never reaches
+fixture teardown.
+
+Process discovery is a no-op off Linux. UDS PID liveness also depends on
+`/proc`; on macOS, recognized PID-named sockets can therefore be classified as
+dead without proof. The session-scoped autouse fixture currently passes those
+candidates directly to `reap_uds()` at teardown. Do not treat its non-Linux
+classification as proof of orphanhood or run concurrent live Tractor sessions
+against the same UDS bindspace.
+
+Use the CLI in inspection-only mode first:
+
+`--shm` and `--shm-only` are Linux/FreeBSD-only and raise
+`NotImplementedError` elsewhere. On other platforms, use the UDS-only command.
+
+```text
+uv run --frozen --no-sync scripts/tractor-reap -n
+uv run --frozen --no-sync scripts/tractor-reap --parent <pytest-pid> -n
+uv run --frozen --no-sync scripts/tractor-reap --shm --uds -n
+uv run --frozen --no-sync scripts/tractor-reap --uds-only -n
+```
+
+Direct `scripts/tractor-reap` execution is acceptable only after verifying its
+`python3` shebang resolves the intended project environment.
+
+Review every candidate before requesting a mutating run:
+
+- default orphan mode is not repository-scoped;
+- `--parent` trusts the supplied PID and can include non-Tractor children;
+- `--shm` scans all current-user candidate files, not just Tractor-named
+  files;
+- `--uds` treats `registry@1616.sock` as removable even if a live default UDS
+  registrar uses it.
+
+Dry-run output prints only the initially matched root PIDs. A mutating run can
+recursively expand those roots to additional descendants when `psutil` is
+available. Inspect the descendant process tree separately; `-n` is not exact
+signal-set parity and does not by itself authorize signaling unseen children.
+
+The canonical skill owns signaling and unlinking authorization.
+
+## Test Layout And Change Mapping
+
+| Changed area | Run first |
+|---|---|
+| `tractor/runtime/_runtime.py`, `_state.py`, `tractor/_root.py` | `tests/test_local.py`, `tests/test_root_runtime.py`, `tests/test_runtime.py`, `tests/test_rpc.py` |
+| `tractor/runtime/_portal.py`, `_rpc.py` | `tests/test_rpc.py`, `tests/test_cancellation.py` |
+| `tractor/runtime/_supervise.py` | `tests/test_cancellation.py`, `tests/test_spawning.py` |
+| `tractor/discovery/` | `tests/discovery/`, `tests/test_local.py` |
+| `tractor/ipc/` | `tests/ipc/`, `tests/test_2way.py`, `tests/test_shm.py` as relevant |
+| `tractor/spawn/` | `tests/test_spawning.py`, `tests/discovery/test_multi_program.py`, `tests/test_cancellation.py` |
+| `tractor/_context.py`, `_streaming.py` | `tests/test_context_stream_semantics.py`, `tests/test_advanced_streaming.py`, `tests/test_legacy_one_way_streaming.py` |
+| `tractor/to_asyncio.py` | `tests/test_infected_asyncio.py`, `tests/test_root_infect_asyncio.py` |
+| `tractor/msg/` | `tests/msg/` |
+| `tractor/devx/` | `tests/devx/`; debugger tests use pexpect and are comparatively slow |
+| `tractor/_exceptions.py` | `tests/test_remote_exc_relay.py`, `tests/test_reg_err_types.py`, `tests/test_inter_peer_cancellation.py`, `tests/test_cancellation.py`, `tests/msg/` |
+
+Current subdirectories include `discovery/`, `ipc/`, `msg/`, `devx/`, and
+`trionics/`. There is no `tests/spawn/` directory.
+
+## Expected Outcomes
+
+Do not maintain a blanket known-flaky exemption list. Classify only current
+explicit skip or xfail marks and exact expected signatures. Notable tracked
+outcomes include:
+
+- duplicate-name `n_dups=4` and `n_dups=8` variants in
+  `tests/discovery/test_multi_program.py` are non-strict xfails;
+- `tests/test_ringbuf.py` is module-skipped;
+- some documentation examples have explicit macOS-CI skips.
+
+A generic `TooSlowError` or `pexpect.TIMEOUT` is not enough to classify a
+failure as pre-existing.

.github/workflows/ci.yml

@@ -1,10 +1,18 @@
 name: CI
 
+# NOTE distilled from,
+# https://github.com/orgs/community/discussions/26276
 on:
-  # any time someone pushes a new branch to origin
+  # any time a new update to 'main'
   push:
+    branches:
+      - main
 
-  # Allows you to run this workflow manually from the Actions tab
+  # for on all (forked) PRs to repo
+  # NOTE, use a draft PR if you just want CI triggered..
+  pull_request:
+
+  # to run workflow manually from the "Actions" tab
   workflow_dispatch:
 
 jobs:
@@ -29,7 +37,12 @@ jobs:
         run: uv build --sdist --python=3.13
 
       - name: Install sdist from .tar.gz
-        run: python -m pip install dist/*.tar.gz
+        # XXX must install under py3.13 (matching the build's
+        # `--python=3.13`); the runner's default `python` is 3.12
+        # which our `requires-python = ">=3.13"` now rejects.
+        run: |
+          uv venv --python 3.13
+          uv pip install dist/*.tar.gz
 
   # ------ type-check ------
   # mypy:
@@ -74,24 +87,49 @@ jobs:
   #       run: mypy tractor/ --ignore-missing-imports --show-traceback
 
 
-  testing-linux:
-    name: '${{ matrix.os }} Python ${{ matrix.python }} - ${{ matrix.spawn_backend }}'
-    timeout-minutes: 10
+  testing:
+    name: '${{ matrix.os }} Python${{ matrix.python-version }} spawn_backend=${{ matrix.spawn_backend }} tpt_proto=${{ matrix.tpt_proto }}'
+    timeout-minutes: 16
     runs-on: ${{ matrix.os }}
+    # Windows support is nascent: its full test suite remains
+    # informational, while setup and the `import tractor` smoke below
+    # are hard signals. Promote the test step to required once the
+    # suite is green.
 
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest]
-        python-version: ['3.13']
+        os: [
+          ubuntu-latest,
+          macos-latest,
+          windows-latest,
+        ]
+        python-version: [
+          '3.13',
+          # '3.14',
+        ]
         spawn_backend: [
           'trio',
           # 'mp_spawn',
           # 'mp_forkserver',
+          # ?TODO^ is it worth it to get these running again?
+          #
+          # - [ ] next-gen backends, on 3.13+
+          #   https://github.com/goodboy/tractor/issues/379
+          # 'subinterpreter',
+          # 'subint',
         ]
+        tpt_proto: [
+          'tcp',
+          'uds',
+        ]
+        exclude:
+          # UDS is POSIX-only; Windows has no `AF_UNIX` so the
+          # backend is intentionally unavailable there.
+          - os: windows-latest
+            tpt_proto: 'uds'
 
     steps:
-
       - uses: actions/checkout@v4
 
       - name: 'Install uv + py-${{ matrix.python-version }}'
@@ -117,8 +155,28 @@ jobs:
       - name: List deps tree
         run: uv tree
 
+      # hard signal for the Windows import-safety fix: `import
+      # tractor` must succeed everywhere, and `HAS_UDS` reflects
+      # platform capability (False on Windows, True on POSIX).
+      - name: 'Smoke: import tractor'
+        run: uv run python -c "import sys; import tractor; from tractor.ipc._uds import HAS_UDS; assert sys.platform != 'win32' or not HAS_UDS; print('import tractor OK | HAS_UDS=', HAS_UDS)"
+
       - name: Run tests
-        run: uv run pytest tests/ --spawn-backend=${{ matrix.spawn_backend }} -rsx
+        # Actor/PTY scheduling on macOS can fail a different
+        # timing-sensitive node between otherwise-green runs. Retry
+        # only that matrix leg; deterministic failures still fail
+        # after the final attempt.
+        continue-on-error: ${{ matrix.os == 'windows-latest' }}
+        run: >
+          uv run
+          pytest
+          tests/
+          -rsx
+          --spawn-backend=${{ matrix.spawn_backend }}
+          --tpt-proto=${{ matrix.tpt_proto }}
+          --capture=fd
+          --reruns=${{ matrix.os == 'macos-latest' && 2 || 0 }}
+          --reruns-delay=1
 
   # XXX legacy NOTE XXX
   #

.github/workflows/docs.yml

@@ -0,0 +1,63 @@
+name: docs
+
+# build sphinx docs on every PR + push to main;
+# deploy to gh-pages only from main pushes.
+# (see goodboy/tractor#123 for the original ask)
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+  # to run workflow manually from the "Actions" tab
+  workflow_dispatch:
+
+# needed by actions/deploy-pages
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+jobs:
+  build:
+    name: 'sphinx build'
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Install latest uv
+        uses: astral-sh/setup-uv@v6
+
+      # NOTE, no `d2` bin is installed in CI (yet) so
+      # the pre-rendered + committed SVGs under
+      # `docs/_diagrams/` are used as-is; see
+      # `docs/_ext/d2diagrams.py` for the fallback
+      # policy.
+      - name: Build html docs
+        run: |
+          uv sync --no-dev --group docs
+          uv run --no-dev --group docs make -C docs html
+
+      - name: Upload pages artifact
+        uses: actions/upload-pages-artifact@v3
+        with:
+          path: docs/_build/html
+
+  deploy:
+    name: 'deploy to gh-pages'
+    if: ${{ github.ref == 'refs/heads/main' && github.event_name == 'push' }}
+    needs: build
+    # serialize deploys but NEVER cancel an in-flight one
+    # mid-upload; queue the next instead (scoped to this job so
+    # PR builds can't cancel a production deploy).
+    concurrency:
+      group: 'pages'
+      cancel-in-progress: false
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    steps:
+      - name: Deploy
+        id: deployment
+        uses: actions/deploy-pages@v4

.gitignore

@@ -102,3 +102,333 @@ venv.bak/
 
 # mypy
 .mypy_cache/
+
+# all files under
+.git/
+
+# require very explicit staging for anything we **really**
+# want put/kept in repo.
+notes_to_self/
+snippets/
+
+# ------- AI shiz -------
+# `ai.skillz` symlinks,
+# (machine-local, deploy via deploy-skill.sh)
+.claude/skills/py-codestyle
+.claude/skills/close-wkt
+.claude/skills/plan-io
+.claude/skills/prompt-io
+.claude/skills/resolve-conflicts
+.claude/skills/inter-skill-review
+
+# /open-wkt specifics
+.claude/skills/open-wkt
+.claude/wkts/
+claude_wkts
+
+# /code-review-changes specifics
+.claude/skills/code-review-changes
+# review-skill ephemeral ctx (per-PR, single-use)
+.claude/review_context.md
+.claude/review_regression.md
+
+# /pr-msg specifics
+.claude/skills/pr-msg/*
+# repo-specific
+!.claude/skills/pr-msg/format-reference.md
+# XXX, so u can nvim-telescope this file.
+# !.claude/skills/pr-msg/pr_msg_LATEST.md
+
+# /commit-msg specifics
+# - any commit-msg gen tmp files
+.claude/*_commit_*.md
+.claude/*_commit*.txt
+.claude/skills/commit-msg/*
+!.claude/skills/commit-msg/style-duie-reference.md
+
+# use prompt-io instead?
+.claude/plans
+
+# nix develop --profile .nixdev
+.nixdev*
+
+# :Obsession .
+Session.vim
+
+# `gish` local `.md`-files
+# TODO? better all around automation!
+# -[ ] it'd be handy to also commit and sync with wtv git service?
+# -[ ] everything should be put under a `.gish/` no?
+gitea/
+gh/
+
+# ------ macOS ------
+# Finder metadata
+**/.DS_Store
+
+# LLM conversations that should remain private
+docs/conversations/
+
+# BEGIN ai.skillz: direct:symlink:claude:run-tests
+/.claude/skills/run-tests/SKILL.md
+# END ai.skillz: direct:symlink:claude:run-tests
+
+# BEGIN ai.skillz: direct:symlink:opencode:run-tests
+/.opencode/skills/run-tests/SKILL.md
+# END ai.skillz: direct:symlink:opencode:run-tests
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:run-tests
+/.opencode/commands/run-tests.md
+# END ai.skillz: direct:symlink:opencode:command:run-tests
+
+# BEGIN ai.skillz: direct:symlink:claude:gish
+/.claude/skills/gish
+# END ai.skillz: direct:symlink:claude:gish
+
+# BEGIN ai.skillz: direct:symlink:opencode:gish
+/.opencode/skills/gish
+# END ai.skillz: direct:symlink:opencode:gish
+
+# BEGIN ai.skillz: direct:symlink:claude:resolve-conflicts
+/.claude/skills/resolve-conflicts
+# END ai.skillz: direct:symlink:claude:resolve-conflicts
+
+# BEGIN ai.skillz: direct:symlink:opencode:resolve-conflicts
+/.opencode/skills/resolve-conflicts
+# END ai.skillz: direct:symlink:opencode:resolve-conflicts
+
+# BEGIN ai.skillz: direct:symlink:claude:git-mgmt
+/.claude/skills/git-mgmt
+# END ai.skillz: direct:symlink:claude:git-mgmt
+
+# BEGIN ai.skillz: direct:symlink:opencode:git-mgmt
+/.opencode/skills/git-mgmt
+# END ai.skillz: direct:symlink:opencode:git-mgmt
+
+# BEGIN ai.skillz: runtime:open-wkt
+/wkts/
+# END ai.skillz: runtime:open-wkt
+
+# BEGIN ai.skillz: direct:symlink:claude:open-wkt
+/.claude/skills/open-wkt
+# END ai.skillz: direct:symlink:claude:open-wkt
+
+# BEGIN ai.skillz: direct:symlink:opencode:open-wkt
+/.opencode/skills/open-wkt
+# END ai.skillz: direct:symlink:opencode:open-wkt
+
+# BEGIN ai.skillz: direct:symlink:claude:close-wkt
+/.claude/skills/close-wkt
+# END ai.skillz: direct:symlink:claude:close-wkt
+
+# BEGIN ai.skillz: direct:symlink:opencode:close-wkt
+/.opencode/skills/close-wkt
+# END ai.skillz: direct:symlink:opencode:close-wkt
+
+# BEGIN ai.skillz: runtime:code-review
+.ai/code-review/reports/
+# END ai.skillz: runtime:code-review
+
+# BEGIN ai.skillz: direct:symlink:claude:code-review
+/.claude/skills/code-review
+# END ai.skillz: direct:symlink:claude:code-review
+
+# BEGIN ai.skillz: direct:symlink:opencode:code-review
+/.opencode/skills/code-review
+# END ai.skillz: direct:symlink:opencode:code-review
+
+# BEGIN ai.skillz: direct:symlink:claude:code-nav-refs
+/.claude/skills/code-nav-refs
+# END ai.skillz: direct:symlink:claude:code-nav-refs
+
+# BEGIN ai.skillz: direct:symlink:opencode:code-nav-refs
+/.opencode/skills/code-nav-refs
+# END ai.skillz: direct:symlink:opencode:code-nav-refs
+
+# BEGIN ai.skillz: runtime:code-review-changes
+.claude/review_context.md
+.claude/review_regression.md
+.claude/review_replies/
+# END ai.skillz: runtime:code-review-changes
+
+# BEGIN ai.skillz: direct:symlink:claude:code-review-changes
+/.claude/skills/code-review-changes
+# END ai.skillz: direct:symlink:claude:code-review-changes
+
+# BEGIN ai.skillz: direct:symlink:opencode:code-review-changes
+/.opencode/skills/code-review-changes
+# END ai.skillz: direct:symlink:opencode:code-review-changes
+
+# BEGIN ai.skillz: runtime:commit-msg
+.claude/skills/commit-msg/msgs/
+.claude/git_commit_msg_LATEST.md
+# END ai.skillz: runtime:commit-msg
+
+# BEGIN ai.skillz: direct:symlink:claude:commit-msg
+/.claude/skills/commit-msg/SKILL.md
+# END ai.skillz: direct:symlink:claude:commit-msg
+
+# BEGIN ai.skillz: direct:symlink:opencode:commit-msg
+/.opencode/skills/commit-msg/SKILL.md
+# END ai.skillz: direct:symlink:opencode:commit-msg
+
+# BEGIN ai.skillz: direct:symlink:claude:commit-plan
+/.claude/skills/commit-plan
+# END ai.skillz: direct:symlink:claude:commit-plan
+
+# BEGIN ai.skillz: direct:symlink:opencode:commit-plan
+/.opencode/skills/commit-plan
+# END ai.skillz: direct:symlink:opencode:commit-plan
+
+# BEGIN ai.skillz: direct:symlink:claude:dep-supersede-scan
+/.claude/skills/dep-supersede-scan
+# END ai.skillz: direct:symlink:claude:dep-supersede-scan
+
+# BEGIN ai.skillz: direct:symlink:opencode:dep-supersede-scan
+/.opencode/skills/dep-supersede-scan
+# END ai.skillz: direct:symlink:opencode:dep-supersede-scan
+
+# BEGIN ai.skillz: direct:symlink:claude:harness-perf
+/.claude/skills/harness-perf
+# END ai.skillz: direct:symlink:claude:harness-perf
+
+# BEGIN ai.skillz: direct:symlink:opencode:harness-perf
+/.opencode/skills/harness-perf
+# END ai.skillz: direct:symlink:opencode:harness-perf
+
+# BEGIN ai.skillz: direct:symlink:claude:inter-skill-review
+/.claude/skills/inter-skill-review
+# END ai.skillz: direct:symlink:claude:inter-skill-review
+
+# BEGIN ai.skillz: direct:symlink:opencode:inter-skill-review
+/.opencode/skills/inter-skill-review
+# END ai.skillz: direct:symlink:opencode:inter-skill-review
+
+# BEGIN ai.skillz: direct:symlink:claude:opencode-cleaning
+/.claude/skills/opencode-cleaning
+# END ai.skillz: direct:symlink:claude:opencode-cleaning
+
+# BEGIN ai.skillz: direct:symlink:opencode:opencode-cleaning
+/.opencode/skills/opencode-cleaning
+# END ai.skillz: direct:symlink:opencode:opencode-cleaning
+
+# BEGIN ai.skillz: direct:symlink:claude:plan-io
+/.claude/skills/plan-io
+# END ai.skillz: direct:symlink:claude:plan-io
+
+# BEGIN ai.skillz: direct:symlink:opencode:plan-io
+/.opencode/skills/plan-io
+# END ai.skillz: direct:symlink:opencode:plan-io
+
+# BEGIN ai.skillz: runtime:pr-msg
+.claude/skills/pr-msg/msgs/
+.claude/skills/pr-msg/pr_msg_LATEST.md
+# END ai.skillz: runtime:pr-msg
+
+# BEGIN ai.skillz: direct:symlink:claude:pr-msg
+/.claude/skills/pr-msg/SKILL.md
+/.claude/skills/pr-msg/references
+/.claude/skills/pr-msg/scripts
+# END ai.skillz: direct:symlink:claude:pr-msg
+
+# BEGIN ai.skillz: direct:symlink:opencode:pr-msg
+/.opencode/skills/pr-msg/SKILL.md
+/.opencode/skills/pr-msg/references
+/.opencode/skills/pr-msg/scripts
+# END ai.skillz: direct:symlink:opencode:pr-msg
+
+# BEGIN ai.skillz: direct:symlink:claude:prompt-io
+/.claude/skills/prompt-io
+# END ai.skillz: direct:symlink:claude:prompt-io
+
+# BEGIN ai.skillz: direct:symlink:opencode:prompt-io
+/.opencode/skills/prompt-io
+# END ai.skillz: direct:symlink:opencode:prompt-io
+
+# BEGIN ai.skillz: direct:symlink:claude:py-codestyle
+/.claude/skills/py-codestyle
+# END ai.skillz: direct:symlink:claude:py-codestyle
+
+# BEGIN ai.skillz: direct:symlink:opencode:py-codestyle
+/.opencode/skills/py-codestyle
+# END ai.skillz: direct:symlink:opencode:py-codestyle
+
+# BEGIN ai.skillz: runtime:taken-export
+.ai/taken/exports/
+# END ai.skillz: runtime:taken-export
+
+# BEGIN ai.skillz: direct:symlink:claude:taken-export
+/.claude/skills/taken-export
+# END ai.skillz: direct:symlink:claude:taken-export
+
+# BEGIN ai.skillz: direct:symlink:opencode:taken-export
+/.opencode/skills/taken-export
+# END ai.skillz: direct:symlink:opencode:taken-export
+
+# BEGIN ai.skillz: direct:symlink:claude:yt-url-lookup
+/.claude/skills/yt-url-lookup
+# END ai.skillz: direct:symlink:claude:yt-url-lookup
+
+# BEGIN ai.skillz: direct:symlink:opencode:yt-url-lookup
+/.opencode/skills/yt-url-lookup
+# END ai.skillz: direct:symlink:opencode:yt-url-lookup
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:gish
+/.opencode/commands/gish.md
+# END ai.skillz: direct:symlink:opencode:command:gish
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:resolve-conflicts
+/.opencode/commands/resolve-conflicts.md
+# END ai.skillz: direct:symlink:opencode:command:resolve-conflicts
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:git-mgmt
+/.opencode/commands/git-mgmt.md
+# END ai.skillz: direct:symlink:opencode:command:git-mgmt
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:open-wkt
+/.opencode/commands/open-wkt.md
+# END ai.skillz: direct:symlink:opencode:command:open-wkt
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:close-wkt
+/.opencode/commands/close-wkt.md
+# END ai.skillz: direct:symlink:opencode:command:close-wkt
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:code-review
+/.opencode/commands/code-review.md
+# END ai.skillz: direct:symlink:opencode:command:code-review
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:code-review-changes
+/.opencode/commands/code-review-changes.md
+# END ai.skillz: direct:symlink:opencode:command:code-review-changes
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:commit-msg
+/.opencode/commands/commit-msg.md
+# END ai.skillz: direct:symlink:opencode:command:commit-msg
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:commit-plan
+/.opencode/commands/commit-plan.md
+# END ai.skillz: direct:symlink:opencode:command:commit-plan
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:dep-supersede-scan
+/.opencode/commands/dep-supersede-scan.md
+# END ai.skillz: direct:symlink:opencode:command:dep-supersede-scan
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:harness-perf
+/.opencode/commands/harness-perf.md
+# END ai.skillz: direct:symlink:opencode:command:harness-perf
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:opencode-cleaning
+/.opencode/commands/opencode-cleaning.md
+# END ai.skillz: direct:symlink:opencode:command:opencode-cleaning
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:pr-msg
+/.opencode/commands/pr-msg.md
+# END ai.skillz: direct:symlink:opencode:command:pr-msg
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:taken-export
+/.opencode/commands/taken-export.md
+# END ai.skillz: direct:symlink:opencode:command:taken-export
+
+# BEGIN ai.skillz: direct:symlink:opencode:command:yt-url-lookup
+/.opencode/commands/yt-url-lookup.md
+# END ai.skillz: direct:symlink:opencode:command:yt-url-lookup

NEWS.rst

@@ -78,6 +78,7 @@ Bug Fixes
   for ``asyncio``-side errors to not propagate due to a race condition.
 
   The implementation fix summary is:
+
   - add state to signal the end of the ``trio`` side task to be
     read by the ``asyncio`` side and always cancel any ongoing
     task in such cases.

ai/conc-anal/fork_thread_semantics_execution_vs_memory.md

@@ -0,0 +1,281 @@
+# `fork()` in a multi-threaded program — execution-side vs. memory-side of the same coin
+
+A reference doc for readers who've encountered one of two
+opposite-sounding framings of POSIX `fork()` semantics in a
+multi-threaded program and are confused by the other.
+
+This is a sibling to
+`subint_fork_blocked_by_cpython_post_fork_issue.md` — that
+doc covers a CPython-level refusal of fork-from-subint;
+this one covers the more general POSIX layer, since
+tractor's main-thread forkserver design rests on it.
+
+## TL;DR
+
+POSIX `fork()` only preserves the *calling* thread as a
+runnable thread in the child — every other thread in the
+parent simply never executes another instruction in the
+child. trio's docs call this "leaked"; tractor's
+`_main_thread_forkserver.py` docstring calls it "gone".
+Both are correct: "gone" is the *execution* side (no
+scheduler entry, no instructions retired), "leaked" is the
+*memory* side (the dead threads' stacks and per-thread
+heap structures still ride into the child's address space
+as orphaned COW pages with no owner and no cleanup hook).
+Same POSIX reality, two halves of the same coin.
+
+## The two framings
+
+[python-trio/trio#1614][trio-1614] (the canonical "trio +
+fork" hazards thread) puts it this way:
+
+> If you use `fork()` in a process with multiple threads,
+> all the other thread stacks are just leaked: there's
+> nothing else you can reasonably do with them.
+
+`tractor.spawn._main_thread_forkserver`'s module docstring
+(specifically the "What survives the fork? — POSIX
+semantics" section) puts it this way:
+
+> POSIX `fork()` only preserves the *calling* thread as a
+> runnable thread in the child. Every other thread in the
+> parent — trio's runner thread, any `to_thread` cache
+> threads, anything else — never executes another
+> instruction post-fork.
+
+A reader bouncing between the two can be forgiven for
+asking: well, *which* is it — leaked or gone?
+
+The answer is "yes". They're describing the same POSIX
+behavior from two different angles:
+
+- trio is talking about the **bytes** the dead threads
+  leave behind — stacks, TLS slots, per-thread arena
+  metadata — and the fact that nothing in the child can
+  drive them forward, free them, or even safely walk
+  them. That's a memory leak in the strict sense: held
+  but unreachable.
+- tractor is talking about the **execution** side
+  relevant to the forkserver design: which threads
+  retire instructions in the child? Exactly one — the
+  one that called `fork()`. Everything else, regardless
+  of the bytes left behind, is dead in a scheduler
+  sense.
+
+Neither framing is wrong; they're just answering
+different questions.
+
+## POSIX `fork()` in a multi-threaded program — what actually happens
+
+Per POSIX (and concretely on Linux glibc), the contract
+of `fork()` in a multi-threaded process is:
+
+1. The kernel creates a new process whose virtual
+   address space is a COW copy of the parent's. *All*
+   pages map across — code, heap, every thread's stack,
+   every malloc arena, every mmap region.
+2. Of the parent's N threads, exactly **one** is
+   reified in the child as a runnable kernel task: the
+   thread that called `fork()`. The other N-1 threads
+   have *no* corresponding task in the child kernel. They
+   were never scheduled, never `clone()`d for the child,
+   never exist as runnable entities.
+3. Their **memory artifacts** — pthread stacks, TLS,
+   `pthread_t` structures, glibc per-thread arena
+   bookkeeping — are still mapped in the child's address
+   space, because (1) duplicates *everything* page-wise.
+   They sit there as inert COW bytes.
+4. The kernel does not clean those bytes up. There is no
+   "phantom-thread cleanup" pass post-fork. The kernel
+   doesn't know which mapped pages "belonged to" which
+   thread — at the kernel level mappings are
+   process-scoped, not thread-scoped.
+5. The surviving thread (the caller of `fork()`) cannot
+   safely access those leaked bytes either. Any state
+   they encoded — held mutexes, in-flight syscalls,
+   half-updated invariants — is frozen at whatever
+   instant the parent's fork-syscall observed it. Some
+   of those mutexes may even still be locked from the
+   child's POV (the canonical "fork-in-multithreaded-
+   program-deadlocks" hazard; see `man pthread_atfork`).
+
+So: from the kernel's PoV, the child has one thread.
+From the address-space's PoV, the child has all the
+parent's bytes — including the corpses of the N-1 dead
+threads' stacks. Both true simultaneously.
+
+## Why trio says "leaked"
+
+trio's framing makes sense from the parent's
+PoV, looking at *what those threads were doing*. In a
+running `trio.run()` process you typically have:
+
+- The trio runner thread itself — owns the `selectors`
+  epoll fd, the signal-wakeup-fd, the run-queue.
+- Threadpool worker threads (`trio.to_thread`'s cache)
+  — blocked in `wait()` on the threadpool's work
+  condvar.
+- Whatever other ad-hoc threads the application
+  started.
+
+Each of those threads owns *real work-state*: epoll
+registrations, file descriptors held in
+soon-to-be-completed reads, half-released locks, posted
+but unconsumed wakeups. After fork, that state is still
+encoded in the child's memory. None of it is invalid in
+a well-formed-bytes sense. It's just that:
+
+- The thread that was driving it is gone.
+- Nothing else in the child knows the layout well
+  enough to take over.
+- Even if it did, the kernel objects backing the work
+  (epoll fd, signalfd) have separate post-fork
+  semantics that don't compose with userland trio
+  state.
+
+So the bytes are *held* (they're in the child's
+address space, they count against RSS, they survive
+until something clobbers them), and they're
+*unreachable* in any meaningful sense — no thread can
+safely drive them forward. That is the textbook
+definition of a leak.
+
+trio's quote is reminding the user that `fork()` from a
+multi-threaded process is a one-way memory hazard:
+whatever those threads were doing, that work-state is
+now garbage you happen to still be carrying.
+
+## Why tractor says "gone"
+
+tractor's `_main_thread_forkserver` framing is concerned
+with a different question: *which thread executes in the
+child, and is it safe?*
+
+The forkserver design rests on POSIX's "calling thread
+is the sole survivor" guarantee. We pick that calling
+thread very deliberately: a dedicated worker that has
+provably never entered trio. So the thread that *does*
+run in the child is one whose locals, TLS, and stack
+contain nothing trio-related. Trio's runner thread —
+the one that owned the epoll fd and the run-queue — is
+*gone* from the child in the execution sense. It will
+never run another instruction. The fact that its stack
+bytes still exist in the child's address space (the
+"leaked" view) is irrelevant to the forkserver, because
+nothing in the child reads or writes those pages.
+
+So when the docstring says "Every other thread … is
+gone the instant `fork()` returns in the child", it's
+being precise about the surface that matters for the
+backend: scheduler-level liveness. Nothing schedules
+those threads ever again. Whether their bytes are
+hanging around is a separate (and, for the design,
+non-load-bearing) fact.
+
+## Cross-table
+
+The same tabular layout the `_main_thread_forkserver`
+docstring uses, expanded with a fourth "what handles
+it" column:
+
+| thread              | parent    | child (executing) | child (memory)               | what handles it             |
+|---------------------|-----------|-------------------|------------------------------|-----------------------------|
+| forkserver worker   | continues | sole survivor     | live stack                   | runs the child's bootstrap  |
+| `trio.run()` thread | continues | not running       | leaked stack (zombie bytes)  | overwritten by child's fresh `trio.run()` |
+| any other thread    | continues | not running       | leaked stack (zombie bytes)  | overwritten / GC'd / clobbered by `exec()` if used |
+
+The "child (executing)" column is the *execution* side
+of the coin — what tractor cares about. The "child
+(memory)" column is the *memory* side — what trio
+cares about.
+
+The "what handles it" column is the deliberate punchline
+of the design: nothing has to handle the leaked bytes
+*explicitly*. They get clobbered by ordinary forward
+progress in the child:
+
+- The fresh `trio.run()` the child boots up allocates
+  its own stack, scheduler, and run-queue, which over
+  time overlaps and overwrites the inherited zombie
+  pages.
+- Python's GC walks live objects only; the dead-thread
+  Python frames aren't reachable from any
+  `PyThreadState`, so they get freed at the next
+  collection cycle.
+- If the child eventually `exec()`s, the entire address
+  space is replaced and the leak vanishes.
+
+## What this means for the forkserver design
+
+The crucial point is that **the design doesn't and
+*can't* prevent the leak**. There is no userland fix
+for COW thread stacks. The kernel hands the child a
+duplicated address space; that's what `fork()` *is*. No
+amount of pre-fork hookery, `pthread_atfork()`
+gymnastics, or post-fork cleanup can un-COW the dead
+threads' pages without unmapping them, and unmapping
+arbitrary regions of a duplicated address space is
+neither portable nor safe.
+
+What the design *does* ensure is the orthogonal
+property: the survivor thread is one that doesn't need
+any of that leaked state to function. Concretely:
+
+- Survivor is the forkserver worker thread.
+- That worker has provably never imported, called into,
+  or held any reference to `trio`. (Enforced by keeping
+  the worker's lifecycle entirely in
+  `_main_thread_forkserver.py` and never letting trio
+  task-state cross into it.)
+- So the leaked pages — trio runner stack, threadpool
+  caches, etc. — are inert relative to the survivor.
+  No code path in the child references them.
+- The child then boots its own fresh `trio.run()`,
+  which allocates new state in new pages. Over the
+  child's lifetime the COW'd zombie pages get
+  overwritten, GC'd, or (if the child eventually
+  `exec()`s) discarded wholesale.
+
+The "leak" is real but inert. It costs RSS until
+clobbered; it doesn't cost correctness. That's exactly
+the property the forkserver pattern is built on, and
+it's also why the design needs the "calling thread is
+trio-free" precondition to be airtight: if the survivor
+were a trio thread, it *would* try to drive the leaked
+trio state, and the leak would no longer be inert.
+
+## See also
+
+- `tractor/spawn/_main_thread_forkserver.py` — module
+  docstring's "What survives the fork? — POSIX
+  semantics" section is the in-tree, code-adjacent
+  prose this doc expands on. The cross-table here is a
+  fourth-column expansion of the table there.
+
+- [python-trio/trio#1614][trio-1614] — the trio issue
+  with the "leaked" framing, and the canonical thread
+  for trio + `fork()` hazards more broadly.
+
+- [`subint_fork_blocked_by_cpython_post_fork_issue.md`](./subint_fork_blocked_by_cpython_post_fork_issue.md)
+  — sibling analysis covering CPython's *post-fork*
+  hooks (`PyOS_AfterFork_Child`,
+  `_PyInterpreterState_DeleteExceptMain`) and why
+  fork-from-non-main-subint is a CPython-level hard
+  refusal. Complementary axis: this doc is about POSIX
+  semantics; that doc is about the CPython runtime
+  layer that runs *after* POSIX `fork()` returns in
+  the child.
+
+- `man pthread_atfork(3)` — canonical "fork in a
+  multithreaded process is dangerous" reference.
+  Especially the rationale section, which is the
+  closest thing to a normative statement of "the
+  surviving thread cannot safely use anything the dead
+  threads were touching."
+
+- `man fork(2)` (Linux) — "Other than [the calling
+  thread], … no other threads are replicated …"
+  paragraph is the kernel-side statement of the
+  execution-side framing this doc opens with.
+
+[trio-1614]: https://github.com/python-trio/trio/issues/1614

ai/conc-anal/ria_nursery_removal_plan.md

@@ -0,0 +1,463 @@
+# `_ria_nursery` removal plan (issue #477 follow-up)
+
+Goal: drop the secondary "run-in-actor" spawn nursery (and
+friends) from `ActorNursery`/spawn internals, now that
+`tractor.to_actor.run()` delivers one-shot semantics purely on
+the daemon-spawn + portal primitives.
+
+## Verified machinery map (2026-07-02, wkt @ a34aaf98)
+
+The entire mechanism is 4 files:
+
+- `runtime/_supervise.py`
+  - `ActorNursery.__init__(.., ria_nursery, ..)` stores
+    `._ria_nursery` (:202, :238); sole read is
+    `run_in_actor()` passing `nursery=self._ria_nursery`
+    (:442) into `start_actor()`'s `nursery:
+    trio.Nursery|None` escape-hatch param (:305, :367).
+  - `._cancel_after_result_on_exit: set` (:244) marks ria
+    portals (:457).
+  - `_open_and_supervise_one_cancels_all_nursery()` nests
+    `da_nursery` (:609) around `ria_nursery` (:622); the
+    `finally:` at the ria->da boundary (:747-766) raises
+    collected `errors` (single exc or BEG).
+- `runtime/_portal.py`
+  - `._expect_result_ctx` (:112) set by `_submit_for_result()`
+    (:142, sole caller `run_in_actor()`); consumed by
+    `wait_for_result()` (:167) + deprecated `result()` (:220).
+    The `None` branch (:184-196) returns the `NoResult`
+    sentinel (`_exceptions.py:1164`).
+- `spawn/_spawn.py`
+  - `exhaust_portal()` (:129): awaits
+    `portal.wait_for_result()`, CATCHES+RETURNS any exc
+    (never raises).
+  - `cancel_on_completion()` (:177): `exhaust_portal()` ->
+    on exc-result stash `errors[uid] = result` (:203) ->
+    ALWAYS `portal.cancel_actor()` (:218).
+- `spawn/_trio.py` (:195-222) + `spawn/_mp.py` (:187-213),
+  identical shape: after shielded
+  `await an._join_procs.wait()`, open a per-child local
+  nursery; IFF `portal in an._cancel_after_result_on_exit`
+  start `cancel_on_completion` alongside `soft_kill()`; when
+  `soft_kill` returns first, `nursery.cancel_scope.cancel()`
+  reaps the result-waiter.
+
+## The load-bearing semantic (already-deferred errors)
+
+Remote ria-child errors NEVER raise into `ria_nursery`:
+
+1. reaper tasks only START after `_join_procs.set()` (block
+   exit or the inner error handler),
+2. `exhaust_portal` swallows the exc into a return value,
+3. `cancel_on_completion` stashes it in `errors` + cancels
+   that child,
+4. the ria->da `finally:` re-raises collected `errors` (and
+   `an.cancel()`s any daemon stragglers).
+
+So mid-block there is NO error propagation from ria children
+(unless user code explicitly `await portal.wait_for_result()`s)
+— the two-nursery nesting only sequences "reap ria results
+BEFORE blocking on daemon join". A single-nursery impl only
+needs to preserve that sequencing, not any ASAP-cancel
+behavior.
+
+## Target design
+
+### step A: single-nursery `run_in_actor()` (mechanical)
+
+- `run_in_actor()` spawns via the DEFAULT (`_da_nursery`)
+  path — drop `nursery=self._ria_nursery`.
+- rename `._cancel_after_result_on_exit` ->
+  `._ria_portals: dict[portal, Actor]` (need the subactor ref
+  for `cancel_on_completion`).
+- move reaper start-up OUT of the backends into
+  `_open_and_supervise...`: immediately after EACH
+  `an._join_procs.set()` call-site (happy path :642, inner
+  error handler :661), start one
+  `cancel_on_completion(portal, subactor, errors)` task per
+  ria portal into `da_nursery`, then (happy path only)
+  `await` their completion BEFORE falling out of the
+  `try:`/`finally:` that raises `errors` — e.g. gather in a
+  dedicated inner `trio.open_nursery()` block replacing
+  today's `ria_nursery` join point.
+- delete the membership branch + local reaper nursery from
+  `_trio.py`/`_mp.py` (keep the `soft_kill()` call; the
+  per-child local nursery collapses to just `soft_kill`).
+- `_trio.py:310` `_children.pop()` etc. unchanged.
+
+### step B: delete the plumbing
+
+- `_open_and_supervise...`: drop the inner
+  `ria_nursery` + merge its `except BaseException` classify
+  logic into ONE handler on the (now single) nursery scope;
+  `ActorNursery.__init__` loses the `ria_nursery` param.
+- `start_actor()` loses the `nursery:` escape-hatch param
+  (the :302-304 TODO).
+- backends: no more `_cancel_after_result_on_exit` refs.
+
+### step C: (separate PRs) deprecate + migrate + excise
+
+- migrate in-repo `.run_in_actor()` usage to
+  `to_actor.run()`: tests 46 hits/9 files (test_cancellation
+  15, test_infected_asyncio 10, test_spawning 8, registrar 3,
+  adv_streaming 4, pubsub 2, rpc 1, runtime 1), examples 28
+  hits/13 files (debugging/* dominate), docs 20 hits/8 rst
+  files. NOTE: many sites also use deprecated
+  `Portal.result()`/`wait_for_result()` — these die with
+  `_expect_result_ctx`, so migration must land FIRST.
+- add `DeprecationWarning` to `run_in_actor()` (+
+  `_submit_for_result`/`wait_for_result`).
+- final excision: `run_in_actor()`, `_submit_for_result`,
+  `_expect_result_ctx`, `wait_for_result`/`result`,
+  `exhaust_portal`, `cancel_on_completion`, `NoResult`.
+
+## Risk register
+
+1. hard-killed ria child: today the backend-local
+   `nursery.cancel_scope.cancel()` discards a still-parked
+   reaper when the proc dies first; a da_nursery-hosted
+   reaper instead sees the transport break ->
+   `exhaust_portal` returns a `TransportClosed`-ish exc ->
+   NEW entry in `errors` that today gets discarded. Guard:
+   reap-gather block must cancel remaining reapers once all
+   ria procs are dead, or filter transport-death excs for
+   already-`cancel_called` children.
+2. error-path ordering: inner handler today sets
+   `_join_procs` THEN `an.cancel()`; reapers race the
+   cancel-RPC. Keep that ordering when moving reaper spawn.
+3. debugger interplay: `maybe_wait_for_debugger()` calls
+   (:654, :730) must stay BEFORE any reap/cancel issuance.
+4. `errors` double-entry: local body error (:646) + child's
+   relayed exc (via reaper) can both land for the same
+   scenario -> BEG shape changes vs today? (today has the
+   same dual-write sites; keep behavior identical.)
+5. mp backend parity: mirror every `_trio.py` edit in
+   `_mp.py` (identical block).
+
+## Step-A first-probe findings (2026-07-02, WIP in tree)
+
+Step A is IMPLEMENTED (uncommitted):
+`run_in_actor()` spawns via da_nursery; new
+`_supervise._reap_ria_portals()` helper; reap awaited after
+happy-path `_join_procs.set()`; error-path runs reap
+CONCURRENT with `an.cancel()` in the shielded block;
+backends stripped of the membership branch + per-child
+reaper nursery (+ dead imports).
+
+Probe history (trio backend):
+- `tests/test_to_actor.py` + `tests/test_spawning.py`:
+  20/20 PASS — incl. all `run_in_actor()` result
+  round-trips + `test_remote_error` (single erroring child,
+  body re-raise -> inner error path).
+- FIRST attempt ran the error-path reap CONCURRENT with
+  `an.cancel()` (mimicking the old backend-side race):
+  `test_cancellation.py::test_multierror` (2 erroring ria
+  children, body re-raises one) DEADLOCKED. Root cause per
+  the sequencing fix below: reap + cancel must NOT race at
+  this layer (suspected `._children` pop-during-iteration
+  and/or double-cancel RPC wedge; not fully root-caused
+  since the fix removes the race wholesale).
+- FIX (2nd attempt, current impl): error path SEQUENCES:
+  (1) snapshot ria `(portal, subactor)` pairs (backend
+  `finally`s pop `._children` as procs reap), (2)
+  `await an.cancel()`, (3) bounded reap over the snapshot.
+  Bound was first 3s -> blew the `fail_after` deadline in
+  `test_cancel_while_childs_child_in_sync_sleep` (hard-
+  killed grandchild never relays => reaper parks the full
+  bound). Tightened to 0.5s: anything collectable is
+  already queued in the local ctx (relayed BEFORE the
+  cancel); a parked reaper self-cleans (`trio.Cancelled`
+  results are never stashed).
+- RESULT: `tests/test_cancellation.py` FULLY GREEN
+  (20 passed, 1 xfailed, 77s); full-suite gate run kicked
+  off same session (see final report/next session).
+
+Remaining risk: on slow CI a relayed-but-undelivered error
+racing the 0.5s bound could drop an `errors` entry
+(BEG-shape flake); if observed, scale the bound via the
+`cpu_perf_headroom()`-style approach or peek
+`Portal._final_result_msg`/ctx queue state instead of
+time-bounding.
+
+## Step-B outcome (2026-07-02, done in tree)
+
+Step A landed as `5cd190c5` (code) + `99310269` (docs).
+Step B implemented on top (uncommitted):
+
+- `._ria_nursery` is GONE — the inner
+  `async with (collapse_eg(), trio.open_nursery() as
+  ria_nursery)` layer in
+  `_open_and_supervise_one_cancels_all_nursery` is deleted;
+  `da_nursery` is now the single nursery for ALL subactors.
+- `ActorNursery.__init__` drops the `ria_nursery` param +
+  the `self._ria_nursery` attr; `start_actor()` drops its
+  `nursery=` escape-hatch param (uses `self._da_nursery`
+  directly).
+- `._cancel_after_result_on_exit` STAYS — it's the
+  ria-child discriminator for `_reap_ria_portals()`.
+
+Deliberately NOT done (deferred to its own higher-risk PR,
+flagged with a TODO at the outer `except`): merging the two
+error handlers into one. Rationale — collapsing the empty
+nursery is provably behavior-preserving (a zero-task
+`trio.open_nursery()` only adds a checkpoint), whereas the
+inner `except BaseException` (swallow-into-`errors`) and
+outer `except (...)` (re-raise, safety-net for the inner
+handler's own non-shielded awaits) have DIFFERENT
+semantics; merging changes error/cancel propagation and
+wants isolated review + its own gate. Both handlers are
+kept, now nested directly under the single nursery.
+
+Why the collapse is safe: post-step-A NOTHING spawns into
+`ria_nursery` (its only reader, `run_in_actor`'s
+`nursery=self._ria_nursery`, was removed in A; the stored
+attr was never read again). So the layer was pure dead
+weight.
+
+Gate (trio backend, all 0-failure):
+- targeted set (`test_cancellation test_spawning test_local
+  test_rpc test_to_actor`) = 49 passed, 1 xfailed.
+- tail set (`test_reg_err_types remote_exc_relay
+  resource_cache ringbuf root_infect_asyncio root_runtime
+  runtime shm task_broadcasting trioisms trionics/`) = 63
+  passed, 1 skipped, 5 xfailed.
+- full-suite head ~73% (subdirs + `test_2way`..`test_pubsub`)
+  = 303 passed before the known-flaky `test_dynamic_pub_sub`
+  TooSlowError stall (pre-existing; same hang in the step-A
+  full run). Suite ran slow this session (~13min vs 555s
+  cold, likely thermal from back-to-back runs), never
+  completing within an 800s bound — but split across the
+  above three runs EVERY module passed under step B.
+
+## Step-B2 outcome (2026-07-02, done in tree)
+
+Step B committed as `9201a2ed` (code) + `d2e812fb` (docs), then
+branched to `drop_ria_nursery`. Step B2 (the deferred
+handler-merge) implemented on top (uncommitted):
+
+- the two nested handlers in
+  `_open_and_supervise_one_cancels_all_nursery` collapse to
+  ONE `except BaseException as _scope_err` + the existing
+  `finally`. The `outer_err`/`inner_err` locals go away.
+
+Why it's safe (trace, not hope): the OLD inner handler records
+`errors[actor.aid.uid]` as its FIRST statement (before any
+await). So whenever an error path runs, `errors` is non-empty.
+The OLD outer handler was only reachable via leakage from the
+inner handler (it catches `BaseException`, so nothing from the
+`yield` scope bypasses it) — and by then `errors` is already
+populated, so the `finally`'s `raise` from `errors` ALWAYS
+superseded the outer handler's own `raise`. i.e. the outer
+`raise` was dead. The outer handler's other effects
+(`_scope_error`, a 2nd debugger-wait, child-cancel) are
+redundant with the merged handler + `finally`. So one handler
++ `finally` is observably equivalent.
+
+Residual nuance (accepted): in the rare "`trio.Cancelled`
+delivered during the non-shielded `maybe_wait_for_debugger`"
+path, the merged form may leave `_cancel_called` False (cancel
+happens after the wait), so `open_nursery`'s tb-hiding guard
+(`not cancel_called and _scope_error`) can show a tb it
+previously hid. More informative, not less; no test asserts on
+it.
+
+Gate (box ran ~2.7x slow this session, load-induced
+`TooSlowError` flakiness on timing tests — NOT code; see
+[[env_cpu_throttle_masquerades_as_regression]]):
+- baseline (pre-B2 tip `9201a2ed`) full suite
+  (`-k 'not dynamic_pub_sub'`) = 300 passed + 1
+  `test_ext_types_over_ipc` `TooSlowError` that passes 6/6 in
+  isolation (4.89s).
+- B2 error/cancel gate (`test_cancellation remote_exc_relay
+  inter_peer_cancellation advanced_faults oob_cancellation
+  to_actor spawning local rpc`) = 71 passed, 1 xfailed
+  (125s).
+- B2 full-suite run: see `b2_full.log` (result appended on
+  completion). RECOMMEND a clean full-suite run on a
+  normal-speed box before this merges.
+
+## Regression + fix: ria-reap hang (2026-07-02)
+
+Human hit a full-suite hang on
+`test_infected_asyncio.py::test_tractor_cancels_aio`. Bisected:
+passes at pre-ria `a34aaf98` (0.59s), hangs at B2 `e617b498`
+(90s+). Root-caused to the STEP-A reaper hoist (`5cd190c5`),
+NOT B2 (`_reap_ria_portals` is byte-identical A->B2).
+
+Bug: the test does `run_in_actor(asyncio_actor)` then a USER
+`portal.cancel_actor()` and exits the block cleanly -> the
+happy path's `await _reap_ria_portals()`, which waits UNBOUNDED
+on `cancel_on_completion -> wait_for_result()`. The child was
+cancelled out-of-band so no final result is relayed -> parked
+forever. The OLD spawn-backend reaper was raced against
+`soft_kill()` (per-child nursery `cancel_scope.cancel()` on
+subproc death); the hoist dropped that race.
+
+Fix: `_reap_ria_portals()` runs each `cancel_on_completion()`
+in a local nursery alongside a `proc.poll()` death-watch that
+cancels the parked reaper once the subproc exits — restoring
+the old race, backend-agnostic (guarded by
+`hasattr(proc, 'poll')` for a future `subint` handle).
+
+Why POLL (`proc.poll()`) not the event-driven `wait_func`:
+the mp waiter (`_spawn.proc_waiter`) does
+`wait_readable(proc.sentinel)`, and `soft_kill()` is ALREADY
+awaiting that same fd concurrently in the daemon nursery — a
+2nd `wait_readable` on one fd raises `trio.BusyResourceError`.
+(`trio.Process.wait()` IS multi-waiter-safe, but mp has no
+async equivalent.) `proc.poll()` — the same liveness check
+`soft_kill` itself falls back to — is the conflict-free common
+denominator. Verified: poll-fix passes on BOTH trio and
+mp_spawn.
+
+Also added a per-test anti-hang guard: wrapped
+`test_tractor_cancels_aio`'s `main()` in
+`with trio.fail_after(9 * cpu_perf_headroom())` — the blessed
+pattern (`pytest-timeout`'s global cap is intentionally off;
+breaks trio under fork backends, see `pyproject` NOTE). So a
+future recurrence FAILS FAST instead of hanging the suite.
+(Several other tests in the file are still guardless —
+`test_aio_simple_error`, `test_trio_error_cancels_intertask_chan`,
+`test_aio_errors_and_channel_propagates_and_closes` — candidate
+follow-up sweep.)
+
+Lesson: the B2 focused gate OMITTED `test_infected_asyncio`
+(and the full runs were clipped/slow), so the step-A hang
+slipped through. Any future ria-touching change MUST gate
+`test_infected_asyncio` explicitly.
+
+Gate: `test_tractor_cancels_aio` green (trio 1.53s, mp 3.98s);
+fix gate (`test_infected_asyncio test_cancellation test_to_actor
+test_spawning`) = 74 passed, 3 xfailed, 0 failures.
+
+## PAUSED (2026-07-02): re-assess the reaper's SCOPE
+
+User's insight (compelling — likely the real root cause of
+the hang, not just the missing proc-death race):
+
+> the "hoisting" of 5cd190c5 was just not really done right
+> — the hoist should have been into the `to_actor` scope,
+> not `_supervise`.
+
+The argument: `.run_in_actor()`'s result-waiting/reaping got
+hoisted into `_supervise._reap_ria_portals` (nursery-machinery
+scope), which has NO natural cancel-scope to bound a parked
+`wait_for_result()` — hence the awkward proc-death race +
+the poll-vs-`proc_waiter` dilemma. If the result-wait instead
+lived in the `to_actor` one-shot scope
+(`to_actor._invoke_in_subactor()`), it would sit right next to
+the caller's `an` + a local `trio` task-nursery + cancel-scope
+(the `trio.to_thread`-style model #477 actually wants) — so
+bounding/cancelling the wait is trivial and the hang
+dissolves from correct scoping rather than a bolt-on race.
+
+Follow-on to re-evaluate on resume:
+- should `_reap_ria_portals` exist AT ALL, or should
+  result-waiting move entirely into
+  `to_actor._invoke_in_subactor()`?
+- reimplement legacy `run_in_actor()` on top of
+  `to_actor.run()` so `_reap_ria_portals` +
+  `_cancel_after_result_on_exit` can be DROPPED from
+  `_supervise` entirely (the true #477 simplification)?
+- the poll-vs-event decision is MOOT under this re-scoping.
+
+State at pause: `test_infected_asyncio` anti-hang guard
+COMMITTED (`d1fb4a1a`, intentionally red w/o the fix — the
+user's failing-test-first convention). The poll-based reap
+fix in `_supervise.py` is UNCOMMITTED and likely SUPERSEDED
+by the re-scoping — do NOT land it as-is.
+
+## RESOLVED (2026-07-06): migrate everything, remove the API
+
+The PAUSED re-assessment concluded decisively: rather than
+re-scope `_reap_ria_portals` (or bolt any hack onto it), the
+`run_in_actor()` API itself was REMOVED — its non-blocking
+"result at teardown" semantic predates streaming and confused
+more than it served. Every in-repo caller was migrated
+per-file/-group (each its own commit, each gated):
+
+- tests: `test_infected_asyncio` `test_runtime` `test_rpc`
+  `test_spawning` `test_pubsub` `test_registrar`
+  `test_cancellation` (3 groups) `test_advanced_streaming`.
+- examples: 4 non-debugging + all 8 `debugging/` REPL scripts
+  (debugger suite byte-identical green, 28p/6s).
+- docs: 8 rst pages + the `experimental/_pubsub` docstring.
+
+Migration patterns (the `run_in_actor` shape -> successor):
+
+- blocking result        -> `to_actor.run(fn, an=an, ...)`
+- fire-&-forget/forever  -> bg `to_actor.run()` task in a local
+                            `trio` task-nursery (or `start_actor`
+                            + bg `Portal.run()` when a portal
+                            handle is needed)
+- concurrent fan-out     -> N bg `to_actor.run()` tasks / or
+                            `gather_contexts([p.open_context(..)])`
+- reap-all-error-collect -> the "collect don't cancel" pattern:
+                            each one-shot catches + stashes its
+                            `RemoteActorError`, group raised
+                            after the task-nursery joins (see
+                            `examples/debugging/multi_subactors.py`)
+- mutual-rendezvous      -> peers must OUTLIVE both dialogs:
+                            `start_actor()` daemons + concurrent
+                            `Portal.run()`s + explicit
+                            `an.cancel()` (eager one-shot reap
+                            races the slower peer's dial of the
+                            winner's dead sockaddr; found via
+                            `test_trynamic_trio` flake).
+
+Semantic deltas (tests loosened accordingly):
+
+- teardown-reap-all BEG-of-N is GONE: local task-nurseries are
+  cancel-on-first, raced siblings' `Cancelled`s are absorbed,
+  and the runtime's `collapse_eg()` unwraps every single-member
+  group at each actor boundary — a fully-raced nested tree
+  relays a bare (annotated) `RemoteActorError` chain.
+- `test_multierror_fast_nursery`'s obsolete BEG-of-25 assertion
+  deleted; `test_concurrent_start_error_reaps_all` retains its
+  high-fan-out startup/cancel/reap stress under caller-scoped
+  semantics.
+- `test_nested_multierrors` re-purposed separately as deep-tree
+  cancel-cascade stress w/ a race-tolerant shape walk.
+
+Final excision (after zero callers remained): `run_in_actor()`,
+`._cancel_after_result_on_exit`, `_reap_ria_portals()`,
+`Portal._submit_for_result/._expect_result_ctx/
+.wait_for_result()/.result()`, `exhaust_portal()`,
+`cancel_on_completion()`, `NoResult` — net -402 lines. The
+reap-hang class (unbounded `wait_for_result` in machinery
+scope) dissolves structurally: the only result-wait left lives
+in the caller's task inside its own cancel-scope; the
+`d1fb4a1a` anti-hang guard test passes by construction. The
+poll-vs-`proc_waiter` debate is moot as predicted.
+
+## Follow-up sketch: `to_actor.open_one_shot()` (run-async parity)
+
+If deferred-result parity is ever wanted, the design that needs
+NO runtime coupling, NO returned `Portal` and NO cancel-relay
+`trio.Event` machinery:
+
+    async with to_actor.open_one_shot(
+        fn, an=an, **kws,
+    ) as one_shot:
+        ...                       # concurrent caller work
+        val = await one_shot.wait()   # optional; errors always
+                                      # propagate at scope exit
+
+an `@acm` that opens a private task-nursery, `start_soon`s ONE
+task running the existing blocking `run()` and stashes the
+value in a slot + sets a done-`trio.Event` (a memo, not a
+cancel relay). Cancellation = plain scope-cancel of the acm's
+nursery (the parked `Portal.run()` unwinds via `Cancelled`, the
+shielded `cancel_actor()` reap still runs); a child error
+raises into the acm scope so an un-`wait()`ed one-shot can
+never silently drop its error. i.e. the old reaper's job is
+done by scoping, not machinery. ~40 lines, all in
+`to_actor/_api.py`, zero `_supervise` involvement.
+
+## Verification gate
+
+- per-migration-commit module gates on `trio` (+ `mp_spawn`
+  spot-gates incl. `test_infected_asyncio` per the B2 lesson);
+  `tests/devx/test_debugger.py` for the REPL flows.
+- full suite on `trio` + `mp_spawn` at branch tip + CI matrix
+  via draft PR #484.

ai/conc-anal/spawn_time_boot_death_dup_name_issue.md

@@ -0,0 +1,142 @@
+# Spawn-time boot-death (`rc=2`) under rapid same-name spawn against a registrar
+
+## Symptom
+
+Spawning N (≥4) sub-actors with the **same name** in tight
+succession against a daemon registrar surfaces as
+`ActorFailure: Sub-actor (...) died during boot (rc=2)
+before completing parent-handshake`.
+
+```
+tests/discovery/test_multi_program.py
+  ::test_dup_name_cancel_cascade_escalates_to_hard_kill[n_dups=4]
+```
+
+```
+tractor._exceptions.ActorFailure:
+  Sub-actor ('doggy', '<uuid>') died during boot (rc=2)
+  before completing parent-handshake.
+    proc: <_ForkedProc pid=<n> returncode=None>
+```
+
+The `proc` repr shows `returncode=None` because the repr is
+captured before `proc.wait()` returns; the actual
+`os.WEXITSTATUS == 2` is reported via `result['died']` in the
+race-helper.
+
+## When it surfaces
+
+- N=2 (`n_dups=2`): **always passes**.
+- N=4 (`n_dups=4`): **consistent fail** under both `tpt-proto=tcp`
+  and `tpt-proto=uds`, MTF backend.
+- N=8 (`n_dups=8`): **passes** (counter-intuitive — see "racing
+  windows").
+- Non-MTF backends: not yet exercised systematically.
+
+## What previously masked it
+
+Pre the spawn-time `wait_for_peer_or_proc_death` race-helper
+(in `tractor.spawn._spawn`), the parent's `start_actor` flow
+ended with a bare:
+
+```python
+event, chan = await ipc_server.wait_for_peer(uid)
+```
+
+That awaits an unsignalled `trio.Event` on `_peer_connected[uid]`.
+If the sub-actor process **dies during boot** (before its
+runtime executes the parent-callback handshake that sets the
+event), the wait parks forever. The dead proc becomes a zombie
+because no one ever calls `proc.wait()` to reap it.
+
+In test contexts the failure presented as a hang or a much
+later `trio.TooSlowError` from an outer `fail_after`. In
+production it'd present as a parent that never makes progress
+past `start_actor`. The death itself was silently masked.
+
+## What surfaces it now
+
+`tractor.spawn._spawn.wait_for_peer_or_proc_death` (used by
+`_main_thread_forkserver_proc`) races the handshake-wait
+against `proc.wait()`. The race-helper raises `ActorFailure`
+on death-first instead of parking, exposing the rc=2.
+
+## Hypothesis: registrar-side same-name contention
+
+The test spawns N actors with name `doggy` sequentially:
+
+```python
+for i in range(n_dups):
+    p: Portal = await an.start_actor('doggy')
+    portals.append(p)
+```
+
+Each spawned doggy:
+
+1. Forks via the forkserver.
+2. Boots its runtime in `_actor_child_main`.
+3. Connects back to the parent for handshake.
+4. Connects to the daemon registrar to call `register_actor`.
+5. Enters its RPC msg-loop.
+
+Step (4) is where the same-name contention lives. The
+registrar's `register_actor` (in
+`tractor.discovery._registry`) accepts duplicate names
+(stores `(name, uuid) -> addr`), but its internal bookkeeping
+may have a non-trivial check (e.g. `wait_for_actor` resolution,
+`_addrs2aids` map updates) that errors out under specific
+ordering between the existing entry and the incoming one.
+
+`rc=2 == os.WEXITSTATUS == 2` corresponds to `sys.exit(2)`
+in the doggy process — typically reached via an unhandled
+exception that's translated to exit code 2 by Python's top-
+level (e.g. `argparse` errors use 2; `SystemExit(2)` etc.).
+So the doggy is hitting an explicit exit path during
+`register_actor` or just-after.
+
+The non-monotonic shape (N=2 OK, N=4 BAD, N=8 OK) suggests a
+specific timing window — likely "the 3rd register-RPC arrives
+while the 1st-or-2nd is in some intermediate state". With
+N=8, the additional procs widen the registration spread
+enough that no two land in the conflicting window.
+
+## Where to dig next
+
+- Add per-actor logging in `_actor_child_main` and
+  `register_actor` to surface the actual exception that
+  triggers the rc=2 exit. Currently the doggy dies before
+  the parent ever sees its stderr (forkserver doesn't
+  marshal child stdio back).
+- Race-test the registrar's `register_actor` /
+  `unregister_actor` /  `wait_for_actor` against same-name
+  concurrent calls in isolation (no spawn).
+- Consider whether `register_actor` should be idempotent
+  under same-name re-register or should explicitly reject
+  same-name (and ideally with a clear `RemoteActorError`,
+  not `sys.exit(2)`).
+
+## Test-suite handling
+
+Currently:
+
+- `tests/discovery/test_multi_program.py
+  ::test_dup_name_cancel_cascade_escalates_to_hard_kill[n_dups=4]`
+  is `pytest.mark.xfail(strict=False, reason=...)` to keep
+  the suite green while this issue is investigated.
+- `n_dups=2` and `n_dups=8` continue to validate the
+  cancel-cascade hard-kill escalation.
+
+Once the underlying race is understood + fixed, drop the
+xfail.
+
+## Related work
+
+- The cancel-cascade fix that introduced this regression
+  test:
+  `tractor/_exceptions.py:ActorTooSlowError`,
+  `tractor/runtime/_supervise.py:_try_cancel_then_kill`,
+  `tractor/runtime/_portal.py:Portal.cancel_actor(
+   raise_on_timeout=...)`.
+- The spawn-time death-detection that exposed this:
+  `tractor/spawn/_spawn.py:wait_for_peer_or_proc_death`,
+  used by `tractor/spawn/_main_thread_forkserver.py`.

ai/conc-anal/test_register_duplicate_name_daemon_connect_race_issue.md

@@ -0,0 +1,273 @@
+# `test_register_duplicate_name` racy connect-failure on `daemon` fixture readiness
+
+## Symptom
+
+`tests/test_multi_program.py::test_register_duplicate_name`
+fails intermittently under BOTH transports + ALL spawn
+backends with connect-refused errors:
+
+```
+# under --tpt-proto=uds
+FAILED tests/test_multi_program.py::test_register_duplicate_name
+- ConnectionRefusedError: [Errno 111] Connection refused
+( ^^^ this exc was collapsed from a group ^^^ )
+
+# under --tpt-proto=tcp
+FAILED tests/test_multi_program.py::test_register_duplicate_name
+- OSError: all attempts to connect to 127.0.0.1:36003 failed
+( ^^^ this exc was collapsed from a group ^^^ )
+```
+
+Distinct from the cancel-cascade `TooSlowError` flake
+class — see
+`cancel_cascade_too_slow_under_main_thread_forkserver_issue.md`.
+This is a **connect-time race** before the daemon is
+fully ready to `accept()`, not a teardown-cascade
+slowness.
+
+## Root cause: blind `time.sleep()` in `daemon` fixture
+
+`tests/conftest.py::daemon` boots a sub-py-process via
+`subprocess.Popen([python, '-c', 'tractor.run_daemon(...)'])`,
+then **blindly sleeps** a fixed delay before yielding
+`proc` to the test:
+
+```python
+# excerpt from tests/conftest.py::daemon
+proc = subprocess.Popen([
+    sys.executable, '-c', code,
+])
+
+bg_daemon_spawn_delay: float = _PROC_SPAWN_WAIT  # 0.6
+if tpt_proto == 'uds':
+    bg_daemon_spawn_delay += 1.6
+if _non_linux and ci_env:
+    bg_daemon_spawn_delay += 1
+
+# XXX, allow time for the sub-py-proc to boot up.
+# !TODO, see ping-polling ideas above!
+time.sleep(bg_daemon_spawn_delay)
+
+assert not proc.returncode
+yield proc
+```
+
+Inherent fragility: the delay is "long enough on dev
+boxes most of the time" but has no actual
+synchronization with the daemon's `bind()` + `listen()`
+completion. Under any of:
+
+- Loaded box (CI parallelism, big rebuild in
+  background, low-cpu-freq)
+- Cold first-run (`importlib` cache miss, JIT warmup)
+- Higher-than-expected `tractor` import cost
+- Filesystem latency (UDS sockfile create, slow
+  tmpfs)
+
+...the sleep finishes BEFORE the daemon has bound its
+listen socket → first test client call to
+`tractor.find_actor()` / `wait_for_actor()` /
+`open_nursery(registry_addrs=[reg_addr])`'s implicit
+connect → `ConnectionRefusedError` (TCP) or
+`FileNotFoundError`/`ConnectionRefusedError` (UDS).
+
+## Reproducer
+
+Easiest: run the suite under load.
+
+```bash
+# create CPU pressure on another core in parallel
+stress-ng --cpu 2 --timeout 600s &
+
+./py313/bin/python -m pytest \
+  tests/test_multi_program.py::test_register_duplicate_name \
+  --spawn-backend=main_thread_forkserver \
+  --tpt-proto=tcp -v
+```
+
+Reproduces ~30-50% of the time on a dev laptop. On a
+quiet idle box, may need 5-10 runs to hit.
+
+## Why the existing `_PROC_SPAWN_WAIT` tuning is
+inadequate
+
+Recent `bg_daemon_spawn_delay` rename
+(de-monotonic-grow fix) just-shipped removed the
+*accumulation* bug where each invocation made the
+NEXT test's wait longer too. Net effect: every
+invocation now uses the SAME `0.6 + 1.6` (UDS) or
+`0.6` (TCP) sleep, no growth. Good — but does
+NOTHING for the underlying race. Each individual
+test still relies on a blind sleep that may or may
+not be sufficient.
+
+Bumping the constant higher pushes flake rate down
+but never to zero AND adds dead time to every
+non-flaking run. Not a fix, just a knob.
+
+## Side effects
+
+- **Inter-test cascade**: a single failure can cascade
+  via leaked subprocesses (the `daemon` fixture's
+  cleanup may not fully tear down a daemon that never
+  reached "ready"). The `_reap_orphaned_subactors`
+  session-end + `_track_orphaned_uds_per_test`
+  per-test fixtures handle most of this now, but the
+  affected test itself still fails.
+- **Worsens under fork-spawn backends**: the daemon
+  has more init work
+  (`_main_thread_forkserver`-coordinator-thread
+  startup, etc.) so the sleep has to cover MORE.
+
+## Fix design — replace blind sleep with active poll
+
+The right primitive is **poll the daemon's bind
+address until it accepts a connection or we time
+out**, with the timeout being a hard ceiling rather
+than a baseline. Two implementation paths:
+
+### Path A — TCP/UDS connect-poll loop
+
+Try `socket.connect(reg_addr)` in a tight loop with
+short backoff (~50ms), succeed on the first non-error
+return, fail-loud on a hard cap (e.g. 10s). Same
+primitive works for both transports because both use
+`socket.connect()` semantics.
+
+Rough shape:
+
+```python
+def _wait_for_daemon_ready(
+    reg_addr,
+    tpt_proto: str,
+    timeout: float = 10.0,
+    poll_interval: float = 0.05,
+) -> None:
+    deadline = time.monotonic() + timeout
+    while True:
+        if tpt_proto == 'tcp':
+            sock = socket.socket(socket.AF_INET)
+            target = reg_addr  # (host, port)
+        else:  # uds
+            sock = socket.socket(socket.AF_UNIX)
+            target = os.path.join(*reg_addr)
+        try:
+            sock.settimeout(poll_interval)
+            sock.connect(target)
+        except (
+            ConnectionRefusedError,
+            FileNotFoundError,
+            socket.timeout,
+        ) as exc:
+            if time.monotonic() >= deadline:
+                raise TimeoutError(
+                    f'Daemon never accepted on {target!r} '
+                    f'within {timeout}s'
+                ) from exc
+            time.sleep(poll_interval)
+        else:
+            sock.close()
+            return
+```
+
+Pros: trivial primitive, no tractor-runtime
+dependency, works pre-yield in the fixture body,
+fail-fast on truly-broken daemon.
+Cons: doesn't actually do an IPC handshake, just
+proves listen-side is up. A daemon that bound but
+hasn't initialized its registrar table yet would
+still race.
+
+### Path B — `tractor.find_actor()` poll
+
+Use the actual discovery API the test would call:
+
+```python
+async def _wait_for_daemon_ready_via_discovery(
+    reg_addr,
+    timeout: float = 10.0,
+    poll_interval: float = 0.05,
+):
+    deadline = trio.current_time() + timeout
+    async with tractor.open_root_actor(
+        registry_addrs=[reg_addr],
+        # ephemeral root just for the probe
+    ):
+        while True:
+            try:
+                async with tractor.find_actor(
+                    'registrar',  # daemon's own name
+                    registry_addrs=[reg_addr],
+                ) as portal:
+                    if portal is not None:
+                        return
+            except Exception:
+                pass
+            if trio.current_time() >= deadline:
+                raise TimeoutError(...)
+            await trio.sleep(poll_interval)
+```
+
+Pros: actually proves the discovery path works,
+handles the "bound but not ready" case naturally.
+Cons: requires booting an ephemeral root actor JUST
+for the probe (overhead), more code, and runs in trio
+which complicates the sync-fixture context. Need a
+`trio.run()` wrapper.
+
+### Recommended: Path A with optional handshake check
+
+Path A is much simpler + handles 95% of the bug
+class. If "bound-but-not-ready" turns out to still
+race (it shouldn't — `tractor.run_daemon` doesn't
+return from `bind()` until the registrar is
+fully populated), escalate to Path B as a focused
+follow-up.
+
+## Workarounds (until fix lands)
+
+1. **Bump `_PROC_SPAWN_WAIT`** higher (current: 0.6).
+   2.0–3.0 hides most flakes at the cost of adding
+   dead time to every test. Not a fix but reduces
+   blast radius while the proper poll lands.
+2. **`pytest-rerunfailures`** with `reruns=1` on the
+   `daemon` fixture's tests specifically. Hides the
+   flake but doesn't address it.
+3. **Mark known-affected tests as `xfail(strict=False)`**
+   under `--ci`. Lets CI go green at the cost of
+   silently hiding regressions.
+
+(Recommend skipping all three — implement the active
+poll instead.)
+
+## Investigation next steps
+
+1. Implement Path A as a `_wait_for_daemon_ready()`
+   helper in `tests/conftest.py`. Replace the
+   `time.sleep(bg_daemon_spawn_delay)` call with it.
+2. Drop the `_PROC_SPAWN_WAIT` constant entirely
+   (active poll obsoletes blind sleep).
+3. Run the suite 5-10 times to validate flake rate
+   drops to 0.
+4. If flakes persist, profile whether the daemon
+   process exits with non-zero before the poll's
+   deadline hits — that'd be a different bug
+   (daemon startup crash) that the blind sleep was
+   masking.
+5. Cross-check `tests/test_multi_program.py::test_*`
+   — multiple tests use the `daemon` fixture; all
+   should benefit from the same poll primitive.
+
+## Related
+
+- `tests/conftest.py::daemon` — the fixture under
+  fix
+- `tests/conftest.py::_PROC_SPAWN_WAIT` — the
+  constant to drop
+- `cancel_cascade_too_slow_under_main_thread_forkserver_issue.md`
+  — distinct flake class (cancel-cascade
+  `TooSlowError` at teardown, not connect-time race)
+- `trio_wakeup_socketpair_busy_loop_under_fork_issue.md`
+  — different bug entirely; this race was masked
+  pre-WakeupSocketpair-patch by the busy-loop
+  hangs.

ai/conc-anal/trio_033_cancel_cascade_slowdown_depth3_issue.md

@@ -0,0 +1,102 @@
+# `trio` 0.29 -> 0.33 slows the depth=3 cancel-cascade
+
+## Symptom
+
+After locking to `trio==0.33.0` (commit `c7741bba`, was
+`0.29.0`), this test reliably trips its `fail_after`
+deadline on the **`trio`** backend:
+
+```
+FAILED tests/test_cancellation.py::test_nested_multierrors[start_method=trio-depth=3]
+  - AssertionError: assert False
+    where False = isinstance(
+      Cancelled(source='deadline', source_task=None, reason=None),
+      tractor.RemoteActorError,
+    )
+```
+
+A `fail_after_w_trace` hang-snapshot is captured for the
+test each run (deadline-injected `Cancelled` wrapped into
+the actor-nursery `BaseExceptionGroup`).
+
+## Root cause (immediate)
+
+The test budgets `fail_after(6)` for the `trio` backend.
+That 6s was chosen (commit `32955db0`, while `trio==0.29`)
+with the assertion that trio finishes "well under" 6s.
+The `trio` 0.29 -> 0.33 bump slowed the depth=3 cascade
+past that budget, so the 6s deadline now fires mid-cascade.
+
+trio 0.33 added **cancel-reason tracking** — every
+`Cancelled` now carries `(source=, reason=, source_task=)`.
+The injected exc is `Cancelled(source='deadline')`, i.e.
+trio itself naming our `fail_after(6)` scope as the cancel
+origin. When that `Cancelled` collapses one branch of the
+nursery BEG, the test's `isinstance(subexc,
+RemoteActorError)` assertion fails. The healthy outcome is
+`BEG = [RemoteActorError, RemoteActorError]`; the
+`Cancelled` is purely an artifact of the deadline cutting
+the cascade short.
+
+## Measurements (standalone, this machine)
+
+```
+depth=1  trio   ~3.15s   PASS  (keeps 6s budget)
+depth=3  trio   ~6.8-8.2s  FAIL @ 6s  (now bumped to 12s)
+```
+
+depth=1 still fits comfortably; only depth=3 (deeper
+recursive spawn-and-error tree => more actors to reap)
+exceeds the old budget. The ~2s/depth-level cost looks
+like serialized per-actor reap / `terminate_after` waits.
+
+## Mitigation applied
+
+`test_nested_multierrors` now splits the `trio` budget:
+
+```python
+case ('trio', 1):
+    timeout = 6
+case ('trio', 3):
+    timeout = 12   # was 6; see this doc
+```
+
+This stops the deadline from firing so the cascade
+completes naturally to `[RAE, RAE]`.
+
+## Also affected — same root cause, different test
+
+`test_echoserver_detailed_mechanics[trio-raise_error=KeyboardInterrupt]`
+(`tests/test_infected_asyncio.py`) tripped the *same*
+slowdown via its much tighter `trio` budget of `1s`. The
+single-aio-subactor teardown now takes ~1s, so the `1s`
+`fail_after` raced the deadline (PASS at 0.99s / FAIL at
+1.03s across back-to-back standalone runs). On a deadline-
+fire the injected `Cancelled(source='deadline')` wraps the
+mid-stream `KeyboardInterrupt` into a `BaseExceptionGroup`,
+which is NOT a `KeyboardInterrupt` so the bare
+`pytest.raises(KeyboardInterrupt)` fails. (The sibling
+`raise_error=Exception` variant only "passes" by accident:
+an `ExceptionGroup` *is-a* `Exception`, so its
+`pytest.raises(Exception)` still matches even when wrapped.)
+
+Mitigation: bump that `trio` budget `1 -> 4s` (matching the
+forking-spawner case). Without a deadline-fire the KBI
+propagates bare and the assertion passes.
+
+## Open follow-up (the actual regression)
+
+The budget bump is a band-aid — the underlying question is
+**why** the depth=3 `trio` cancel-cascade went from <6s to
+~7-8s across `trio` 0.29 -> 0.33. Candidate avenues:
+
+- which scope owns the per-actor `terminate_after` wait,
+  and are the tree's reaps concurrent or serialized?
+- did trio 0.33's abort/reschedule or cancel-reason
+  bookkeeping change checkpoint timing on the cancel path?
+
+If/when the cascade speeds back up under-budget, depth=3
+will start completing well under 12s — at which point the
+budget can be tightened back toward 6s as a regression
+tripwire. Related (different backend, same cascade class):
+`cancel_cascade_too_slow_under_main_thread_forkserver_issue.md`.

ai/conc-anal/trio_wakeup_socketpair_busy_loop_under_fork_issue.md

@@ -0,0 +1,221 @@
+# trio `WakeupSocketpair.drain()` busy-loop in forked child (peer-closed missed-EOF)
+
+## Reproducer
+
+```bash
+./py313/bin/python -m pytest \
+  tests/test_multi_program.py::test_register_duplicate_name \
+  --tpt-proto=tcp \
+  --spawn-backend=main_thread_forkserver \
+  -v --capture=sys
+```
+
+Subactor pegs a CPU core indefinitely; parent test
+hangs waiting for the subactor.
+
+## Empirical evidence (caught alive)
+
+```
+$ sudo strace -p <subactor-pid>
+recvfrom(6, "", 65536, 0, NULL, NULL)   = 0
+recvfrom(6, "", 65536, 0, NULL, NULL)   = 0
+recvfrom(6, "", 65536, 0, NULL, NULL)   = 0
+... (no `epoll_wait`, no other syscalls, just this back-to-back)
+```
+
+Pattern: tight C-level `recvfrom` loop returning 0
+each call. No `epoll_wait` between iterations →
+**not trio's task scheduler**. Pure synchronous C
+loop.
+
+```
+$ sudo readlink /proc/<subactor-pid>/fd/6
+socket:[<inode>]
+
+$ sudo lsof -p <subactor-pid> | grep ' 6u'
+<cmd> <pid> goodboy 6u unix 0xffff... 0t0 <inode> type=STREAM (CONNECTED)
+```
+
+fd=6 is an **AF_UNIX socket** in CONNECTED state.
+Even though the test uses `--tpt-proto=tcp`, this fd
+is NOT a tractor IPC channel — it's an internal
+trio socketpair.
+
+## Root-cause: `WakeupSocketpair.drain()`
+
+`/site-packages/trio/_core/_wakeup_socketpair.py`:
+
+```python
+class WakeupSocketpair:
+    def __init__(self) -> None:
+        self.wakeup_sock, self.write_sock = socket.socketpair()
+        self.wakeup_sock.setblocking(False)
+        self.write_sock.setblocking(False)
+        ...
+
+    def drain(self) -> None:
+        try:
+            while True:
+                self.wakeup_sock.recv(2**16)
+        except BlockingIOError:
+            pass
+```
+
+`socket.socketpair()` on Linux defaults to AF_UNIX
+SOCK_STREAM. Both ends non-blocking. Normal flow:
+
+1. Signal/wake event → `write_sock.send(b'\x00')`
+   queues a byte.
+2. `wakeup_sock` becomes readable → trio's epoll
+   triggers.
+3. Trio calls `drain()` to flush the buffer.
+4. drain loops on `wakeup_sock.recv(64KB)`.
+5. Eventually buffer empty → non-blocking socket
+   raises `BlockingIOError` → except → break.
+
+**Bug surface — peer-closed missed-EOF**:
+
+Non-blocking socket semantics:
+- buffer has data → `recv` returns N>0 bytes (loop continues)
+- buffer empty → `recv` raises `BlockingIOError`
+- **peer FIN'd → `recv` returns 0 bytes (NEITHER exception NOR
+  break — infinite tight loop)**
+
+`drain()` does not handle the `b''` return-value
+(EOF) case. If `write_sock` has been closed (or the
+process holding it is gone), every iteration returns
+0 → infinite loop → 100% CPU on a single core.
+
+## Why this triggers under `main_thread_forkserver`
+
+Under `os.fork()` from the forkserver-worker thread:
+
+1. Parent has a `WakeupSocketpair` instance with
+   `wakeup_sock=fdN`, `write_sock=fdM`. Both fds
+   open in parent.
+2. Fork → child inherits BOTH fds (kernel-level fd
+   table dup).
+3. `_close_inherited_fds()` runs in child →
+   closes everything except stdio. `wakeup_sock` and
+   `write_sock` of the parent's `WakeupSocketpair`
+   ARE closed in child.
+4. Child's trio (running fresh) creates its OWN
+   `WakeupSocketpair` → NEW fd numbers (e.g. fd 6, 7).
+5. **In `infect_asyncio` mode** the asyncio loop is
+   the host; trio runs as guest via
+   `start_guest_run`. trio still creates its
+   `WakeupSocketpair` in the I/O manager but its
+   role is different.
+
+The race window: somewhere between (3) and (5), if a
+`WakeupSocketpair` Python object reference inherited
+via COW (from parent's pre-fork heap) survives long
+enough that `drain()` is called on it AFTER its fds
+were closed but BEFORE the child's NEW socketpair
+takes over the recycled fd numbers — the recycled fd
+will be one of the child's NEW socketpair ends, whose
+peer might be FIN-flagged (e.g. parent-process
+peer-end is closed).
+
+Or simpler: the `wait_for_actor`/`find_actor` discovery
+flow in `test_register_duplicate_name` triggers an
+unusual code path where a stale `WakeupSocketpair`
+gets `drain()`-called on a fd whose peer has already
+closed.
+
+## Why `drain()` shouldn't loop indefinitely on EOF
+(upstream trio bug)
+
+Even WITHOUT fork, `drain()` should treat `b''` as
+EOF and break. The current code is correct for the
+"buffer drained on a healthy socketpair" scenario but
+incorrect for the "peer is gone" scenario. It's a
+defensive-programming gap in trio.
+
+A one-line patch upstream:
+
+```python
+def drain(self) -> None:
+    try:
+        while True:
+            data = self.wakeup_sock.recv(2**16)
+            if not data:
+                break  # peer-closed; nothing more to drain
+    except BlockingIOError:
+        pass
+```
+
+## Workarounds (until the underlying issue lands)
+
+1. **Skip-mark on the fork backend**:
+   `tests/test_multi_program.py` →
+   `pytest.mark.skipon_spawn_backend('main_thread_forkserver',
+   reason='trio WakeupSocketpair.drain busy-loop, see ai/conc-anal/trio_wakeup_socketpair_busy_loop_under_fork_issue.md')`.
+
+2. **Defensive monkey-patch in tractor's
+   forkserver-child prelude** — wrap
+   `WakeupSocketpair.drain` to handle `b''`:
+
+   ```python
+   # in `_actor_child_main` or `_close_inherited_fds`'s
+   # post-fork prelude:
+   from trio._core._wakeup_socketpair import WakeupSocketpair
+   _orig_drain = WakeupSocketpair.drain
+   def _safe_drain(self):
+       try:
+           while True:
+               data = self.wakeup_sock.recv(2**16)
+               if not data:
+                   return  # peer closed
+       except BlockingIOError:
+           pass
+   WakeupSocketpair.drain = _safe_drain
+   ```
+
+   Tracks upstream — remove once trio fixes.
+
+3. **Upstream the fix**: 1-line PR to `python-trio/trio`
+   adding `if not data: break` to `drain()`.
+
+## Investigation next steps
+
+1. **Confirm via py-spy**: when caught alive, detach
+   strace first then
+   `sudo py-spy dump --pid <subactor> --locals`. The
+   busy thread should show `drain` from `WakeupSocketpair`
+   in the call chain.
+2. **Identify which write-end peer is closed**: from
+   the inode of fd 6, look up the matching peer
+   inode via `ss -xp` and see whose process it
+   was/is.
+3. **Verify the missed-EOF hypothesis**: hand-craft a
+   minimal `WakeupSocketpair` repro:
+
+   ```python
+   from trio._core._wakeup_socketpair import WakeupSocketpair
+   ws = WakeupSocketpair()
+   ws.write_sock.close()  # simulate peer-gone
+   ws.drain()             # should hang forever
+   ```
+
+## Sibling bug
+
+`tests/test_infected_asyncio.py::test_aio_simple_error`
+hangs under the same backend with a DIFFERENT
+fingerprint (Mode-A deadlock, both parties in
+`epoll_wait`, no busy-loop). Distinct root cause —
+see `infected_asyncio_under_main_thread_forkserver_hang_issue.md`.
+
+Both share the broader theme: **trio internal-state
+initialization isn't fully fork-safe under
+`main_thread_forkserver`** for the more exotic
+dispatch paths.
+
+## See also
+
+- [#379](https://github.com/goodboy/tractor/issues/379) — subint umbrella
+- python-trio/trio#1614 — trio + fork hazards
+- `trio._core._wakeup_socketpair.WakeupSocketpair`
+  source (the smoking gun)
+- `ai/conc-anal/fork_thread_semantics_execution_vs_memory.md`
+- `ai/conc-anal/infected_asyncio_under_main_thread_forkserver_hang_issue.md`

ai/prompt-io/claude/20260406T172848Z_02b2ef1_prompt_io.md

@@ -0,0 +1,54 @@
+---
+model: claude-opus-4-6
+service: claude
+session: (ad-hoc, not tracked via conf.toml)
+timestamp: 2026-04-06T17:28:48Z
+git_ref: 02b2ef1
+scope: tests
+substantive: true
+raw_file: 20260406T172848Z_02b2ef1_prompt_io.raw.md
+---
+
+## Prompt
+
+User asked to extend `tests/test_resource_cache.py` with a test
+that reproduces the edge case fixed in commit `02b2ef18` (per-key
+locking+user tracking in `maybe_open_context()`). The bug was
+originally triggered in piker's `brokerd.kraken` backend where the
+same `acm_func` was called with different kwargs, and the old
+global `_Cache.users` counter caused:
+
+- teardown skipped for one `ctx_key` bc another key's users kept
+  the global count > 0
+- re-entry hitting `assert not resources.get(ctx_key)` during the
+  teardown window
+
+User requested a test that would fail under the old code and pass
+with the fix.
+
+## Response summary
+
+Designed and implemented `test_per_ctx_key_resource_lifecycle`
+which verifies per-`ctx_key` resource isolation by:
+
+1. Holding resource `'a'` open in a bg task
+2. Opening+closing resource `'b'` (same `acm_func`, different
+   kwargs) while `'a'` is still alive
+3. Re-opening `'b'` and asserting cache MISS — proving `'b'` was
+   torn down independently despite `'a'` keeping its own user
+   count > 0
+
+With the old global counter, phase 3 would produce a stale cache
+HIT (leaked resource) or crash on the assert.
+
+Also added a trivial `acm_with_resource(resource_id)` ACM helper
+at module level.
+
+## Files changed
+
+- `tests/test_resource_cache.py` — add `acm_with_resource` ACM +
+  `test_per_ctx_key_resource_lifecycle` test fn
+
+## Human edits
+
+None — committed as generated (pending user review).

ai/prompt-io/claude/20260406T193125Z_85f9c5d_prompt_io.md

@@ -0,0 +1,57 @@
+---
+model: claude-opus-4-6
+service: claude
+session: (ad-hoc, not tracked via conf.toml)
+timestamp: 2026-04-06T19:31:25Z
+git_ref: 85f9c5d
+scope: tests
+substantive: true
+raw_file: 20260406T193125Z_85f9c5d_prompt_io.raw.md
+---
+
+## Prompt
+
+User asked to reproduce the exact `assert not resources.get(ctx_key)`
+crash originally seen in piker's `brokerd.kraken` backend via
+`open_cached_client('kraken')`. Key constraints from user:
+
+- In piker, kwargs were the **same** (empty) so all callers
+  share one `ctx_key = (fid, ())`
+- The root issue is `_Cache.locks` being indexed by `fid`
+  rather than `ctx_key`, plus the race window between
+  `values.pop()` and `resources.pop()` in `_Cache.run_ctx`
+  (the acm `__aexit__` has checkpoints in between)
+- Mark the test `@pytest.mark.xfail` so it can be committed
+  against the unpatched code and later adjusted once the fix
+  from `9e49eddd` is applied
+
+User pointed to piker's `open_cached_client('kraken')` usage
+as the real-world pattern to model.
+
+## Response summary
+
+Added `test_moc_reentry_during_teardown` which deterministically
+reproduces the teardown race:
+
+1. A `cached_client()` acm (no params, like
+   `kraken.api.get_client()`) signals via `in_aexit` event
+   when its `__aexit__` starts, then sleeps to hold the
+   window open
+2. `use_and_exit` task enters+exits the cached resource
+3. `reenter_during_teardown` waits on `in_aexit` then
+   re-enters — landing exactly in the window where
+   `values` is popped but `resources` still exists
+4. The old `assert not resources.get(ctx_key)` fires
+5. `@pytest.mark.xfail(raises=AssertionError)` catches it
+
+Test confirmed: `XFAIL` with the exact traceback at
+`_mngrs.py:327`.
+
+## Files changed
+
+- `tests/test_resource_cache.py` — add
+  `test_moc_reentry_during_teardown` xfail test fn
+
+## Human edits
+
+None — committed as generated (pending user review).

ai/prompt-io/claude/20260406T222648Z_f086222_prompt_io.md

@@ -0,0 +1,52 @@
+---
+model: claude-opus-4-6
+service: claude
+session: (ad-hoc, not tracked via conf.toml)
+timestamp: 2026-04-06T22:26:48Z
+git_ref: f086222
+scope: code
+substantive: true
+raw_file: 20260406T222648Z_f086222_prompt_io.raw.md
+---
+
+## Prompt
+
+User applied the per-key locking patch (`f086222d`),
+removed the `@pytest.mark.xfail` from
+`test_moc_reentry_during_teardown`, and found the
+test STILL fails — now with `RuntimeError` instead
+of `AssertionError`. Asked to diagnose why the fix
+doesn't actually resolve this race.
+
+## Response summary
+
+Traced the exact scheduling sequence across three
+concurrent tasks to show that the per-`ctx_key`
+locking change doesn't close the fundamental gap:
+
+- `maybe_open_context`'s finally pops the lock and
+  sets `no_more_users` in the **exiting caller's
+  task**
+- But the actual resource cleanup (`values.pop`,
+  acm `__aexit__`, `resources.pop`) runs in the
+  **`run_ctx` task** inside `service_tn`
+- A re-entering task creates a fresh lock (old one
+  popped), acquires immediately, and hits stale
+  `resources[ctx_key]` while `run_ctx` is still in
+  `__aexit__`
+
+Identified three possible fix strategies:
+1. Let `run_ctx` own lock release/pop after
+   `resources.pop()`
+2. Add a "teardown complete" event that new callers
+   wait on
+3. Move `resources.pop()` into `run_ctx`'s inner
+   finally (before acm `__aexit__`)
+
+## Files changed
+
+No code changes — analysis only.
+
+## Human edits
+
+N/A — diagnostic output, no code generated.

ai/prompt-io/claude/20260406T222648Z_f086222_prompt_io.raw.md

@@ -0,0 +1,32 @@
+---
+model: claude-opus-4-6
+service: claude
+timestamp: 2026-04-06T22:26:48Z
+git_ref: f086222
+diff_cmd: git diff f086222~1..f086222
+---
+
+Here's the problem. The per-key locking doesn't fix this race because the **lock lifetime is decoupled from `run_ctx`'s lifetime**. Trace through what happens:
+
+**`maybe_open_context` finally** (lines 468-495):
+1. `users[ctx_key] -= 1` → 0
+2. `no_more_users.set()` → wakes `run_ctx` (but it doesn't run yet, just becomes ready)
+3. `_Cache.locks.pop(ctx_key)` → **lock is gone**
+4. `maybe_open_context` exits
+
+**`run_ctx`** now gets scheduled (lines 237-241):
+5. Inner finally: `values.pop(ctx_key)` → values gone
+6. `mng.__aexit__()` runs → `in_aexit.set()`, `await trio.sleep(10)` → **CHECKPOINT**
+
+**`reenter_during_teardown`** wakes:
+7. Enters `maybe_open_context` → no lock exists for `ctx_key` → creates **fresh** lock, acquires immediately
+8. `values[ctx_key]` → `KeyError`
+9. `resources.get(ctx_key)` → **still exists!** (step 6 hasn't reached `resources.pop()` in `run_ctx`'s outer finally yet)
+10. `raise RuntimeError('Caching resources ALREADY exist?!')`
+
+The core issue: `no_more_users.set()` (step 2) and `locks.pop()` (step 3) happen in the **exiting caller's task**, but the actual resource cleanup (`values.pop`, acm `__aexit__`, `resources.pop`) happens in the **`run_ctx` task** inside `service_tn`. There's no synchronization between "lock removed, new callers welcome" and "run_ctx has actually finished tearing down". The per-`ctx_key` lock change just made the lock more granular — it didn't close this gap.
+
+The fix needs to ensure a new caller can't acquire a lock for `ctx_key` until `run_ctx` has fully exited (including `resources.pop()`). Either:
+- **don't pop the lock in `maybe_open_context`** — let `run_ctx` release/pop it after `resources.pop()`
+- **wait on a "teardown complete" event** in the cache-miss path when stale `resources` exist
+- **pop `resources` before the acm `__aexit__`** (move `resources.pop()` into the inner finally alongside `values.pop()`)

ai/prompt-io/claude/20260413T192116Z_f851f28_prompt_io.md

@@ -0,0 +1,74 @@
+---
+model: claude-opus-4-6
+service: claude
+session: (ad-hoc, not tracked via conf.toml)
+timestamp: 2026-04-13T19:21:16Z
+git_ref: f851f28
+scope: code
+substantive: true
+---
+
+## Prompt
+
+User provided a detailed implementation plan for
+a new `test_tpt_bind_addrs.py` test suite covering
+`open_root_actor(tpt_bind_addrs=...)` — the three
+runtime code paths in `_root.py:385-450`:
+
+1. Non-registrar, no explicit bind -> random addrs
+2. Registrar, no explicit bind -> registry_addrs
+3. Explicit bind given -> `wrap_address()` + merge
+
+Plan specified 6 test functions (~10 parametrized
+variants), predicted a type-mixing bug at line 446,
+and asked for an in-flight fix if confirmed.
+
+## Response summary
+
+Created `tests/discovery/test_tpt_bind_addrs.py`
+with 9 collected test variants across 6 functions:
+
+- `test_registrar_root_tpt_bind_addrs` (3 variants:
+  `bind-eq-reg`, `bind-subset-reg`,
+  `bind-disjoint-reg`)
+- `test_non_registrar_root_tpt_bind_addrs`
+  (2 variants: `bind-same-bindspace`,
+  `bind-disjoint`)
+- `test_non_registrar_default_random_bind`
+  (baseline, no explicit bind)
+- `test_tpt_bind_addrs_as_maddr_str`
+  (multiaddr string input)
+- `test_registrar_merge_binds_union`
+  (registrar + disjoint bind -> union)
+- `test_open_nursery_forwards_tpt_bind_addrs`
+  (`open_nursery(**kwargs)` forwarding)
+
+Confirmed and fixed the predicted bug at
+`_root.py:446`: the registrar merge path mixed
+`Address` objects (`tpt_bind_addrs`) with raw tuples
+(`uw_reg_addrs`) inside `set()`, preventing
+deduplication and causing double-bind `OSError`.
+
+Fix: wrap `uw_reg_addrs` before the set union:
+```python
+# before (broken)
+tpt_bind_addrs = list(set(
+    tpt_bind_addrs + uw_reg_addrs
+))
+# after (fixed)
+tpt_bind_addrs = list(set(
+    tpt_bind_addrs
+    + [wrap_address(a) for a in uw_reg_addrs]
+))
+```
+
+All 9 tests pass after the fix.
+
+## Files changed
+
+- `tests/discovery/test_tpt_bind_addrs.py` (new)
+- `tractor/_root.py:446` (bug fix, 1 line)
+
+## Human edits
+
+N/A — pending review.

ai/prompt-io/claude/20260413T205048Z_269d939c_prompt_io.md

@@ -0,0 +1,50 @@
+---
+model: claude-opus-4-6
+service: claude
+session: 76154e65-d8e1-4b5f-9275-0ea45ba7e98a
+timestamp: 2026-04-13T20:50:48Z
+git_ref: 269d939c
+scope: code
+substantive: true
+raw_file: 20260413T205048Z_269d939c_prompt_io.raw.md
+---
+
+## Prompt
+
+Implement a `parse_endpoints()` API in
+`tractor.discovery._multiaddr` that lets downstream
+projects declare per-actor transport bind addresses
+as a service table of actor-name -> multiaddr
+strings (loaded from config, e.g. TOML `[network]`
+section). Add type aliases `EndpointsTable` and
+`ParsedEndpoints`, plus 7 unit tests covering TCP,
+UDS, tuples, mixed input, unsupported protocols, and
+edge cases.
+
+A detailed implementation plan was provided
+specifying insertion points, type signatures, test
+names, and verification commands.
+
+## Response summary
+
+Add `parse_endpoints()` function with two supporting
+type aliases to `_multiaddr.py`. The function
+iterates a `dict[str, list[str|tuple]]` service
+table and delegates each entry to the existing
+`wrap_address()` helper, which already handles maddr
+strings, raw tuples, and pre-wrapped `Address`
+objects. Seven pure unit tests (no trio runtime)
+validate all input variants and edge cases. All 22
+multiaddr tests pass.
+
+## Files changed
+
+- `tractor/discovery/_multiaddr.py` — add
+  `EndpointsTable`, `ParsedEndpoints` type aliases
+  and `parse_endpoints()` function after line 133
+- `tests/discovery/test_multiaddr.py` — add
+  `parse_endpoints` import and 7 new test functions
+
+## Human edits
+
+None — committed as generated.

ai/prompt-io/claude/20260413T205048Z_269d939c_prompt_io.raw.md

@@ -0,0 +1,32 @@
+---
+model: claude-opus-4-6
+service: claude
+timestamp: 2026-04-13T20:50:48Z
+git_ref: 269d939c
+diff_cmd: git diff 269d939c~1..269d939c
+---
+
+## Implementation output
+
+### `tractor/discovery/_multiaddr.py`
+
+> `git diff 269d939c~1..269d939c -- tractor/discovery/_multiaddr.py`
+
+Type aliases `EndpointsTable` and `ParsedEndpoints`
+plus `parse_endpoints()` helper that wraps each
+entry via `wrap_address()`.
+
+### `tests/discovery/test_multiaddr.py`
+
+> `git diff 269d939c~1..269d939c -- tests/discovery/test_multiaddr.py`
+
+Add import of `parse_endpoints` to existing import
+block, then 7 new test functions.
+
+### Test run output
+
+```
+22 passed, 1 warning in 0.05s
+```
+
+All 22 tests pass (15 existing + 7 new).

ai/prompt-io/claude/20260414T163300Z_befedc49_prompt_io.md

@@ -0,0 +1,38 @@
+---
+model: claude-opus-4-6
+service: claude
+session: multiaddr-support-rename-prefer
+timestamp: 2026-04-14T16:33:00Z
+git_ref: befedc49
+scope: code
+substantive: true
+raw_file: 20260414T163300Z_befedc49_prompt_io.raw.md
+---
+
+## Prompt
+
+Create a helper function that determines the best transport given
+actor locality (distributed vs same host). Use PID/hostname
+comparison for locality detection, apply at registry addr selection
+only (not spawn-time).
+
+## Response summary
+
+New `prefer_addr()` + `_is_local_addr()` helpers
+in `_api.py` using `socket.getaddrinfo()` and
+`ipaddress` for PID/hostname locality detection.
+Preference: UDS > local TCP > remote TCP.
+Integrated into `query_actor()` and
+`wait_for_actor()`. Also changed
+`Registrar.find_actor()` to return full addr list
+so callers can apply preference.
+
+## Files changed
+
+- `tractor/discovery/_discovery.py` → `_api.py`
+  — renamed + added `prefer_addr()`,
+  `_is_local_addr()`; updated `query_actor()` and
+  `wait_for_actor()` call sites
+- `tractor/discovery/_registry.py`
+  — `Registrar.find_actor()` returns
+  `list[UnwrappedAddress]|None`

ai/prompt-io/claude/20260414T163300Z_befedc49_prompt_io.raw.md

@@ -0,0 +1,62 @@
+---
+model: claude-opus-4-6
+service: claude
+timestamp: 2026-04-14T16:33:00Z
+git_ref: befedc49
+diff_cmd: git diff befedc49~1..befedc49
+---
+
+### `tractor/discovery/_api.py`
+
+> `git diff befedc49~1..befedc49 -- tractor/discovery/_api.py`
+
+Add `_is_local_addr()` and `prefer_addr()` transport
+preference helpers.
+
+#### `_is_local_addr(addr: Address) -> bool`
+
+Determines whether an `Address` is reachable on the
+local host:
+
+- `UDSAddress`: always returns `True`
+  (filesystem-bound, inherently local)
+- `TCPAddress`: checks if `._host` is a loopback IP
+  via `ipaddress.ip_address().is_loopback`, then
+  falls back to comparing against the machine's own
+  interface IPs via
+  `socket.getaddrinfo(socket.gethostname(), None)`
+
+#### `prefer_addr(addrs: list[UnwrappedAddress]) -> UnwrappedAddress`
+
+Selects the "best" transport address from a
+multihomed actor's address list. Wraps each
+candidate via `wrap_address()` to get typed
+`Address` objects, then classifies into three tiers:
+
+1. **UDS** (same-host guaranteed, lowest overhead)
+2. **TCP loopback / same-host IP** (local network)
+3. **TCP remote** (only option for distributed)
+
+Within each tier, the last-registered (latest) entry
+is preferred. Falls back to `addrs[-1]` if no
+heuristic matches.
+
+### `tractor/discovery/_registry.py`
+
+> `git diff befedc49~1..befedc49 -- tractor/discovery/_registry.py`
+
+`Registrar.find_actor()` return type broadened from
+single addr to `list[UnwrappedAddress]|None` — full
+addr list lets callers apply transport preference.
+
+#### Integration
+
+`query_actor()` and `wait_for_actor()` now call
+`prefer_addr(addrs)` instead of `addrs[-1]`.
+
+### Verification
+
+All discovery tests pass (13/13 non-daemon).
+`test_local.py` and `test_multi_program.py` also
+pass (daemon fixture teardown failures are
+pre-existing and unrelated).

ai/prompt-io/claude/20260417T034918Z_9703210_prompt_io.md

@@ -0,0 +1,101 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+session: subints-spawner-design-kickoff
+timestamp: 2026-04-17T03:49:18Z
+git_ref: 9703210
+scope: docs
+substantive: true
+raw_file: 20260417T034918Z_9703210_prompt_io.raw.md
+---
+
+## Prompt
+
+Drive the "first big boi, from GH issue" task seeded by
+`ai/prompt-io/prompts/subints_spawner.md`: design, plan
+and implement sub-interpreter (subint) spawn-backend
+support per issue #379, including (1) modularizing
+`tractor.spawn._spawn` into per-backend submods, (2) a new
+`._subint` backend, and (3) harness parametrization via the
+existing `--spawn-backend` / `start_method` pytest fixture
+in `tractor._testing.pytest`.
+
+Follow-up clarifications from the user (this turn):
+1. Pin `<3.15` on this dev branch and feature-gate subint
+   tests — chose option (a).
+2. Split Phase A (modularization) into its own PR first.
+3. Defer the `fork()`-via-subint hack to a follow-up.
+4. Harness flag is `pytest --spawn-backend <key>` CLI →
+   `start_method` fixture (prompt file updated to match).
+
+## Response summary
+
+Produced a three-phase plan and a concrete Phase A (pure
+modularization) file-split plan for user review; no code
+written yet — the green-light to start Phase A was given
+in this same turn conditional on logging this prompt-io
+entry first.
+
+Phases:
+- **A — modularize** `tractor/spawn/_spawn.py` (847 LOC):
+  keep generic machinery in `_spawn.py`, extract
+  `trio_proc` → `spawn/_trio.py`, `mp_proc` →
+  `spawn/_mp.py`. No pin bump.
+- **B — `_subint` backend**: bump `pyproject.toml`
+  `requires-python` upper to `<3.15`; add `'subint'` to
+  `SpawnMethodKey`; reuse existing UDS transport; shm
+  escape-hatch deferred.
+- **C — harness**: drive the valid-backend tuple in
+  `tractor/_testing/pytest.py:345-349` from
+  `typing.get_args(SpawnMethodKey)`; skip subint tests on
+  Python < 3.14.
+
+Key findings surfaced to the user:
+- `pyproject.toml:12` currently pins `<3.14`; PEP 734
+  `concurrent.interpreters` only ships in 3.14 — the
+  load-bearing constraint.
+- `_testing/pytest.py:345-349` hardcodes valid backends
+  as a string tuple (`'mp_spawn'`, `'mp_forkserver'`,
+  `'trio'`) — should be `get_args(SpawnMethodKey)`.
+- `_testing/pytest.py:228` already imports
+  `try_set_start_method` from `tractor.spawn._spawn` —
+  keeping the `_spawn.py` path as the "core" module
+  avoids breaking external refs during Phase A.
+
+Phase A commit breakdown (each green via `/run-tests`):
+1. Drive `_testing/pytest.py` backend-tuple from
+   `get_args(SpawnMethodKey)` — prep.
+2. Extract `trio_proc` → `spawn/_trio.py`; `_methods`
+   registry imports it.
+3. Extract `mp_proc` → `spawn/_mp.py`; `_methods` imports
+   it.
+4. Tidy `spawn/__init__.py` docstring for the new layout.
+
+Workflow: Phase A runs in a new worktree via `/open-wkt
+spawn_modularize` (branch `wkt/spawn_modularize`) and
+lands as its own PR before Phase B/C returns to the
+current `claude_bun_opus4.7_subint_spawner_backend`
+branch.
+
+## Files changed
+
+None this turn — design/planning only. The user updated
+`ai/prompt-io/prompts/subints_spawner.md` themselves to
+complete the truncated sentence on line 32 (harness-flag
+reference).
+
+Upcoming (Phase A, next turn):
+- `tractor/spawn/_spawn.py` — trimmed to "core"
+- `tractor/spawn/_trio.py` — new; holds `trio_proc()`
+- `tractor/spawn/_mp.py` — new; holds `mp_proc()`
+- `tractor/spawn/__init__.py` — docstring refresh
+- `tractor/_testing/pytest.py` — backend tuple from
+  `get_args(SpawnMethodKey)`
+
+## Human edits
+
+None — committed as generated. User completed the
+truncated sentence on line 32 of
+`ai/prompt-io/prompts/subints_spawner.md` themselves
+*before* this prompt-io log was written; that edit was
+the user's authoring, not an edit to AI output.

ai/prompt-io/claude/20260417T034918Z_9703210_prompt_io.raw.md

@@ -0,0 +1,117 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+timestamp: 2026-04-17T03:49:18Z
+git_ref: 9703210
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+No code was produced in this turn — this is a design/planning
+exchange for subint spawner support per issue #379. Captured
+verbatim below are the two substantive AI responses, which
+together fix the phased plan (A: modularize, B: subint, C:
+harness) and the concrete Phase A file split before any
+implementation begins.
+
+## Turn 1 — initial triage + gating questions
+
+> Read: issue #379, current `spawn/_spawn.py` (847 lines),
+> and pyproject pins.
+>
+> ## Task summary
+> Design/plan/implement subint backend per #379, plus
+> modularize `spawn/_spawn.py` into per-backend submods, plus
+> harness flag for test parametrization.
+>
+> ## Key finding
+> `pyproject.toml:12` pins `requires-python = ">=3.12, <3.14"`
+> — but `concurrent.interpreters` / PEP 734 **only lands in
+> 3.14**. This is the load-bearing design question before
+> coding.
+>
+> ## Proposed phasing (incremental commits)
+> - **Phase A — modularize** `_spawn.py`: extract `trio_proc`
+>   → `spawn/_trio.py`, `mp_proc` → `spawn/_mp.py`, keep
+>   generic helpers (`hard_kill`/`soft_kill`/`new_proc`
+>   dispatcher/`SpawnMethodKey`) in a slim core. Tests green
+>   at each step.
+> - **Phase B — `_subint` backend**: extend `SpawnMethodKey`
+>   with `'subint'`; reuse UDS transport by default; shm is
+>   escape-hatch follow-up.
+> - **Phase C — harness**: add backend-parametrize flag, skip
+>   subint when Python < 3.14.
+>
+> ## Gating questions before I start
+> 1. **Python pin**: bump `<3.15` and feature-gate subint
+>    tests (a), require 3.14 outright (b), or backport shim
+>    (c)? I'd recommend (a).
+> 2. **PR strategy**: one branch, commits per phase (current
+>    `claude_bun_opus4.7_subint_spawner_backend`)? Or split
+>    Phase A as its own PR first?
+> 3. **`fork()`-via-subint hack** from the issue's "Our own
+>    thoughts" section — v1 or follow-up? I'd defer to
+>    follow-up.
+> 4. **Prompt file line 32 is truncated** — what's the
+>    existing harness flag name you had in mind ("using the
+>    existing ___")?
+>
+> Which of these are already decided vs. open?
+
+## User answers (for the raw record)
+
+1. bump `<3.15`, feature-gate — option (a).
+2. split Phase A as its own PR first.
+3. defer `fork()`-via-subint hack.
+4. harness flag is `--spawn-backend` CLI → `start_method`
+   fixture, per `tractor._testing._pytest`. Prompt file
+   updated accordingly.
+
+## Turn 2 — concrete Phase A plan
+
+> Good — prompt file now ends cleanly; harness flag
+> confirmed as `--spawn-backend` CLI → `start_method`
+> fixture (via `_testing/pytest.py:193,355`). Also noted
+> `pytest.py:345-349` hardcodes valid backends as a string
+> tuple — `typing.get_args(SpawnMethodKey)` should drive
+> that list.
+>
+> ## Phase A — concrete plan (separate PR)
+>
+> **File split** in `tractor/spawn/`:
+> - `_spawn.py` stays as "core": keeps `SpawnMethodKey`,
+>   `_spawn_method`, `proc_waiter`, `try_set_start_method`,
+>   `exhaust_portal`, `cancel_on_completion`, `hard_kill`,
+>   `soft_kill`, `new_proc` dispatcher, `_methods`
+>   registry. Kept at this path so `_testing/pytest.py:228`
+>   and other external imports don't break.
+> - `_trio.py` ← `trio_proc()` (lines 448–671, ~224 lines)
+> - `_mp.py` ← `mp_proc()` (lines 672–842, ~171 lines)
+> - `_methods` dict imports the two backend fns from the
+>   new modules.
+>
+> **Commits** (each green via `/run-tests`):
+> 1. Drive `_testing/pytest.py` backend-tuple from
+>    `get_args(SpawnMethodKey)` — tiny prep.
+> 2. Extract `trio_proc` → `spawn/_trio.py`; `_methods`
+>    imports it.
+> 3. Extract `mp_proc` → `spawn/_mp.py`; `_methods`
+>    imports it.
+> 4. Tidy `spawn/__init__.py` docstring to describe new
+>    layout.
+>
+> No pyproject pin bump in Phase A (that's Phase B).
+>
+> **Workflow**: I'd run this in a new worktree via
+> `/open-wkt spawn_modularize` (branch
+> `wkt/spawn_modularize`), then open a PR before returning
+> to the current `claude_bun_opus4.7_subint_spawner_backend`
+> branch for Phase B/C.
+>
+> Green-light to `/open-wkt` and start Phase A?
+
+## User final response
+
+> all this first then yes, green light.
+
+— log this `/prompt-io` entry first, then proceed with
+Phase A in a worktree.

ai/prompt-io/claude/20260417T035800Z_61a73ba_prompt_io.md

@@ -0,0 +1,86 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+session: subints-spawner-phase-a-impl
+timestamp: 2026-04-17T03:58:05Z
+git_ref: 61a73ba
+scope: code
+substantive: true
+raw_file: 20260417T035800Z_61a73ba_prompt_io.raw.md
+---
+
+## Prompt
+
+Follow-up from the design session logged in
+`20260417T034918Z_9703210_prompt_io.md`: the user
+greenlit Phase A after `/prompt-io` tracking, so
+execute the Phase A "modularize `spawn._spawn`" plan
+end-to-end in a new worktree — four logical steps
+(pytest harness prep, `trio_proc` extraction,
+`mp_proc` extraction, docstring refresh), each green
+via `/run-tests`.
+
+User directives during execution:
+- **Option 3** for the test cadence ("continue A.2–A.4
+  first, then run the full suite once at the end of
+  Phase A").
+- **One commit** for the whole phase ("can't we just
+  commit the whole patch in one commit?") instead of
+  the 3/4-commit split I initially proposed.
+- **Don't pre-draft** commit messages — wait for the
+  user to invoke `/commit-msg` (captured as feedback
+  memory `feedback_no_auto_draft_commit_msgs.md`).
+
+## Response summary
+
+Produced the cohesive Phase A modularization patch,
+landed as commit `61a73bae` (subject: `Mv
+trio_proc`/`mp_proc` to per-backend submods`). Five
+files changed, +565 / -418 lines.
+
+Key pieces of the patch (generated by claude,
+reviewed by the human before commit):
+- `tractor/spawn/_trio.py` — **new**; receives
+  `trio_proc()` verbatim from `_spawn.py`; imports
+  cross-backend helpers back from `._spawn`.
+- `tractor/spawn/_mp.py` — **new**; receives
+  `mp_proc()` verbatim; uses `from . import _spawn`
+  for late-binding access to the mutable `_ctx` /
+  `_spawn_method` globals (design decision made
+  during impl, not the original plan).
+- `tractor/spawn/_spawn.py` — shrunk 847 → 448 LOC;
+  import pruning; bottom-of-module late imports for
+  `trio_proc` / `mp_proc` with a one-line comment
+  explaining the circular-dep reason.
+- `tractor/spawn/__init__.py` — docstring refresh
+  describing the new layout.
+- `tractor/_testing/pytest.py` — the valid-backend
+  set now comes from `typing.get_args(SpawnMethodKey)`
+  so future additions (`'subint'`) don't need harness
+  edits.
+
+## Files changed
+
+See `git diff 61a73ba~1..61a73ba --stat`:
+
+```
+ tractor/_testing/pytest.py |  12 +-
+ tractor/spawn/__init__.py  |  31 +++-
+ tractor/spawn/_mp.py       | 235 ++++++++++++++++++++++++
+ tractor/spawn/_spawn.py    | 413 +-------------------------------
+ tractor/spawn/_trio.py     | 292 ++++++++++++++++++++++++++++
+ 5 files changed, 565 insertions(+), 418 deletions(-)
+```
+
+Validation:
+- import probe + `_methods` wiring check — OK
+- spawn-relevant test subset — 37 passed, 1 skipped
+- full suite — 350 passed, 14 skipped, 7 xfailed, 1
+  xpassed
+
+## Human edits
+
+None — committed as generated by claude (no diff
+between `.claude/git_commit_msg_LATEST.md` and the
+committed body, as far as the assistant could
+observe).

ai/prompt-io/claude/20260417T035800Z_61a73ba_prompt_io.raw.md

@@ -0,0 +1,138 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+timestamp: 2026-04-17T03:58:05Z
+git_ref: 61a73ba
+diff_cmd: git diff 61a73ba~1..61a73ba
+---
+
+Code generated in this turn was committed verbatim as
+`61a73bae` ("Mv `trio_proc`/`mp_proc` to per-backend
+submods"). Per diff-ref mode, per-file code is captured
+via the pointers below, each followed by a prose
+summary of what the AI generated. Non-code output
+(sanity-check results, design rationale) is included
+verbatim.
+
+## Per-file generated content
+
+### `tractor/spawn/_trio.py` (new, 292 lines)
+
+> `git diff 61a73ba~1..61a73ba -- tractor/spawn/_trio.py`
+
+Pure lift-and-shift of `trio_proc()` out of
+`tractor/spawn/_spawn.py` (previously lines 448–670).
+Added AGPL header + module docstring describing the
+backend; imports include local `from ._spawn import
+cancel_on_completion, hard_kill, soft_kill` which
+creates the bottom-of-module late-import pattern in
+the core file to avoid a cycle. All call sites,
+log-format strings, and body logic are byte-identical
+to the originals — no semantic change.
+
+### `tractor/spawn/_mp.py` (new, 235 lines)
+
+> `git diff 61a73ba~1..61a73ba -- tractor/spawn/_mp.py`
+
+Pure lift-and-shift of `mp_proc()` out of
+`tractor/spawn/_spawn.py` (previously lines 672–842).
+Same AGPL header convention. Key difference from
+`_trio.py`: uses `from . import _spawn` (module
+import, not from-import) for `_ctx` and
+`_spawn_method` references — these are mutated at
+runtime by `try_set_start_method()`, so late binding
+via `_spawn._ctx` / `_spawn._spawn_method` is required
+for correctness. Also imports `cancel_on_completion`,
+`soft_kill`, `proc_waiter` from `._spawn`.
+
+### `tractor/spawn/_spawn.py` (modified, 847 → 448 LOC)
+
+> `git diff 61a73ba~1..61a73ba -- tractor/spawn/_spawn.py`
+
+- removed `trio_proc()` body (moved to `_trio.py`)
+- removed `mp_proc()` body (moved to `_mp.py`)
+- pruned imports now unused in core: `sys`,
+  `is_root_process`, `current_actor`,
+  `is_main_process`, `_mp_main`, `ActorFailure`,
+  `pretty_struct`, `_pformat`
+- added bottom-of-file late imports
+  `from ._trio import trio_proc` and
+  `from ._mp import mp_proc` with a one-line
+  comment explaining why (circular dep)
+- `_methods` dict unchanged structurally; still binds
+  `'trio' → trio_proc`, `'mp_spawn' → mp_proc`,
+  `'mp_forkserver' → mp_proc`
+
+### `tractor/spawn/__init__.py` (modified)
+
+> `git diff 61a73ba~1..61a73ba -- tractor/spawn/__init__.py`
+
+Docstring rewrite only — no code. New text describes
+the per-backend `Layout` with `- ._spawn` (core),
+`- ._trio` (default), `- ._mp` (two mp variants), and
+reiterates the "no eager import" NOTE.
+
+### `tractor/_testing/pytest.py` (modified)
+
+> `git diff 61a73ba~1..61a73ba -- tractor/_testing/pytest.py`
+
+Two-line logical change (plus a comment):
+- `from typing import` grows `get_args`
+- `pytest_generate_tests()` replaces the hardcoded
+  `('mp_spawn','mp_forkserver','trio')` tuple with
+  `get_args(SpawnMethodKey)` via lazy import
+  `from tractor.spawn._spawn import SpawnMethodKey`
+  inside the function body (matching the project's
+  existing lazy-import convention for the spawn
+  module)
+
+## Non-code output (verbatim)
+
+### Design rationale for the file split
+
+From my Phase A plan turn (already captured in
+`20260417T034918Z_9703210_prompt_io.md`):
+- `_spawn.py` stays as the "core" rather than renamed
+  to `_core.py` so external imports
+  (`_testing/pytest.py:228` imports
+  `try_set_start_method` from `tractor.spawn._spawn`)
+  keep working without churn.
+- Per-backend extraction chosen over alternatives
+  (e.g. splitting generic helpers further) because
+  the immediate motivation is hosting a 3rd
+  `_subint.py` sibling cleanly in Phase B.
+
+### Sanity-check output (verbatim terminal excerpts)
+
+Post-extraction import probe:
+```
+extraction OK
+_methods: {'trio': 'tractor.spawn._trio.trio_proc',
+           'mp_spawn': 'tractor.spawn._mp.mp_proc',
+           'mp_forkserver': 'tractor.spawn._mp.mp_proc'}
+```
+
+Spawn-relevant test subset (`tests/test_local.py
+test_rpc.py test_spawning.py test_multi_program.py
+test_discovery.py`):
+```
+37 passed, 1 skipped, 14 warnings in 55.37s
+```
+
+Full suite:
+```
+350 passed, 14 skipped, 7 xfailed, 1 xpassed,
+151 warnings in 437.73s (0:07:17)
+```
+
+No regressions vs. `main`. One transient `-x`
+early-stop `ERROR` on
+`test_close_channel_explicit_remote_registrar[trio-True]`
+was flaky (passed solo, passed without `-x`), not
+caused by this refactor.
+
+### Commit message
+
+Also AI-drafted (via `/commit-msg`) — the 40-line
+message on commit `61a73bae` itself. Not reproduced
+here; see `git log -1 61a73bae`.

ai/prompt-io/claude/20260601T231429Z_0e3e008b_prompt_io.md

@@ -0,0 +1,146 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+session: trio-0.33-subproc-supervisor-retroactive
+timestamp: 2026-06-01T23:14:29Z
+git_ref: 0e3e008b
+scope: code
+substantive: true
+raw_file: 20260601T231429Z_0e3e008b_prompt_io.raw.md
+---
+
+## Prompt
+
+**RETROACTIVE LOG** — original session prompts not
+preserved; reconstructed from the staged work product.
+
+The work designs a `trio.Nursery.start()`-style wrapper
+around `trio.run_process()` for SC-friendly subprocess
+supervision. From the resulting code shape, the
+prompting intent was:
+
+1. Surface rc!=0 `CalledProcessError` DETERMINISTICALLY,
+   without the nursery-eg-wrapping that complicates
+   `collapse_eg()` usage and races the relay reader on
+   trio's `check=True`-driven cancel cascade.
+2. ALWAYS isolate the parent controlling-tty so a
+   spawned child can't emit terminal control-seqs onto
+   the launching tty (clobbering scrollback). Default
+   `stdin=DEVNULL`; default `stdout=DEVNULL` unless
+   explicitly relayed/overridden; distinguish "caller
+   passed nothing" from "caller passed `None` for
+   inherit".
+3. Optional live per-line relay of child std-streams to
+   the `tractor` log — STREAMED (not
+   buffered-until-exit) so long-lived daemon output is
+   visible during the run. Pick a custom log level that
+   shows at usual `info`/`devx` console levels but is
+   separately filterable.
+4. Concurrent pipe-drain reader MANDATORY when piping
+   without `capture_*` — without it the child blocks on
+   `write()` once the OS pipe buffer fills (~64KiB),
+   causing deadlocks on output bursts.
+5. Non-blocking `tn.start()` semantics: hand the live
+   `trio.Process` to the parent immediately;
+   supervise/relay run to completion in the supervisor
+   coro.
+6. Hermetic `trio`-only unit tests (no actor-runtime)
+   covering each of: per-line relay, tty isolation,
+   no-deadlock on >64KiB unnewlined output, CPE
+   rebuild w/ stderr relay, CPE rebuild on the silent
+   drain+capture path.
+
+## Response summary
+
+Adds `tractor/trionics/_subproc.py` (296 LOC) +
+`tests/trionics/test_subproc.py` (230 LOC) + a
+re-export in `tractor/trionics/__init__.py`.
+
+**`supervise_run_process()`** (public, re-exported)
+- `check=False` is forced to `trio.run_process`; the
+  rc-check runs in the supervisor coro AFTER `own_tn`
+  unwinds (both the child AND the relay readers have
+  hit EOF + fully drained). A BARE
+  `subprocess.CalledProcessError` is rebuilt + raised
+  from there, with `.stderr` bytes passed in the
+  constructor AND attached as an `add_note()`'d
+  `|_.stderr:` block for legible teardown logs.
+- `stdin=DEVNULL` always. `stdout` default chosen via a
+  `_UNSET` sentinel: `relay_stdout=True` → PIPE,
+  explicit `stdout=...` → as given, else `DEVNULL`.
+  `stderr` defaults to PIPE whenever we relay OR need
+  the CPE note (when `check=True`), else `DEVNULL`.
+- `relay_level='io'` (custom level 21; sorts just
+  above stdlib `INFO`=20 so it shows at usual
+  `info`/`devx` levels and stays separately
+  filterable). `runtime`=15 would silently filter at
+  default levels, so it's rejected as a default.
+- `task_status.started(trio_proc)` delivers the live
+  process immediately. The internal `own_tn`
+  supervises `trio.run_process` + any relay readers to
+  completion.
+- `**run_process_kwargs` forward verbatim;
+  `stdin/stdout/stderr/check` are MANAGED keys
+  (override on conflict).
+- Crash-handling deliberately NOT baked in — compose
+  `maybe_open_crash_handler()` on top at the call-site.
+
+**`_relay_stream_lines()`** (internal helper)
+- Three modes (combinable): `emit`-only (live per-line
+  relay), `accum`-only (silent drain+capture for a CPE
+  note), or both (live relay AND capture).
+- Per-line split handles cross-chunk residuals via a
+  rolling `residual` bytes buffer; flushes any trailing
+  un-newline-term'd line at EOF.
+- `async with stream:` ensures aclose at EOF/cancel
+  (mirrors trio's internal `_subprocess` drain idiom).
+
+**`_add_stderr_note()`** (internal helper)
+- `add_note()`s a `textwrap.indent(...)`'d
+  `|_.stderr:` block onto a `CalledProcessError` for
+  teardown logs.
+
+**Tests** (5 hermetic, trio-only) — `_capture_relay`
+fixture monkeypatches `_subproc.log.<level>` to a list:
+- `test_stdout_relayed_per_line`: per-line stdout
+  relay carries each `line=N` to the records.
+- `test_parent_tty_isolated`: `readlink /proc/self/fd/0`
+  and `fd/1` from the child show `pipe:` (fd1) +
+  `/dev/null` (fd0); NO `/dev/pts/*`.
+- `test_no_deadlock_on_big_unnewlined_output`: 200KiB
+  of `x` with no newlines completes inside
+  `fail_after(2)` — exercises the concurrent drain.
+- `test_stderr_relay_and_cpe_rebuild`: rc=3 with
+  `relay_stderr=True` raises bare CPE
+  (via `collapse_eg()`) with `b'boom' in cpe.stderr`,
+  the note attached, AND per-line live relay.
+- `test_nonrelay_cpe_note`: rc=7 with no relay still
+  produces CPE with `.stderr` + note via the silent
+  drain+capture path.
+
+## Files changed
+
+- `tractor/trionics/_subproc.py` — NEW. Public
+  `supervise_run_process()` + helpers
+  `_relay_stream_lines()` / `_add_stderr_note()` + the
+  `_UNSET` sentinel.
+- `tests/trionics/test_subproc.py` — NEW. 5 hermetic
+  trio-only tests + `_capture_relay` monkeypatch
+  fixture.
+- `tractor/trionics/__init__.py` — re-export
+  `supervise_run_process`.
+
+## Human edits
+
+**RETROACTIVE**: this log is being written from the
+staged diff, not from a live session. The code as
+staged is the canonical artifact; any human edits the
+user made during the originating design session are
+already integrated and cannot be separated post-hoc.
+The `.raw.md` sibling is a diff-pointer placeholder,
+NOT a pre-edit transcript.
+
+Future prompt-io entries for in-flight work should be
+written DURING the design session per the skill
+contract so the pre-edit `.raw.md` captures the
+unedited model output for genuine provenance.

ai/prompt-io/claude/20260601T231429Z_0e3e008b_prompt_io.raw.md

@@ -0,0 +1,106 @@
+---
+model: claude-opus-4-7[1m]
+service: claude
+timestamp: 2026-06-01T23:14:29Z
+git_ref: 0e3e008b
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+# RETROACTIVE — original model output not preserved
+
+This `.raw.md` would normally contain the verbatim
+pre-human-edit response from the design session that
+produced the staged `_subproc.py` module + tests. That
+session's transcript is not available, so this file
+serves as a diff-pointer placeholder + transparency
+note.
+
+## Authoritative artifact
+
+The committed code IS the artifact of record. Once the
+companion commit lands, the unified diff is:
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_subproc.py`
+> `git diff HEAD~1..HEAD -- tests/trionics/test_subproc.py`
+> `git diff HEAD~1..HEAD -- tractor/trionics/__init__.py`
+
+Before committing, substitute `--cached` for the
+pre-commit form.
+
+## What is NOT here
+
+Because this is retroactive:
+- No verbatim chain-of-thought / discussion prose from
+  the design session.
+- No rejected alternatives the model considered before
+  arriving at the final shape (e.g. whether the
+  rc-check should live inside `own_tn` vs after it; the
+  `_UNSET` sentinel vs a `None`-means-DEVNULL
+  convention; `io` vs `info` as the default relay
+  level).
+- No pre-edit code blocks as the model first emitted
+  them, separable from any user cleanup applied before
+  the diff was staged.
+
+## Inferred design choices visible in the final code
+
+(Documented here because they're the kind of decision
+detail an unedited raw transcript would have captured.)
+
+1. **Post-drain rc-check in the supervisor coro body,
+   AFTER `own_tn.__aexit__`.** Placing the
+   `CalledProcessError` raise here (not inside
+   `own_tn`) means the EG-unwrap happens at the OUTER
+   `tn.start()` boundary — callers do `collapse_eg()`
+   if they want bare. Doing the raise INSIDE `own_tn`
+   would cancel the still-draining relay reader
+   mid-flight and lose stderr lines.
+
+2. **`_UNSET` sentinel for `stdout`.** A plain default
+   of `None` couldn't distinguish "use the safe
+   `DEVNULL` default" from "caller explicitly passed
+   `None` (inherit, presumably knowingly)". The
+   sentinel keeps the SAFE default while letting power
+   users opt into inherit.
+
+3. **`relay_level='io'` (custom level 21).** Chosen to
+   sort just above stdlib `INFO`=20 so a default
+   `--ll info` shows the relay, but it remains a
+   distinct level so users can filter
+   `tractor.trionics:io` separately. Picking
+   `runtime`=15 would have made the relay invisible at
+   default verbosity (a footgun for daemon supervisors
+   whose whole point is "I want to see this output").
+
+4. **Reader is MANDATORY, not opt-in cosmetic.** With
+   `stdout=PIPE` / `stderr=PIPE` we OWN the drain
+   responsibility — there's no `trio.capture_*` running
+   under the hood here. The ~64KiB OS pipe buffer
+   means a child writing more than that without us
+   reading hangs at `write()` — a deadlock that won't
+   show up in small-output tests, which is why the
+   200KiB-no-newline test is in the suite.
+
+5. **`task_status.started(trio_proc)` BEFORE the
+   `own_tn` exits.** Without this, `tn.start()` would
+   block until the child exits — losing the "start a
+   long-lived daemon and continue with parent work"
+   use case. With it, the parent gets the live process
+   handle immediately and the supervise+relay tasks
+   run in the supervisor coro until the child exits.
+
+6. **`__notes__` via `add_note()` for the CPE
+   `.stderr`.** The `.stderr` attribute is what
+   `subprocess` callers expect; the `add_note()` is
+   what trio's exception-rendering shows. Both wired so
+   programmatic AND human consumers see the stderr at
+   teardown.
+
+## Honesty statement
+
+This file's content is RECONSTRUCTED from the staged
+code, not extracted from a verbatim model transcript.
+The prompt-io skill's intent is for the `.raw.md` to
+be a pre-edit fossil; that's not possible here. Future
+work should write the prompt-io entry DURING the
+design session.

ai/prompt-io/claude/20260611T175152Z_8526985c_prompt_io.md

@@ -0,0 +1,93 @@
+---
+model: claude-fable-5[1m]
+service: claude
+session: 0780a862-e19a-4f0a-86cd-c8afc0997757
+timestamp: 2026-06-11T17:51:52Z
+git_ref: 8526985c
+scope: docs+code+config
+substantive: true
+raw_file: 20260611T175152Z_8526985c_prompt_io.raw.md
+---
+
+## Prompt
+
+> we need big boi docs; ours are way out of date and generally
+> terrible. i'd like to use a sphinx theme in the vein and/or
+> one-of/nearly-the-same-as used in one or all of the `msgspec`,
+> `numpy`, `ray`, `xonsh`, `polars` projects.
+>
+> see the following oustanding but now very old issues to guide
+> you: #175, #126, #123, #157
+>
+> we should try to target a 3 columned sphinx theme with,
+> - document index-nav on left,
+> - body content in middle,
+> - diagrams (ideally in d2lang if possible) on RHS
+>
+> optimize for (if possible),
+> - reusing examples code without duplication in docs,
+> - generating new examples from todos throughout code base,
+> - distilling the essence of SC across processes in the simplest
+>   and most friendly way possible.
+>
+> do this work in a /open-wkt and do you best without my guidance
+> for the first major pass - we will refine asap once you are
+> complete.
+
+Session settings: `/effort max` + ultracode (multi-agent workflow
+orchestration). Fully autonomous first pass; two orchestrated
+agent fleets were used (6-agent recon survey, then 10-agent
+content fan-out: 9 section writers + 1 examples smith), with the
+orchestrator authoring the landing page, SC essay, sphinx
+scaffold, d2 pipeline + diagram sources, CI workflow and all
+integration/fix passes directly.
+
+## Response summary
+
+Complete sphinx docs revamp on branch `wkt/big_boi_docs`:
+`pydata-sphinx-theme` 0.18 (3-column: left nav / content /
+page-toc, with d2 diagrams + asides floated into the RHS margin
+via custom CSS), sphinx 9.1, a local `.. d2::` directive
+rendering `docs/diagrams/*.d2` sources with committed-SVG
+fallback, a vendored `.. margin::` directive, ~25 new doc pages
+(landing, start/, explain/, 12 guides, 10 api-ref pages,
+project/), 5 new auto-tested examples + 3 modernized + 1
+renamed, and a gh-pages deploy workflow (issue #123). All
+example code is `literalinclude`d from `examples/` (zero
+duplication, CI-verified). Build: green, 24 warnings all
+pre-existing-docstring/NEWS sourced.
+
+## Files changed
+
+See the branch diff (uncommitted at entry-write time):
+
+> `git diff test_cpu_throttling..wkt/big_boi_docs`
+> `git -C <wkt> status --short` (pre-commit working tree)
+
+- `docs/conf.py` — full rewrite for pydata theme + ext stack
+- `docs/_ext/d2diagrams.py` — new `.. d2::` sphinx directive
+- `docs/_ext/marginalia.py` — new `.. margin::` directive
+- `docs/_static/css/custom.css` — b&w skin + RHS margin floats
+- `docs/diagrams/*.d2` (7) — diagram sources (sketch/grayscale)
+- `docs/_diagrams/*.svg` (7) — committed rendered fallbacks
+- `docs/index.rst` — new landing (replaces dead-API doc)
+- `docs/start/*.rst` (3), `docs/explain/*.rst` (3),
+  `docs/guide/*.rst` (13), `docs/api/*.rst` (10),
+  `docs/project/*.rst` (3) — new content tree
+- `docs/dev_tips.rst` — removed (ported to project/dev-tips)
+- `examples/{typed_payloads,nested_actor_tree,
+  service_daemon_discovery,uds_transport_actor_tree,
+  streaming_broadcast_fanout}.py` — new, smoke-tested
+- `examples/{a_trynamic_first_scene,
+  actor_spawning_and_causality,parallelism/single_func}.py` —
+  `.result()` -> `.wait_for_result()` modernization
+- `examples/parallelism/concurrent_futures_primes.py` — renamed
+  from leading-underscore + trio-runner shim added
+- `pyproject.toml` — `docs` dependency-group filled in
+- `uv.lock` — relock for docs group
+- `.github/workflows/docs.yml` — build + gh-pages deploy
+
+## Human edits
+
+None yet — entry written pre-commit; the author reviews, stages
+and commits manually (per repo workflow policy).

ai/prompt-io/claude/20260611T175152Z_8526985c_prompt_io.raw.md

@@ -0,0 +1,49 @@
+---
+model: claude-fable-5[1m]
+service: claude
+timestamp: 2026-06-11T17:51:52Z
+git_ref: 8526985c
+diff_cmd: git diff test_cpu_throttling..wkt/big_boi_docs
+---
+
+# Raw output pointers (diff-ref mode)
+
+All generated content is code/config/docs committed alongside
+this entry on branch `wkt/big_boi_docs`; per the prompt-io
+diff-ref decision rule each file's verbatim content is the diff
+itself rather than a copy here:
+
+> `git diff test_cpu_throttling..wkt/big_boi_docs -- docs/`
+> `git diff test_cpu_throttling..wkt/big_boi_docs -- examples/`
+> `git diff test_cpu_throttling..wkt/big_boi_docs -- pyproject.toml uv.lock .github/`
+
+## Generation notes (non-code output summary)
+
+- Theme research (web, agent-verified 2026-06-11): msgspec=furo,
+  xonsh=furo, numpy/ray/polars=pydata-sphinx-theme (ray migrated
+  off sphinx-book-theme); sphinx-book-theme 1.2.0 hard-pins
+  pydata 0.16.1 (stale) -> chose pydata 0.18 + sphinx 9.1.
+- d2 ecosystem: no production-grade pypi extension exists
+  (sphinxcontrib-d2lang 0.0.5 ignores returncodes, uuid4 output
+  names; sphinx-d2 is an empty stub) -> wrote local
+  `docs/_ext/d2diagrams.py` (~230 LOC) with D2_BIN env
+  discovery, mtime caching, committed-SVG fallback and
+  literal-block last resort.
+- Diagrams authored in d2 (theme-id 1 "Neutral Grey" + sketch
+  mode + ELK layout, validated by render + headless-firefox
+  screenshot loop): actor_tree, context_handshake (real
+  msg-spec names Start/StartAck/Started/Yield/Stop/Return),
+  streaming_pipeline, runtime_stack, debug_lock,
+  error_propagation, infected_aio.
+- API truth enforced from a 6-agent recon pass over the
+  reorganized package tree (runtime/, discovery/, spawn/, ipc/,
+  msg/, devx/, trionics/): docs teach `.wait_for_result()`,
+  registrar (not arbiter) naming, `@tractor.context` +
+  `open_context()` as the core model, `run_in_actor()` as
+  convenience only.
+- All ~30 literalincluded example scripts verified present; 9
+  touched/new example files smoke-run green (exit 0, <16s).
+- Final build: `sphinx-build -b html` succeeded; 24 residual
+  warnings, every one sourced from pre-existing library
+  docstring rst-isms or legacy NEWS.rst content (left untouched
+  by design; flagged for a follow-up docstring lint pass).

ai/prompt-io/claude/20260625T232506Z_0a3c48fc_prompt_io.md

@@ -0,0 +1,65 @@
+---
+model: claude-fable-5[1m]
+service: claude
+session: 638d5a98-abd2-4217-a447-19f2330eb334
+timestamp: 2026-06-25T23:25:06Z
+git_ref: 0a3c48fc
+scope: code
+substantive: true
+raw_file: 20260625T232506Z_0a3c48fc_prompt_io.raw.md
+---
+
+## Prompt
+
+Same `/code-review` follow-up session (PR #460). Finding #10:
+the new `api/` autodoc reference pages surfaced 22 docutils
+warnings sourced from informal reST in public docstrings. The
+user opted to fix it on this branch,
+
+> well we're on a clean main so why not just do these now too?
+
+Goal: drive the docs build warning-free (24 -> 0) without any
+code/behaviour change — docstring content only.
+
+## Response summary
+
+Surgical docstring-RST lint across 9 library modules + the one
+offending `NEWS.rst` changelog entry,
+
+- bullet lists given a blank line + base-column indent
+  (`Context`, `Context.cancel_called`/`.cancelled_caught`/
+  `.outcome`, `ActorNursery.cancel_called`, `query_actor`,
+  `open_crash_handler`, the `#318` `NEWS.rst` entry),
+- the under-short `Behaviour:` underline in `Context.cancel`
+  demoted to a `**bold**` label,
+- unbalanced/pluralized-role backticks fixed — closed the
+  `wait_for_actor` summary backtick; applied the `` `role`\ s ``
+  escaped-plural idiom in `gather_contexts`, `mk_pdb`,
+  `MsgCodec`, msg `Error`, `open_context_from_portal`,
+- the `|_` method-tree in `ContextCancelled.canceller` made a
+  literal block (bare `|` was parsed as a substitution ref).
+
+Verified: build `24 -> 0` warnings; `import tractor` clean under
+`-W error::SyntaxWarning`; the `\ s` idiom renders as e.g.
+"acms" (no backslash/space leak in HTML); `ruff` clean; diff
+confirmed docstring-content-only.
+
+Initial generation was delegated to a constrained subagent
+(docstring-only edits + a rebuild-to-zero verification gate);
+the orchestrator independently re-built, reviewed the full
+diff, and added the `NEWS.rst` fix to reach zero.
+
+## Files changed
+
+- `tractor/_context.py`, `tractor/_exceptions.py`,
+  `tractor/devx/debug/_post_mortem.py`,
+  `tractor/devx/debug/_repl.py`, `tractor/discovery/_api.py`,
+  `tractor/msg/_codec.py`, `tractor/msg/types.py`,
+  `tractor/runtime/_supervise.py`,
+  `tractor/trionics/_mngrs.py` — docstring reST fixes,
+- `NEWS.rst` — blank line before a bullet list in the `#318`
+  entry.
+
+## Human edits
+
+None — committed as generated (`0a3c48fc`).

ai/prompt-io/claude/20260625T232506Z_0a3c48fc_prompt_io.raw.md

@@ -0,0 +1,35 @@
+---
+model: claude-fable-5[1m]
+service: claude
+timestamp: 2026-06-25T23:25:06Z
+git_ref: 0a3c48fc
+diff_cmd: git diff 0a3c48fc~1..0a3c48fc
+---
+
+# Raw output pointers (diff-ref mode)
+
+The generated patch is committed; the verbatim content is the
+diff:
+
+> `git diff 0a3c48fc~1..0a3c48fc`
+
+## Generation notes (non-code, verbatim)
+
+- Pure docstring-content edits; no signatures, logic, or
+  non-docstring lines touched (confirmed via `git diff` review).
+- reST fix patterns applied: blank-line + base-column indent
+  for bullet lists ("Unexpected indentation" / "Definition
+  list ..." / "Block quote ..."); `**bold**` in place of an
+  under-length section underline ("Title underline too short");
+  closing/escaping backticks ("Inline interpreted text ...
+  without end-string"); literal-block for an ASCII tree whose
+  `|` chars tripped "Inline substitution_reference ...".
+- The pluralized-role idiom is written `\\ s` in the Python
+  source (so the runtime docstring holds `` `role`\ s ``); a
+  bare `\ ` would raise a 3.13 invalid-escape SyntaxWarning.
+  Verified clean via `python -W error::SyntaxWarning -c
+  "import tractor"`.
+- Verification gate: `sphinx-build` warning count 24 -> 2
+  (subagent, docstrings only) -> 0 after the orchestrator added
+  the `NEWS.rst` blank-line fix; rendered-HTML spot-check
+  confirmed no literal `\ s` leak; `ruff check` clean.

ai/prompt-io/claude/20260625T232506Z_3a5cfde0_prompt_io.md

@@ -0,0 +1,64 @@
+---
+model: claude-fable-5[1m]
+service: claude
+session: 638d5a98-abd2-4217-a447-19f2330eb334
+timestamp: 2026-06-25T23:25:06Z
+git_ref: 3a5cfde0
+scope: code
+substantive: true
+raw_file: 20260625T232506Z_3a5cfde0_prompt_io.raw.md
+---
+
+## Prompt
+
+Follow-up to a `/code-review` of PR #460 (this branch) which
+surfaced lower-severity findings in the `.. d2::` sphinx
+directive. The user's driving instructions across the session,
+
+> yup do both *at least*, then report back.
+
+(re: "implement #7 (+ optionally #8) on the branch") and,
+
+> well we're on a clean main so why not just do these now too?
+
+(re: also doing #9). The three findings being addressed,
+
+- #7: a `d2` render that is *attempted and fails* (binary
+  present but errors on the source) silently degraded to the
+  stale committed SVG and the build still exited 0,
+- #8: `parallel_*_safe = True` was declared while `run()`
+  writes the shared output SVG during the read phase (torn
+  write under `sphinx-build -j`),
+- #9: output keyed only by `src.stem` so two `.d2` sources
+  with the same stem collide.
+
+## Response summary
+
+Reworked `docs/_ext/d2diagrams.py`,
+
+- `render_svg()` now returns a `RenderResult` tristate
+  (`OK` / `FELL_BACK` / `NO_OUTPUT` / `FAILED`) so `run()` can
+  distinguish a graceful no-binary fallback from a real render
+  failure; `FAILED` emits a `reporter.error` (build fails under
+  `-W`),
+- the render writes into a sibling temp-file then `os.replace()`
+  (atomic) so a failed/torn render can never clobber a good
+  committed SVG,
+- a per-build `_seen_outputs` map (reset on `builder-inited`)
+  errors on a same-stem output collision.
+
+Verified: a corrupted `.d2` errors under `-W` (exit 1) with the
+committed SVG byte-unchanged; the collision guard fires on a
+duplicate stem; `nix run nixpkgs#d2` render path + atomic swap
+leave no temp residue; `ruff` clean.
+
+## Response summary (cont.) — Files changed
+
+- `docs/_ext/d2diagrams.py` — tristate render result, atomic
+  temp-file render, output-collision guard, doc-string policy
+  update.
+
+## Human edits
+
+None — committed as generated (`3a5cfde0`). The user noted
+follow-up refinements may come in later commits.

ai/prompt-io/claude/20260625T232506Z_3a5cfde0_prompt_io.raw.md

@@ -0,0 +1,35 @@
+---
+model: claude-fable-5[1m]
+service: claude
+timestamp: 2026-06-25T23:25:06Z
+git_ref: 3a5cfde0
+diff_cmd: git diff 3a5cfde0~1..3a5cfde0
+---
+
+# Raw output pointers (diff-ref mode)
+
+The generated patch is committed; per the diff-ref decision
+rule the verbatim content is the diff itself:
+
+> `git diff 3a5cfde0~1..3a5cfde0 -- docs/_ext/d2diagrams.py`
+
+## Generation notes (non-code, verbatim)
+
+- `RenderResult` enum added; `render_svg()` return type changed
+  `bool -> RenderResult`. Mapping in `run()`: `FAILED` ->
+  `state_machine.reporter.error(...)`, `NO_OUTPUT` -> raw `.d2`
+  source as a `literal_block`, `OK`/`FELL_BACK` -> `image`.
+- Atomicity via `tempfile.mkstemp(dir=out.parent, ...)` +
+  `os.replace(tmp, out)`; temp unlinked on any failure path.
+- Collision guard: module-level `_seen_outputs: dict[str,str]`
+  keyed by output basename, cleared by a `builder-inited`
+  handler connected in `setup()`.
+- `parallel_read_safe`/`parallel_write_safe` kept `True` but
+  now justified by the atomic swap (documented inline).
+- Empirical verification performed before commit:
+  - broken `.d2` + `sphinx-build -W` -> exit 1, ERROR node
+    rendered, committed `actor_tree.svg` md5 unchanged,
+  - duplicate-stem orphan page -> "d2 output collision" error,
+  - `nix run nixpkgs#d2` forced re-render -> deterministic
+    (no git drift), no `.tmp` residue,
+  - `ruff check` clean.

ai/prompt-io/claude/20260702T154255Z_65bf9df5_prompt_io.md

@@ -0,0 +1,79 @@
+---
+model: claude-fable-5
+service: claude
+session: f6c84722-471a-4458-9a80-e453fea9029f
+timestamp: 2026-07-02T15:42:55Z
+git_ref: 65bf9df5
+scope: code
+substantive: true
+raw_file: 20260702T154255Z_65bf9df5_prompt_io.raw.md
+---
+
+## Prompt
+
+Driver prompt file `ai/prompt-io/prompts/issue_477.md`:
+
+> attempt to resolve
+> https://github.com/goodboy/tractor/issues/477
+> do it with /open-wkt.
+
+(plus a hard stop-for-human-review deadline of 12:50PM
+EST the same day)
+
+Issue #477 asks to factor `ActorNursery.run_in_actor()`
+(and possibly `Portal.run()`) out of the nursery
+internals into a new `tractor.to_actor` wrapper
+subpackage of "higher level one shot" single-remote-task
+APIs, adopting the `trio.to_thread`/`anyio.to_process`
+parlance, so that error collection/propagation moves up
+into the caller's local `trio` scope and the nursery's
+spawn machinery can eventually drop the
+`._ria_nursery` coupling.
+
+## Response summary
+
+First-cut `tractor.to_actor` subpkg delivering the
+one-shot API composed purely from the existing
+daemon-spawn + portal primitives (`start_actor()` +
+`Portal.run()` + `Portal.cancel_actor()`), leaving the
+legacy `.run_in_actor()` machinery untouched (formal
+deprecation deferred until in-repo usage migrates):
+
+- `to_actor.run(fn, **fn_kwargs) -> Any`: spawn a
+  subactor, schedule `fn` as its lone remote task, wait
+  on and return its result, ALWAYS reaping the subactor
+  (shield-safe `finally`). Remote errors raise in the
+  caller's task as boxed `RemoteActorError`s.
+- placement variants: `portal=` reuses a running actor
+  (no spawn/reap), `an=` spawns from a caller-managed
+  actor-nursery, neither opens a call-scoped private
+  `open_nursery()` (implicitly booting the runtime,
+  configurable via `runtime_kwargs`).
+- fail-fast validation before any spawn: non-streaming
+  async fn required; `portal=`/`an=` mutually
+  exclusive; `runtime_kwargs` rejected alongside any
+  placement opt.
+- `run_in_actor()` TODO/docstring now cross-reference
+  the successor API.
+
+## Files changed
+
+- `tractor/to_actor/__init__.py` — new subpkg,
+  re-exports `run`
+- `tractor/to_actor/_api.py` — `run()` +
+  `_invoke_in_subactor()` + `_validate_one_shot_fn()`
+- `tractor/__init__.py` — top-level `to_actor`
+  re-export
+- `tractor/runtime/_supervise.py` — comment/docstring
+  pointers from `run_in_actor()` to the successor
+- `tests/test_to_actor.py` — 11-test suite covering
+  all placement variants, error relay, the concurrent
+  worker-pool-ish pattern and arg validation
+- `examples/parallelism/to_actor_one_shots.py` —
+  runnable demo (auto-collected by
+  `test_docs_examples.py`)
+
+## Human edits
+
+None yet — pending human review (work paused before the
+12:50PM EST deadline per the driver prompt).

ai/prompt-io/claude/20260702T154255Z_65bf9df5_prompt_io.raw.md

@@ -0,0 +1,100 @@
+---
+model: claude-fable-5
+service: claude
+timestamp: 2026-07-02T15:42:55Z
+git_ref: 65bf9df5
+diff_cmd: git diff main..wkt/to_actor_subpkg
+---
+
+# Raw AI output (diff-ref mode)
+
+All generated code is committed on the
+`wkt/to_actor_subpkg` branch; per diff-ref mode each
+file's verbatim content is reachable via the pointers
+below rather than duplicated here.
+
+## Generated files
+
+> `git diff main..wkt/to_actor_subpkg -- tractor/to_actor/__init__.py`
+
+New subpackage init: module docstring establishing the
+`trio.to_thread`/`anyio.to_process` "run it over there"
+parlance for actors, plus the single public re-export
+`run as run` from `._api`.
+
+> `git diff main..wkt/to_actor_subpkg -- tractor/to_actor/_api.py`
+
+The one-shot invocation impl, composed entirely from the
+lower level daemon-spawn + portal primitives as
+prescribed by issue #477:
+
+- `_validate_one_shot_fn()`: the `Portal.run()`
+  non-streaming-async-fn constraint checked up-front,
+  before any subactor is spawned.
+- `_invoke_in_subactor()`: `an.start_actor()` ->
+  `Portal.run()` -> always-reap via
+  `Portal.cancel_actor()` in a `finally` (the cancel
+  req's bounded wait is internally shielded so the reap
+  also runs under caller-scope cancellation).
+- `run()`: the public API. Placement options:
+  `portal=` (reuse a running actor, no spawn/reap),
+  `an=` (spawn from a caller-managed nursery), or
+  neither (private `open_nursery()` scoped to the call,
+  implicitly booting the runtime when needed, tunable
+  via pass-through `runtime_kwargs`). Spawn opts mirror
+  `ActorNursery.start_actor()`; `**fn_kwargs` are
+  relayed to the remote task. Errors raise in the
+  caller's task as boxed `RemoteActorError`s.
+  `runtime_kwargs` alongside any placement opt is a
+  hard `ValueError`, never silently ignored.
+
+> `git diff main..wkt/to_actor_subpkg -- tractor/__init__.py`
+
+Top-level `from . import to_actor as to_actor`
+re-export.
+
+> `git diff main..wkt/to_actor_subpkg -- tractor/runtime/_supervise.py`
+
+Comment/docstring-only: the `run_in_actor()` deprecation
+TODO now points at the implemented `.to_actor.run()`
+successor (checkbox ticked) and the method docstring
+gains a NOTE steering users to the new API; remaining
+TODO items are the `DeprecationWarning` emission +
+in-repo usage migration.
+
+> `git diff main..wkt/to_actor_subpkg -- tests/test_to_actor.py`
+
+11-test suite: private-nursery one-shot, implicit
+runtime boot via `runtime_kwargs`, remote-error relay to
+the caller's task (bare + caller-managed nursery),
+caller-nursery spawn, portal reuse w/o implicit reap,
+the concurrent worker-pool-ish pattern (local `trio`
+nursery x shared `an`), and the four validation
+rejections (sync fn, async-gen fn, `portal`+`an`
+combo, `runtime_kwargs`+placement combo).
+
+> `git diff main..wkt/to_actor_subpkg -- examples/parallelism/to_actor_one_shots.py`
+
+Runnable example (auto-collected by
+`test_docs_examples.py`): the fully-implicit one-shot
+plus the concurrent worker-pool-ish prime-check pattern
+against a shared caller-managed actor-nursery.
+
+## Test runs (verbatim)
+
+```
+tests/test_to_actor.py ..........                [100%]
+============= 10 passed in 4.29s =============
+```
+
+Regression subset for touched modules
+(`test_local.py test_rpc.py test_spawning.py
+test_cancellation.py`):
+
+```
+38 passed, 1 xfailed, 24 warnings in 80.71s (0:01:20)
+```
+
+(warnings are pre-existing stdlib `os.fork()`
+DeprecationWarnings from the mp spawn backends, not
+introduced by this change)

ai/prompt-io/claude/20260702T155006Z_65bf9df5_prompt_io.md

@@ -0,0 +1,77 @@
+---
+model: claude-fable-5
+service: claude
+session: b8ec2ffe-1e47-4db5-8bc3-5bebc5feaaf6
+timestamp: 2026-07-02T15:50:06Z
+git_ref: 65bf9df5
+scope: code
+substantive: true
+raw_file: 20260702T155006Z_65bf9df5_prompt_io.raw.md
+---
+
+## Prompt
+
+From `ai/prompt-io/prompts/issue_473.md`:
+
+> NOTE: you MUST pause this work at 12:50PM EST (BEFORE
+> your weekly limit reset) for review by a human!
+>
+> resolve GH #473: https://github.com/goodboy/tractor/issues/474
+> do it with /open-wkt.
+
+(The prompt's issue-number/URL mismatch was resolved in
+favor of #473 — file name, session name and prompt text
+all say #473; the `/474` URL is a typo.)
+
+GH #473: "Root-cause UDS-on-macOS —
+`uds_transport_actor_tree` example fails on CI".
+
+## Response summary
+
+Worked all four #473 next-step checkboxes in wkt
+`uds_macos_473`:
+
+1. **stderr surfacing**: `test_docs_examples.py` now
+   raises with the full subproc stderr (+stdout) on ANY
+   non-zero exit. The old last-line-'Error' check could
+   never fire for `tractor` crashes since root-actor
+   stderr always ends with the strict-EG collapse note
+   `( ^^^ this exc was collapsed from a group ^^^ )` —
+   proven against the real PR #460 macOS CI log (bare
+   `assert 1 == 0`, no traceback).
+2. **root-cause (linux-provable layer)**: macOS-only
+   addr corruption in
+   `MsgpackUDSStream.get_stream_addrs()` — no
+   `SO_PASSCRED`/autobind on darwin means the accept
+   side's `getpeername()` is `''`, and the
+   `(str(), str())` arm took `peername` unconditionally →
+   `Path('')` garbage addrs on every accepted conn.
+   Proven + fixed via linux no-autobind simulation.
+   Possibly not the final macOS crasher (non-fatal on
+   linux-sim); the diagnostic patch guarantees the next
+   macOS CI run shows any remaining layer.
+3. **CI matrix**: removed the `macos-latest`+`uds`
+   exclude.
+4. **un-skip**: dropped the macOS+CI skip of the example.
+
+Also: `start_listener()` bindspace mkdir hardened
+(`parents=True, exist_ok=True`), example docstring
+peer-pid mechanism corrected for macOS.
+
+## Files changed
+
+- `tests/test_docs_examples.py` — surface full stderr on
+  non-zero exit; remove macOS skip of the UDS example
+- `tractor/ipc/_uds.py` — fix no-autobind
+  `get_stream_addrs()` arm; harden bindspace mkdir;
+  document autobind semantics
+- `.github/workflows/ci.yml` — add macOS UDS leg (remove
+  matrix exclude)
+- `examples/uds_transport_actor_tree.py` — platform-
+  correct peer-pid docs
+
+## Human edits
+
+None yet — pending human review (hard-stop review
+checkpoint at 12:50PM EST per prompt); commit staging and
+any edits are the human's.

ai/prompt-io/claude/20260702T155006Z_65bf9df5_prompt_io.raw.md

@@ -0,0 +1,101 @@
+---
+model: claude-fable-5
+service: claude
+timestamp: 2026-07-02T15:50:06Z
+git_ref: 65bf9df5
+diff_cmd: git diff main..wkt/uds_macos_473
+---
+
+# Raw output — GH #473 UDS-on-macOS root-cause session
+
+NOTE: code output is diff-referenced (not copied) per
+prompt-io diff-ref mode; all generated code is on branch
+`wkt/uds_macos_473` relative to `main` (65bf9df5).
+
+## Diagnostic narrative (verbatim)
+
+Root-cause work proceeded by linux-side simulation since no
+macOS host is available:
+
+1. Pulled the actual macOS CI failure log from PR #460 run
+   28343878933 (job 84115107992): the failure is a bare
+   `assert 1 == 0` on `proc.returncode` with NO stderr
+   surfaced — the harness only re-raised when the LAST
+   stderr line contained 'Error'.
+
+2. Reproduced the stderr-swallow mechanism on linux: any
+   `tractor` root-actor crash ends stderr with the
+   strict-EG collapse note
+   `( ^^^ this exc was collapsed from a group ^^^ )` which
+   never matches 'Error' — so EVERY possible crash was
+   swallowed. (Verified by sabotaging the runtime dir via
+   an over-long `XDG_RUNTIME_DIR` → `OSError: AF_UNIX path
+   too long` → rc=1 + swallowed.)
+
+3. Found + proved a macOS-only addr-corruption bug in
+   `MsgpackUDSStream.get_stream_addrs()`: the
+   `(str(), str())` match-arm unconditionally took
+   `peername`, but on no-autobind platforms (macOS lacks
+   linux's `SO_PASSCRED`-triggered autobind) the accept
+   side's `getpeername()` is `''` → `Path('')` garbage
+   laddr/raddr on EVERY accepted UDS conn. Simulated on
+   linux by nulling `SO_PASSCRED` (no autobind → same `''`
+   shape): pre-fix the example printed
+   `listener sock file: .`; post-fix it prints the real
+   registry sockpath. Non-fatal on linux-sim (rc=0), so
+   possibly not the final macOS crasher — the diagnostic
+   patch guarantees the next macOS CI run reveals any
+   remaining layer.
+
+4. Falsified the missing-parent-dir theory:
+   `get_rt_dir()` already `mkdir(parents=True,
+   exist_ok=True)`s at import (and macOS TCP CI passes),
+   so `~/Library/Caches/TemporaryItems` absence cannot be
+   the crasher. Hardened `start_listener()`'s bindspace
+   mkdir anyway (custom `filedir` case + racing actors).
+
+## Generated changes (diff pointers)
+
+> `git diff main..wkt/uds_macos_473 -- tests/test_docs_examples.py`
+
+- always raise with FULL subproc stderr (+stdout) on any
+  non-zero example exit; keep legacy last-line 'Error'
+  check for zero-rc cases; drop the macOS+CI skip of
+  `uds_transport_actor_tree.py` (GH #473 next-step).
+
+> `git diff main..wkt/uds_macos_473 -- tractor/ipc/_uds.py`
+
+- `get_stream_addrs()`: document the autobind semantics
+  (bytes = linux abstract-ns autobind artifact), add
+  no-autobind `(str, str)` arm picking the non-empty name
+  (`peername` connect-side, `sockname` accept-side) with
+  an empty-pair `ValueError` guard.
+- `start_listener()`: `bs.mkdir(parents=True,
+  exist_ok=True)`.
+
+> `git diff main..wkt/uds_macos_473 -- .github/workflows/ci.yml`
+
+- remove the `macos-latest`+`uds` matrix exclude so
+  UDS-on-macOS is exercised by CI (GH #473 next-step).
+
+> `git diff main..wkt/uds_macos_473 -- examples/uds_transport_actor_tree.py`
+
+- docs nit: peer-pid mechanism is `SO_PEERCRED` on linux,
+  `LOCAL_PEERPID` on macOS.
+
+## Verification (verbatim summary)
+
+- macOS-shape sim (no autobind) + fix: example rc=0 with
+  correct listener sockpath.
+- native linux post-fix: example rc=0, autobind arms
+  unchanged.
+- sabotage smoke-test: diagnostic patch surfaces the full
+  EG traceback incl. collapse-note line.
+- `pytest tests/ipc/ tests/test_2way.py tests/discovery/
+  --tpt-proto uds`: 61 passed, 1 `TooSlowError` flake
+  (test_simple_rpc) that passes solo in 2.2s — attributed
+  to a concurrent full-suite run (other session) loading
+  the host.
+- full `tests/test_docs_examples.py` run pending at
+  raw-file write time (queued behind the concurrent
+  session's suite).

ai/prompt-io/claude/20260702T155626Z_65bf9df5_prompt_io.md

@@ -0,0 +1,83 @@
+---
+model: claude-fable-5
+service: claude
+session: b6b42e23-0454-4d9b-be1c-0a89d65aed1b
+timestamp: 2026-07-02T15:56:26Z
+git_ref: 65bf9df5
+scope: code
+substantive: true
+raw_file: 20260702T155626Z_65bf9df5_prompt_io.raw.md
+---
+
+## Prompt
+
+From `ai/prompt-io/prompts/issue_470.md`:
+
+> attempt to resolve
+> https://github.com/goodboy/tractor/issues/470
+> do it with /open-wkt, ensure you /run-tests on all
+> changes
+
+(plus a hard pause-for-human-review deadline of
+12:50PM EST.)
+
+Issue #470 asks to trim the ~0.42s `import tractor`
+cost — which dominates per-actor spawn latency on the
+`trio` backend — by lazy-importing heavy/optional deps
+(`pdbp`, `stackscope`, `multiaddr`, + audit of
+`colorlog`/`bidict`/`wrapt`/`setproctitle`).
+
+## Response summary
+
+Profiling showed the issue's dep-list only accounted
+for ~20ms; the dominant cost (~244ms) was
+`log.get_logger()`'s `get_caller_mod()` calling
+`inspect.stack()` at module level in ~39 modules —
+each call walks every stack frame (deep during nested
+imports) and scans `sys.modules` per frame via
+`inspect.getmodule()`.
+
+Changes, in impact order:
+
+1. `get_caller_mod()` -> `sys._getframe()` +
+   `f_globals['__name__']` `sys.modules` lookup
+   (~240ms saved).
+2. Issue's lazy-import checklist: `bidict`,
+   `multiaddr`, `colorlog`, `wrapt` moved to
+   `TYPE_CHECKING`/function-local imports;
+   `platformdirs` function-local; `asyncio` +
+   `.to_asyncio` deferred out of the `devx.debug` +
+   `spawn._entry` eager paths (~15ms saved).
+3. PEP 562 `__getattr__` on `tractor/__init__.py`
+   preserving public `tractor.to_asyncio` attr access.
+
+Results: `import tractor` 0.42s -> ~0.145s (~65%);
+sequential `start_actor` latency 0.40-0.44s ->
+~0.179s/actor. `pdbp` (needs `_repl.py` class-base
+restructure) + `platformdirs` (needs
+`UDSAddress.def_bindspace` protocol rework) documented
+as follow-ups.
+
+## Files changed
+
+- `tractor/log.py` — `get_caller_mod()` perf fix +
+  lazy `colorlog`
+- `tractor/__init__.py` — PEP 562 lazy `to_asyncio`
+- `tractor/discovery/_addr.py` — `bidict` ->
+  `TYPE_CHECKING`
+- `tractor/discovery/_multiaddr.py` — lazy `multiaddr`
+- `tractor/ipc/_tcp.py`, `tractor/ipc/_uds.py` —
+  `Multiaddr` -> `TYPE_CHECKING`
+- `tractor/runtime/_state.py` — lazy `platformdirs`
+- `tractor/devx/_frame_stack.py` — lazy `pdbp` +
+  `wrapt`
+- `tractor/devx/debug/_trace.py`,
+  `tractor/devx/debug/_tty_lock.py` — lazy `asyncio` +
+  `.to_asyncio`
+- `tractor/spawn/_entry.py` — lazy
+  `run_as_asyncio_guest`
+
+## Human edits
+
+None yet — pending user review at the 12:50PM EST
+pause gate (test-suite results reported in-session).

ai/prompt-io/claude/20260702T155626Z_65bf9df5_prompt_io.raw.md

@@ -0,0 +1,122 @@
+---
+model: claude-fable-5
+service: claude
+timestamp: 2026-07-02T15:56:26Z
+git_ref: 65bf9df5
+diff_cmd: git diff main..wkt/boot_latency_470
+---
+
+# Raw AI output — gh #470 `import tractor` latency trim
+
+All generated code is committed on the
+`wkt/boot_latency_470` branch; per diff-ref mode each
+file's content is referenced via its diff instead of
+copied verbatim.
+
+## Profiling findings (verbatim analysis output)
+
+Baseline: `import tractor` ~0.39-0.42s wall.
+
+`python -X importtime` + `cProfile` traced the cost NOT
+primarily to third-party deps (the issue's hypothesis)
+but to `tractor/log.py:get_logger()` calling
+`get_caller_mod()` -> `inspect.stack()` at module level
+in ~39 tractor modules:
+
+- `inspect.stack()` builds `FrameInfo` (incl. src-file
+  and line-context resolution) for EVERY frame on the
+  stack; during nested imports the stack is dozens of
+  importlib frames deep.
+- each `FrameInfo` resolution calls
+  `inspect.getmodule()` which scans all of
+  `sys.modules` per frame (1.4M `ismodule()` calls in
+  one profiled import).
+- aggregate: ~244ms of tractor-own module "self" time
+  vs ~20ms for ALL the issue-listed third-party deps
+  (`pdbp` ~10ms, `bidict` ~4.5ms, `multiaddr` ~3.5ms,
+  `wrapt`/`colorlog` ~1ms each); `trio` itself is
+  ~70-100ms and unavoidable.
+
+## Generated changes
+
+> `git diff main..wkt/boot_latency_470 -- tractor/log.py`
+
+`get_caller_mod()` rewritten from `inspect.stack()` +
+`inspect.getmodule()` to `sys._getframe(frames_up)` +
+`frame.f_globals['__name__']` -> `sys.modules` lookup
+(O(1) vs O(stack x sys.modules)). Unused `inspect`
+imports dropped; `FrameType` imported from `types`.
+Also `colorlog` lazy-imported inside
+`get_console_log()`.
+
+> `git diff main..wkt/boot_latency_470 -- tractor/discovery/_addr.py`
+
+`bidict` import moved under `TYPE_CHECKING`
+(annotation-only use; `_address_types` is a plain dict
+literal).
+
+> `git diff main..wkt/boot_latency_470 -- tractor/discovery/_multiaddr.py`
+
+`from __future__ import annotations` added; `multiaddr`
+import moved under `TYPE_CHECKING` + function-local
+imports in `mk_maddr()`/`parse_maddr()`.
+
+> `git diff main..wkt/boot_latency_470 -- tractor/ipc/_tcp.py tractor/ipc/_uds.py`
+
+`Multiaddr` imports moved under `TYPE_CHECKING`
+(annotation-only in both transports).
+
+> `git diff main..wkt/boot_latency_470 -- tractor/runtime/_state.py`
+
+`platformdirs` lazy-imported inside `get_rt_dir()`
+(NOTE: still imported eagerly via
+`UDSAddress.def_bindspace` class-var eval; see
+follow-ups).
+
+> `git diff main..wkt/boot_latency_470 -- tractor/devx/_frame_stack.py`
+
+`pdbp` + `wrapt` lazy-imported inside
+`hide_runtime_frames()` / `api_frame()` respectively.
+
+> `git diff main..wkt/boot_latency_470 -- tractor/devx/debug/_trace.py tractor/devx/debug/_tty_lock.py`
+
+`asyncio` moved to `TYPE_CHECKING` + call-site local
+imports (`asyncio.current_task()` sites);
+`tractor.to_asyncio.run_trio_task_in_future` imports
+moved into the infected-aio runtime branches.
+
+> `git diff main..wkt/boot_latency_470 -- tractor/spawn/_entry.py`
+
+`run_as_asyncio_guest` import moved into the
+`infect_asyncio=True` branches of `_mp_main()` /
+`_trio_main()`.
+
+> `git diff main..wkt/boot_latency_470 -- tractor/__init__.py`
+
+PEP 562 module `__getattr__` added so
+`tractor.to_asyncio` attr-access still works (required
+by `tests/test_child_manages_service_nursery.py` and
+any downstream user) while keeping `asyncio` off the
+eager import path.
+
+## Measured results (verbatim)
+
+- `import tractor`: 0.39-0.42s -> ~0.145s (~65% cut)
+- `start_actor` spawn+boot+reg+cancel: ~0.40-0.44s ->
+  ~0.179s/actor (n=5 sequential, warm parent)
+- post-change eager-module check: only `pdbp` +
+  `platformdirs` of the issue's list remain eager.
+
+## Known follow-ups (not implemented, deadline-bound)
+
+- `pdbp` (~10ms): still eager via
+  `devx/debug/_repl.py` class bases
+  (`class PdbREPL(pdbp.Pdb)`) + `_tty_lock.py`
+  module-level `@pdbp.hideframe`; needs `_repl`
+  restructure + PEP 562 in `devx.debug.__init__`.
+- `platformdirs` (~1.5ms): eager via
+  `UDSAddress.def_bindspace: ClassVar = get_rt_dir()`
+  class-body call; needs `Address`-protocol rework of
+  `def_bindspace` to a lazy accessor.
+- `stackscope` + `setproctitle`: already lazy/absent —
+  no change needed.

ai/prompt-io/claude/20260702T161624Z_65bf9df5_prompt_io.md

@@ -0,0 +1,80 @@
+---
+model: claude-fable-5
+service: claude
+session: f4da6654-63f9-4641-b567-f200da46d697
+timestamp: 2026-07-02T16:16:24Z
+git_ref: 65bf9df5
+scope: tests
+substantive: true
+raw_file: 20260702T161624Z_65bf9df5_prompt_io.raw.md
+---
+
+## Prompt
+
+> attempt to resolve
+> https://github.com/goodboy/tractor/issues/474
+> do it with /open-wkt.
+
+Issue #474: "We need test suite for
+`.trionics.start_or_cancel()`" — dig out the use case
+from `modden` as a unit test (and provide motivation
+for upstream `trio` to possibly provide better
+porcelain around this case). The wrapper landed via
+PR #464 with no tests.
+
+## Response summary
+
+New `tests/trionics/test_taskc.py` (9 tests) covering
+`start_or_cancel()`, authored in wkt
+`start_or_cancel_tests_474`.
+
+Behaviour was first probed empirically (trio 0.29):
+the lossy `RuntimeError('child exited without calling
+task_status.started()')` only fires when the child
+exits pre-`.started()` WITHOUT propagating the ambient
+`Cancelled` — i.e. when the child (or lib code it
+calls) absorbs the cancel in a graceful-teardown
+pattern; a well-behaved child surfaces `Cancelled`
+straight out of `.start()`. The `modden`
+`progman.open_wks()` use case was reconstructed from
+`modden/runtime/progman.py` accordingly.
+
+Tests (each `use_start_or_cancel` parametrization also
+pins upstream trio's current lossy behaviour as
+wart-documentation):
+
+- `test_sibling_err_not_masked_by_startup_rte` — the
+  `modden` case: sibling error OOB-cancels the shared
+  nursery scope; with the wrapper ONLY the root-cause
+  `ValueError` escapes; bare `.start()` adds the lossy
+  RTE alongside.
+- `test_pure_oob_cancel_not_morphed_to_rte` — plain
+  ancestor `cs.cancel()`: wrapper → clean exit; bare
+  → eg-wrapped RTE.
+- `test_genuine_startup_rte_still_raised` — no
+  cancellation → protocol-bug RTE re-raised same as
+  bare.
+- `test_childs_own_rte_never_demoted_to_cancel` — a
+  child's own `RuntimeError('never got started!')` /
+  `RuntimeError(1234)` under ambient cancel is never
+  demoted to `Cancelled` (exact-msg-match + str-guard
+  regression cover).
+- `test_started_value_and_args_passthru` — happy path:
+  positional args, `name=`, `.started()` value.
+
+Verified: 9/9 pass; 0 flakes across 50 hammer runs;
+two impl mutations (checkpoint removed; guard relaxed
+to substring match) each caught by exactly the
+targeted tests; `tests/trionics/` +
+`tests/test_trioisms.py` subset green (23 passed,
+5 xfailed); ruff clean; 69-col style.
+
+## Files changed
+
+- `tests/trionics/test_taskc.py` — new
+  `start_or_cancel()` unit-test suite (gh #474).
+
+## Human edits
+
+Pending review — session paused pre-commit per user
+deadline; nothing committed as of this entry.

ai/prompt-io/claude/20260702T161624Z_65bf9df5_prompt_io.raw.md

@@ -0,0 +1,107 @@
+---
+model: claude-fable-5
+service: claude
+timestamp: 2026-07-02T16:16:24Z
+git_ref: 65bf9df5
+diff_cmd: git diff main..wkt/start_or_cancel_tests_474
+---
+
+# Raw output — gh #474 `start_or_cancel()` test suite
+
+## Generated test code
+
+> `git diff main..wkt/start_or_cancel_tests_474 -- tests/trionics/test_taskc.py`
+
+Prose summary of the generated module
+(`tests/trionics/test_taskc.py`):
+
+- module docstring framing the `trio.Nursery.start()`
+  startup-cancellation wart, the wrapper's repair, and
+  the intent that `use_start_or_cancel=False` params
+  double as upstream-trio wart-documentation (break on
+  a trio upgrade → upstream may have shipped porcelain,
+  re-audit the wrapper); cites gh #474 / PR #464 and
+  `modden`'s `progman.open_wks()` as the source use
+  case.
+- shared children: `absorbs_cancel_pre_started()` (the
+  graceful-teardown cancel-absorber which triggers the
+  lossy RTE path) + `raise_val_err()` (fast-erroring
+  sibling).
+- `test_sibling_err_not_masked_by_startup_rte`
+  (parametrized `use_start_or_cancel`): asserts eg
+  contains exactly one `ValueError` and, wrapper-case,
+  NO residual RTE (`eg.split(ValueError)` remainder is
+  `None`); bare-case, the residual RTE carries trio's
+  exact "child exited without calling" wording.
+- `test_pure_oob_cancel_not_morphed_to_rte`
+  (parametrized): wrapper-case runs clean and asserts
+  `cs.cancelled_caught`; bare-case asserts the
+  eg-wrapped RTE.
+- `test_genuine_startup_rte_still_raised`
+  (parametrized): no-cancel protocol bug → RTE with
+  trio's wording from both call forms.
+- `test_childs_own_rte_never_demoted_to_cancel`
+  (parametrized `rte_arg` in `'never got started!'`,
+  `1234`): child cancels the ambient scope then raises
+  its own RTE synchronously (no checkpoint between →
+  deterministically under-cancellation at catch time);
+  asserts the RTE survives with `args[0]` intact.
+- `test_started_value_and_args_passthru`: `.started()`
+  value, positional args and the `name=` kwarg (via
+  `trio.lowlevel.current_task().name`) all forward.
+
+## Non-code output (verbatim highlights)
+
+Behaviour probe (trio 0.29, scratchpad scripts) — the
+decision basis for the test shapes:
+
+```
+== B-sibling-err use_soc=False
+  start raised: RuntimeError('child exited without
+    calling task_status.started()')
+  top-level: ExceptionGroup([ValueError('sibling blew
+    up!'), RuntimeError('child exited without calling
+    task_status.started()')])
+== B-cs-cancel use_soc=False
+  top-level: ExceptionGroup([RuntimeError('child
+    exited without calling task_status.started()')])
+== B-sibling-err use_soc=True
+  start raised: Cancelled()
+  top-level: ExceptionGroup([ValueError('sibling blew
+    up!')])
+== B-cs-cancel use_soc=True
+  start raised: Cancelled()
+  top-level: clean return
+== own-rte-under-cancel (both) -> RTE('never got
+  started!') propagates unchanged
+```
+
+Key finding: with a WELL-BEHAVED (non-absorbing) child
+an OOB ancestor cancel surfaces `Cancelled` directly
+from `.start()` on trio 0.29 — the lossy RTE requires
+the child to absorb its cancel pre-`.started()`, which
+is what `modden`'s `open_from_wks` teardown did. Trio's
+nursery-exit wait defers cancel delivery to children,
+so all tested shapes are deterministic (0 flakes / 50
+runs).
+
+Mutation verification:
+
+```
+mutation 1 (checkpoint_if_cancelled removed):
+  FAILED test_sibling_err_not_masked_by_startup_rte[True]
+  FAILED test_pure_oob_cancel_not_morphed_to_rte[True]
+mutation 2 (guard relaxed to 'started' substring,
+            isinstance dropped):
+  FAILED test_childs_own_rte_never_demoted_to_cancel[never got started!]
+  FAILED test_childs_own_rte_never_demoted_to_cancel[1234]
+```
+
+Final runs:
+
+```
+tests/trionics/test_taskc.py: 9 passed in 0.03s
+hammer: 0/50 runs failed
+tests/trionics/ + tests/test_trioisms.py:
+  23 passed, 5 xfailed in 3.02s
+```

ai/prompt-io/claude/20260702T165806Z_a34aaf98_prompt_io.md

@@ -0,0 +1,78 @@
+---
+model: claude-fable-5
+service: claude
+session: f6c84722-471a-4458-9a80-e453fea9029f
+timestamp: 2026-07-02T16:58:06Z
+git_ref: a34aaf98
+scope: code
+substantive: true
+raw_file: 20260702T165806Z_a34aaf98_prompt_io.raw.md
+---
+
+## Prompt
+
+Follow-up round in the same session as the
+`tractor.to_actor` landing (see
+`20260702T154255Z_65bf9df5_prompt_io.md`). After
+committing that work the user green-lit the deferred
+items:
+
+> go go go on this with what time you have left, in
+> particular see if you can get the _ria_nursery
+> removal going!
+
+then extended the deadline twice to iterate on the
+discovered hang:
+
+> continue on this up until a 12:58:30 deadline
+
+and finally chose "Commit step A now" from the
+next-steps prompt.
+
+## Response summary
+
+Step A of the `._ria_nursery` removal (issue #477): hoist
+`.run_in_actor()` result-reaping out of the spawn
+backends into the `ActorNursery` machinery so ria
+children spawn via the default daemon nursery,
+
+- new `_supervise._reap_ria_portals()` runs one
+  `_spawn.cancel_on_completion()` task per ria child
+  AFTER `._join_procs` is set; happy path awaits it
+  right after `._join_procs.set()`.
+- error path SEQUENCES: snapshot ria
+  `(portal, subactor)` pairs -> `await an.cancel()` ->
+  0.5s-bounded reap. Two failed intermediates informed
+  this: a concurrent reap+cancel DEADLOCKED
+  `test_multierror`; a 3s bound blew
+  `test_cancel_while_childs_child_in_sync_sleep`'s
+  `fail_after` deadline.
+- backends (`spawn/_trio.py`, `spawn/_mp.py`) lose the
+  `._cancel_after_result_on_exit` membership branch,
+  per-child reaper nursery + dead imports.
+- design/probe-history doc:
+  `ai/conc-anal/ria_nursery_removal_plan.md` (from an
+  agent-verified machinery map).
+
+Verification: `test_cancellation.py` fully green
+(20 passed, 1 xfailed) incl. the previously-hung
+`test_multierror`; `test_to_actor`+`test_spawning`
+20/20; bounded full-suite gate SIGINT'd ~30s early at
+303 passed / 0 failures (user opted to commit on that
+signal, deferring the unbounded re-run to step-B
+verification).
+
+## Files changed
+
+- `tractor/runtime/_supervise.py` — `_reap_ria_portals()`
+  + two call-sites; `run_in_actor()` off the ria nursery
+- `tractor/spawn/_trio.py` — reaper branch + import drop
+- `tractor/spawn/_mp.py` — same as `_trio.py`
+- `ai/conc-anal/ria_nursery_removal_plan.md` — plan +
+  probe history
+
+## Human edits
+
+None yet — committed via the drafted
+`.claude/git_commit_msg_ria_step_a.md` (user-driven
+`git commit --edit`).

ai/prompt-io/claude/20260702T165806Z_a34aaf98_prompt_io.raw.md

@@ -0,0 +1,55 @@
+---
+model: claude-fable-5
+service: claude
+timestamp: 2026-07-02T16:58:06Z
+git_ref: a34aaf98
+diff_cmd: git diff a34aaf98..wkt/to_actor_subpkg
+---
+
+# Raw AI output (diff-ref mode)
+
+Step-A code is committed on `wkt/to_actor_subpkg`
+directly after `a34aaf98`; per diff-ref mode the verbatim
+content is reachable via the pointers below.
+
+## Generated files
+
+> `git diff a34aaf98..wkt/to_actor_subpkg -- tractor/runtime/_supervise.py`
+
+New `_reap_ria_portals(an, errors, ria_children=None)`
+helper (one `_spawn.cancel_on_completion()` task per ria
+child under `collapse_eg()` + a local nursery);
+`run_in_actor()` drops `nursery=self._ria_nursery`; happy
+path awaits the reap right after `._join_procs.set()`;
+inner error handler snapshots ria pairs, runs
+`await an.cancel()` then a `move_on_after(0.5)`-bounded
+reap over the snapshot.
+
+> `git diff a34aaf98..wkt/to_actor_subpkg -- tractor/spawn/_trio.py`
+> `git diff a34aaf98..wkt/to_actor_subpkg -- tractor/spawn/_mp.py`
+
+Both backends: the post-`_join_procs` block collapses to
+a bare `soft_kill()` (membership branch, per-child reaper
+nursery, reaper-cancel logging and the now-unused
+`cancel_on_completion` imports all removed).
+
+> `git diff a34aaf98..wkt/to_actor_subpkg -- ai/conc-anal/ria_nursery_removal_plan.md`
+
+Agent-verified machinery map, 3-step design (A/B/C),
+probe history (deadlock -> sequencing fix -> bound
+tightening) and risk register.
+
+## Test runs (verbatim)
+
+```
+tests/test_cancellation.py: 20 passed, 1 xfailed in 77.28s
+tests/test_to_actor.py + tests/test_spawning.py: 20 passed
+full-suite (SIGINT at 540s bound, ~30s short):
+303 passed, 9 skipped, 1 xfailed, 2 xpassed in 539.42s
+```
+
+Failed intermediates (not in final diff): concurrent
+reap+`an.cancel()` deadlocked `test_multierror`; a 3s
+reap bound blew
+`test_cancel_while_childs_child_in_sync_sleep`'s
+`fail_after` deadline.

ai/prompt-io/claude/20260702T172233Z_5cd190c5_prompt_io.md

@@ -0,0 +1,60 @@
+---
+model: claude-opus-4-8
+service: claude
+session: c7860302-6a33-4d92-821c-0807d68be464
+timestamp: 2026-07-02T17:22:33Z
+git_ref: 5cd190c5
+scope: code
+substantive: true
+raw_file: 20260702T172233Z_5cd190c5_prompt_io.raw.md
+---
+
+## Prompt
+
+Branched session (`to_actor_subpkg_POST_WEEKLY_CUTOFF_2OPUS`,
+now on Opus 4.8) continuing the `._ria_nursery` removal after
+step A committed + pushed (`5cd190c5`/`99310269`):
+
+> ok continue where your superior left off please
+
+then, after the commit-split discussion:
+
+> yup committed and pushed so take a look ma fren.
+
+i.e. green-lighting step-B prep against the committed step-A
+tree.
+
+## Response summary
+
+Step B of the `._ria_nursery` removal (issue #477): delete
+the now-vestigial secondary nursery,
+
+- collapse the inner `async with (collapse_eg(),
+  trio.open_nursery() as ria_nursery)` layer in
+  `_open_and_supervise_one_cancels_all_nursery` — `da_nursery`
+  becomes the single nursery for ALL subactors.
+- `ActorNursery.__init__` drops the `ria_nursery` param + the
+  `self._ria_nursery` attr; `start_actor()` drops its
+  `nursery=` escape-hatch param.
+- `._cancel_after_result_on_exit` kept (ria-child
+  discriminator for `_reap_ria_portals()`).
+
+Verified behavior-preserving via a first-principles argument
+(zero-task nursery = a bare checkpoint) + the targeted gate
+(`test_cancellation test_spawning test_local test_rpc
+test_to_actor` = 49 passed, 1 xfailed on trio). The two
+error handlers were deliberately NOT merged — that changes
+propagation semantics and is deferred to its own PR (TODO
+left at the outer `except`).
+
+## Files changed
+
+- `tractor/runtime/_supervise.py` — collapse the ria nursery
+  layer + drop the ctor/`start_actor` params + refresh the
+  now-stale nursery comments
+
+## Human edits
+
+None yet — committed via the drafted
+`.claude/git_commit_msg_ria_step_b.md` (user-driven
+`git commit --edit`).

ai/prompt-io/claude/20260702T172233Z_5cd190c5_prompt_io.raw.md

@@ -0,0 +1,51 @@
+---
+model: claude-opus-4-8
+service: claude
+timestamp: 2026-07-02T17:22:33Z
+git_ref: 5cd190c5
+diff_cmd: git diff 5cd190c5..wkt/to_actor_subpkg
+---
+
+# Raw AI output (diff-ref mode)
+
+Step-B code lives on `wkt/to_actor_subpkg` after `5cd190c5`;
+per diff-ref mode the verbatim content is reachable via the
+pointer below.
+
+## Generated files
+
+> `git diff 5cd190c5..wkt/to_actor_subpkg -- tractor/runtime/_supervise.py`
+
+- `ActorNursery.__init__`: `ria_nursery` param removed;
+  `self._ria_nursery = ria_nursery` block deleted;
+  `_cancel_after_result_on_exit` comment refreshed.
+- `start_actor()`: `nursery=` param removed; body uses
+  `self._da_nursery.start(...)` directly.
+- `_open_and_supervise_one_cancels_all_nursery()`: the inner
+  `async with (collapse_eg(), trio.open_nursery() as
+  ria_nursery)` layer removed; `an = ActorNursery(actor,
+  da_nursery, errors)` constructed once under the single
+  `da_nursery`; the inner-try body de-indented one level;
+  both error handlers retained; the da-nursery lead comment
+  and the outer-`except` TODO refreshed to describe the
+  single-nursery reality + flag the (deferred) handler-merge.
+
+> `git diff 5cd190c5..wkt/to_actor_subpkg -- ai/conc-anal/ria_nursery_removal_plan.md`
+
+Added a "Step-B outcome" section (collapse rationale,
+handler-merge deferral, safety argument, gate result).
+
+## Test runs (verbatim)
+
+```
+targeted gate (trio):
+tests/test_cancellation.py tests/test_spawning.py
+tests/test_local.py tests/test_rpc.py tests/test_to_actor.py
+-> 49 passed, 1 xfailed in 88.62s
+
+signature checks:
+ActorNursery.__init__ params: ['self', 'actor', 'da_nursery', 'errors']
+start_actor has nursery param: False
+
+full-suite gate: in progress (see stepB_full.log)
+```

ai/prompt-io/claude/20260702T222544Z_9201a2ed_prompt_io.md

@@ -0,0 +1,68 @@
+---
+model: claude-opus-4-8
+service: claude
+session: c7860302-6a33-4d92-821c-0807d68be464
+timestamp: 2026-07-02T22:25:44Z
+git_ref: 9201a2ed
+scope: code
+substantive: true
+raw_file: 20260702T222544Z_9201a2ed_prompt_io.raw.md
+---
+
+## Prompt
+
+Continuing on branch `drop_ria_nursery` after step B
+(`9201a2ed`). User green-lit the deferred error-handler merge:
+
+> yes, i checked us out on a new branch for the ria-tn removal;
+> work off the new wkt branch drop_ria_nursery.
+
+and (re the migration) "nope, later we'll do this", plus:
+
+> do ensure the test suite is green in the current state and
+> those going forward ya?
+
+and the standing reminder: no commits without a human ack.
+
+## Response summary
+
+Step B2 of the `._ria_nursery` removal (issue #477): merge the
+two nested error handlers in
+`_open_and_supervise_one_cancels_all_nursery` into one, now
+that the 2ndary nursery (whose `__aexit__` the outer handler
+originally guarded) is gone.
+
+Trace-driven justification (see
+`ai/conc-anal/ria_nursery_removal_plan.md` Step-B2 section):
+the old inner handler records `errors[uid]` as its FIRST
+action, so `errors` is always non-empty by the time control
+could reach the outer handler — making the outer handler's
+`raise` DEAD (always superseded by the `finally`'s
+raise-from-`errors`). Collapsing to a single
+`except BaseException as _scope_err` + the existing `finally`
+is therefore behavior-preserving.
+
+- removed the outer `try` + the
+  `except (Exception, BaseExceptionGroup, trio.Cancelled)`
+  handler; de-indented the inner handler one level.
+- dropped the now-unused `outer_err`/`inner_err` locals
+  (single `_scope_err`).
+- net ~30 lines lighter.
+
+Gated hard: baseline (pre-B2 tip) = 300 passed + 1
+load-induced `TooSlowError` flake (passes 6/6 in isolation);
+B2 error/cancel gate (`test_cancellation remote_exc_relay
+inter_peer_cancellation advanced_faults oob_cancellation
+to_actor spawning local rpc`) = 71 passed, 1 xfailed; B2
+full-suite run following.
+
+## Files changed
+
+- `tractor/runtime/_supervise.py` — collapse the two handlers
+  into one; drop `outer_err`/`inner_err`
+
+## Human edits
+
+None yet — committed via the drafted
+`.claude/git_commit_msg_ria_b2.md` (user-driven
+`git commit --edit`).

ai/prompt-io/claude/20260702T222544Z_9201a2ed_prompt_io.raw.md

@@ -0,0 +1,55 @@
+---
+model: claude-opus-4-8
+service: claude
+timestamp: 2026-07-02T22:25:44Z
+git_ref: 9201a2ed
+diff_cmd: git diff 9201a2ed..drop_ria_nursery
+---
+
+# Raw AI output (diff-ref mode)
+
+Step-B2 code lives on `drop_ria_nursery` after `9201a2ed`; per
+diff-ref mode the verbatim content is reachable via the pointer
+below.
+
+## Generated files
+
+> `git diff 9201a2ed..drop_ria_nursery -- tractor/runtime/_supervise.py`
+
+`_open_and_supervise_one_cancels_all_nursery`:
+- removed the outer `try:` wrapper and the
+  `except (Exception, BaseExceptionGroup, trio.Cancelled) as
+  _outer_err:` safety-net handler.
+- the former inner `except BaseException` is now THE handler,
+  renamed local `_inner_err` -> `_scope_err`, de-indented one
+  level; it sets `an._scope_error`, records `errors[uid]`,
+  waits on the debugger, `_join_procs.set()`, then a shielded
+  classify/log + snapshot-ria + `an.cancel()` + 0.5s-bounded
+  `_reap_ria_portals()`. No re-raise (the `finally` raises
+  from `errors`).
+- `finally` block unchanged.
+- dropped the `outer_err`/`inner_err` local decls at fn top.
+
+(The diff is large — ~119+/149- — because de-indenting the
+handler body one level rewrites every line in the block; the
+logic delta is just "two handlers -> one".)
+
+## Test runs (verbatim)
+
+```
+baseline (pre-B2, step-B tip 9201a2ed), full suite
+(dynamic_pub_sub deselected):
+1 failed, 300 passed, 9 skipped, 2 deselected, 1 xfailed,
+2 xpassed in 1499.49s
+  -> the 1 failure = test_ext_types_over_ipc[...] trio.TooSlowError
+     (load-induced; passes 6/6 in isolation in 4.89s)
+
+B2 error/cancel gate:
+tests/test_cancellation test_remote_exc_relay
+test_inter_peer_cancellation test_advanced_faults
+test_oob_cancellation test_to_actor test_spawning test_local
+test_rpc
+-> 71 passed, 1 xfailed in 125.26s
+
+B2 full-suite run: see b2_full.log
+```

ai/prompt-io/claude/20260706T172818Z_ad42871e_prompt_io.md

@@ -0,0 +1,81 @@
+---
+model: claude-fable-5
+service: claude
+session: 6db64ac6-6986-4505-9343-df4ee31e67db
+timestamp: 2026-07-06T17:28:18Z
+git_ref: ad42871e
+scope: code
+substantive: true
+raw_file: 20260706T172818Z_ad42871e_prompt_io.raw.md
+---
+
+## Prompt
+
+Resuming the `drop_ria_nursery` (wkt) branch after the human's
+review pause at `ba4af0ad`, with the directive:
+
+> we need to resolve the remaining outstandings ... to finish
+> this endeavour we ideally,
+> - finish the ria-tn hoist into the machinery in `to_actor` if
+>   possible.
+> - avoid supporting the `.run_in_actor()` non-blocking API
+>   going forward other then if we can re-wrap `to_actor.run()`
+>   per prior discussion with opus4.8.
+> - get the test harness completely green locally and in CI -
+>   AND ESPECIALLY without any hangs due to the "the old
+>   spawn-backend reaper raced `soft_kill()`'s scope-cancel"
+> - update all docs to reflect these changes once we're happy.
+
+plus mid-flight additions:
+
+> make sure both all the test and example code uses the naming
+> `an: ActorNursery` for variable names
+
+> where we at in terms of a potential hackaround design for
+> supporting a to_actor.run_async() ... without requiring the
+> same malarky ... with a `trio.Event` relay ... i'm totes fine
+> with rejigging the iface to NOT return a Portal as well.
+
+## Response summary
+
+Completed the entire #477 `run_in_actor` removal on
+`drop_ria_nursery` (16 commits this session), gated per-commit:
+
+- ported the remaining `test_cancellation` groups
+  (`some_cancels_all`, `nested_multierrors`+`spawn_and_error`,
+  SIGINT + sync-sleep) — one-shots become concurrent
+  `to_actor.run()` tasks in local task-nurseries; BEG-shape
+  assertions loosened for cancel-on-first + the runtime's
+  `collapse_eg()` single-member unwrap (a fully-raced nested
+  tree relays a bare annotated `RemoteActorError` chain).
+- fixed a pre-existing `UnboundLocalError` (`timeout` `match`
+  had no default arm for non-trio/MTF backends).
+- ported `test_dynamic_pub_sub`, 4 non-debugging examples, all
+  8 `debugging/` examples (debugger suite byte-identical,
+  28p/6s; `multi_subactors` introduces the "collect don't
+  cancel" reap-all replacement pattern), 8 docs pages + the
+  `experimental/_pubsub` docstring.
+- EXCISED the API + cluster: `run_in_actor`,
+  `_reap_ria_portals`, `_cancel_after_result_on_exit`,
+  `Portal._submit_for_result/_expect_result_ctx/
+  wait_for_result/result`, `exhaust_portal`,
+  `cancel_on_completion`, `NoResult` — net -402 lines. The
+  reap-hang class dissolves structurally (result-waits now only
+  in caller task-scope).
+- found + fixed a real migration race: mutual-rendezvous peers
+  (`test_trynamic_trio`, `a_trynamic_first_scene.py`) flaked
+  because an eagerly-reaped one-shot dies while its peer still
+  dials the registry-resolved (dead) sockaddr — such peers now
+  pin lifetimes via `start_actor()` + concurrent `Portal.run()`
+  + explicit `an.cancel()`.
+- `an: ActorNursery` naming sweep across tests/examples (±82
+  lines, scoped renames, prose untouched).
+- parked a `to_actor.open_one_shot()` design sketch (acm +
+  private task-nursery over blocking `run()`; done-Event as
+  memo not cancel-relay; no Portal) in the plan doc.
+
+## Files changed
+
+See commits `d01a2123..ad42871e` on `drop_ria_nursery`
+(tests, examples, docs, `tractor/{runtime,spawn,to_actor,msg}`
++ `_exceptions/_context/experimental`).

ai/prompt-io/claude/20260706T172818Z_ad42871e_prompt_io.raw.md

@@ -0,0 +1,39 @@
+---
+model: claude-fable-5
+service: claude
+timestamp: 2026-07-06T17:28:18Z
+git_ref: ad42871e
+diff_cmd: git diff ba4af0ad..ad42871e
+---
+
+# Raw AI output (diff-ref mode)
+
+This session's output spans the 16 migration/excision commits
+`d01a2123..ad42871e` on `drop_ria_nursery`; per diff-ref mode
+the verbatim content is reachable via the pointer below.
+
+## Generated files
+
+> `git diff ba4af0ad..ad42871e`
+
+Commit-wise (each `Gate:`-footed msg documents its own module
+gate):
+
+- `d01a2123` port `test_some_cancels_all`
+- `697c6152` fix unbound `timeout` (non-trio/MTF `match` arm)
+- `fa8799d5` port `test_nested_multierrors`
+- `f11754ce` port SIGINT + sync-sleep cancel tests
+- `cb6202e3` port `test_dynamic_pub_sub`
+- `d8af5f12` port non-debugging examples
+- `a3057cb2` port debugging examples (+ `test_debugger`
+  nested-nurseries final-shape expectations)
+- `d6bed7c4` port docs (8 rst pages)
+- `07e1669e` fix stale `@pub` docstring example
+- `2a59cefb` REMOVE `run_in_actor()` + the ria reap cluster
+  (net -402 lines)
+- `a297a32a` fix mutual-rendezvous premature-reap race
+- `ad42871e` `an: ActorNursery` naming sweep
+
+Plan/design record updated in
+`ai/conc-anal/ria_nursery_removal_plan.md` (RESOLVED section +
+the `to_actor.open_one_shot()` follow-up sketch).

ai/prompt-io/claude/20260813T001102Z_27c34aeb_prompt_io.md

@@ -0,0 +1,136 @@
+---
+model: claude-opus-5
+service: claude
+session: 7b9c97c4-fff7-4ac4-97fb-35720453308e
+timestamp: 2026-08-13T00:11:02Z
+git_ref: 27c34aeb
+scope: docs+code
+substantive: true
+raw_file: 20260813T001102Z_27c34aeb_prompt_io.raw.md
+---
+
+## Prompt
+
+> draft hyper detailed implementation plans for [three]
+> prospective new transport (tpt) backends for tractor's `.ipc`
+> layer, from four GitHub issues: TIPC (gh #378) using built-in
+> linux socket API w/ `trio` interfacing, leveraging TIPC's
+> built-in discovery machinery; QUIC (gh #353) using the `iroh`
+> lib, ideally with the py asyncio support (via ffi) rewritten
+> for trio; wg (gh #482 and/or #443) with other shuttle-able
+> tpts, using `pyroute2`, as much trio wrapping as possible
+> where any other async support can be replaced.
+
+With constraints: "be only slightly speculative"; realistic given
+the linked info; the plans must be workable "by another
+model/provider without much discrepancy on the design or lib
+selections needed"; do it all in a new `/open-wkt` off main; and
+be "as modular and modern-async-python-as-nearly-functional-style
+as possible."
+
+Key follow-up prompts that changed the design mid-session:
+
+> wg requires binding a socket on both ends of the tunnel, the
+> new support we added to `py-multiaddr` in their #108 enables
+> declaring this in the maddr? […] so the
+> `/ip4/10.0.0.1/udp/666/ip4/192.168.0.13/tcp/80` bullet example
+> from #443 isn't correct for a wg tunnel maddr and would
+> actually end up looking like,
+> `/ip4/10.0.0.1/udp/666/wg/<pub-key>/ip4/192.168.0.13/tcp/80` no?
+
+> right, we're likely going to need less `tuple`-formed
+> `UnwrappedAddress` usage to make them all work and/or to extend
+> the unwrapped form with a proto-str, likely the key from
+> `multiaddrs` no?
+
+> better encapsulating `UnwrappedAddress` for use externally is
+> really way overdue; we should likely be encouraging (if not
+> preventing unwrapped) wrapped addrs always much like other
+> (std) libs, `ipaddress` comes to mind.
+
+> so iirc netns mgmt becomes a lower-level runtime config-api
+> rather then an actor-app-code API which can be dynamically
+> changed once the actor is up ya?
+
+## Response summary
+
+A research-and-planning branch, `ng_tpts_planning` — **no runtime
+code touched**. Five plan docs under `ai/tpt-backends/` plus a
+corrected `wg` example set under `examples/multihost/wg_lan/`,
+across 9 commits.
+
+Three design conclusions came out of the planning that outlive
+the branch:
+
+1. **`wg` is a bindspace, not a `MsgTransport`.** The `/wg/`
+   maddr segment is *infix*: bearer (underlay ip+udp
+   `ListenPort`, bound by the kernel) → `/wg/u<key>` (an
+   identity, bound by nothing) → overlay (the only part tractor
+   binds). Verified empirically by installing
+   `baudco/py-multiaddr@wg_support` (py-multiaddr#108) in a
+   throwaway venv and round-tripping each candidate form; gh
+   #482's original suffix form parses but is semantically
+   inverted.
+2. **Proto-key the `UnwrappedAddress` form**, spelled with the
+   `multiaddr` protocol names, dispatched via
+   `_address_types[addr[0]]`. Kills a whole collision class
+   (TIPC's `(str, int)` ≡ `TCPAddress`; iroh's `(str, str)`
+   swallowed by the UDS case) and is the recommended migration
+   *before* any new backend lands.
+3. **netns is a runtime/boot-time config API, not an app-code
+   one** — `setns(2)` is per-thread and won't move
+   already-created sockets, so there is deliberately no
+   `await actor.enter_netns(...)`.
+
+Also verified that `trio.SocketStream`/`SocketListener` are
+address-family agnostic (no `AF_*` check anywhere), which is what
+makes TIPC the cheapest of the three backends to add.
+
+Four related issues were annotated with the results (#378, #353,
+#482, #443); #443's body was rewritten to reflect the corrected
+grammar, with no existing checkbox state changed.
+
+## Files changed
+
+- `ai/tpt-backends/00_shared_backend_contract.md` — normative
+  backend duck-type contract, registration checklist, §1.1
+  proto-key conclusion
+- `ai/tpt-backends/01_tipc_backend.md` — TIPC plan; service
+  addressing, `TIPC_TOP_SRV` push registry, instance-collision
+  hazard, step-0 probe
+- `ai/tpt-backends/02_quic_iroh_backend.md` — `iroh` plan;
+  `uniffi`→`trio` bridge, listener/stream adapters, API-truth
+  table
+- `ai/tpt-backends/03_wg_tunnel_bindspace.md` — `wg`-as-bindspace
+  plan; verified maddr grammar, 3-owner split, netns reality
+- `ai/tpt-backends/README.md` — index
+- `examples/multihost/wg_lan/wg_maddr.py` — frozen `msgspec`
+  tunnelled addr + pure parse/render helpers; impure
+  `verify_wg_peer()` kept separate
+- `examples/multihost/wg_lan/host_a_srv.py` — host-A actor tree
+- `examples/multihost/wg_lan/host_b_client.py` — host-B dialer
+- `examples/multihost/wg_lan/README.md` — grammar, owner table,
+  setup, "what changed vs #482"
+
+## Human edits
+
+Substantial human steering rather than post-hoc editing; the
+corrections were applied by the model in-session after being
+challenged:
+
+- rejected an initial claim that `wg` has "nothing to bind at the
+  tunnel layer" and supplied the correct composed maddr form,
+  which forced a rewrite of plan 03 §3.2 and a retraction in the
+  already-posted #443 comment
+- rejected a supporting claim that `/ip4/../udp/443/quic-v1` was
+  "also composed"
+- directed the proto-key/`ipaddress`-discipline conclusion and
+  the netns-as-runtime-config framing, both of which were then
+  folded back into the docs
+- chose the commit boundaries and authored all commits; ran every
+  `git` mutation (commit, rebase, push) themselves
+
+One model-initiated correction pre-publication: a self-review
+downgraded two overconfident claims (the `uniffi`/asyncio thesis
+and TIPC duplicate-binder behaviour) to explicitly-flagged
+assumptions before the #353/#378 comments were posted.

ai/prompt-io/claude/20260813T001102Z_27c34aeb_prompt_io.raw.md

@@ -0,0 +1,165 @@
+---
+model: claude-opus-5
+service: claude
+timestamp: 2026-08-13T00:11:02Z
+git_ref: 27c34aeb
+diff_cmd: git diff main..ng_tpts_planning
+---
+
+# Raw output — next-gen tpt-backend implementation plans
+
+## Generated planning docs
+
+> `git diff main..ng_tpts_planning -- ai/tpt-backends/`
+
+Five markdown docs. `00_shared_backend_contract.md` is normative
+and the other three are written against it so they can be worked
+independently:
+
+- **`00_shared_backend_contract.md`** — the backend duck-type
+  (`<Proto>Address(msgspec.Struct, frozen=True)` + module-level
+  `start_listener()`/`close_listener()` + a
+  `Msgpack<Proto>Stream(MsgpackTransport)`), the
+  `inspect.getmodule(self.addr)` reflection in
+  `Endpoint.start_listener()` that forces the Address class and
+  its listener fns to share a module, a 10-item registration
+  checklist, the dep policy, the test-harness shape, and §1.1's
+  proto-key conclusion (below).
+- **`01_tipc_backend.md`** — service addressing via
+  `TIPC_ADDR_NAMESEQ` (bind/publish) and `TIPC_ADDR_NAME`
+  (connect/lookup), `TIPC_TOP_SRV` topology subscriptions as a
+  push-based registry, the `get_random()` instance-collision
+  hazard, and a step-0 capability-probe spike.
+- **`02_quic_iroh_backend.md`** — `iroh` over
+  `aioquic`/`quiche`/`trio-asyncio`, a `_uniffi_trio.py` bridge
+  built on `TrioToken.run_sync_soon()`, `trio.abc.Listener`/
+  `HalfCloseableStream` adapters, and an API-truth table to fill
+  in during step 0.
+- **`03_wg_tunnel_bindspace.md`** — `wg` as a *bindspace* rather
+  than a `MsgTransport`, a `TunnelledAddress` wrapper delegating
+  `.proto_key`/`.unwrap()` to `.inner`, `pyroute2` for layer B,
+  and `@acm`-managed netns/iface for layer C.
+- **`README.md`** — index.
+
+## Generated example code
+
+> `git diff main..ng_tpts_planning -- examples/multihost/wg_lan/`
+
+- `wg_maddr.py` — `WGTunnelledAddr(msgspec.Struct, frozen=True)`
+  carrying `bearer: tuple[str, int]`, `peer_pubkey: str`,
+  `inner: tuple[str, int]`, `inner_proto: Literal['tcp']`, plus a
+  `.maddr` property that re-renders the canonical form. Pure
+  helpers `mb_pubkey()`, `wg8_pubkey()`, `parse_wg_maddr()`, and
+  `_segments()` (with a marked stopgap for when the `wg` codec
+  isn't installed). `verify_wg_peer()` is impure **by design** and
+  kept out of the parse path.
+- `host_a_srv.py` / `host_b_client.py` — the two-host runs; both
+  pass only `addr.inner` to `open_nursery()`/`open_root_actor()`.
+- `README.md` — grammar, owner table, `#108`-branch install line,
+  tunnel setup, "what changed vs #482".
+
+## Verified findings (non-code, verbatim)
+
+### `trio` is address-family agnostic
+
+Read against the installed `trio`. `SocketStream`/`SocketListener`
+ctor checks are only "is a trio sock object" + `type ==
+SOCK_STREAM`, plus an `OSError`-**suppressed** `SO_ACCEPTCONN`
+probe. No `AF_*` check anywhere; `TCP_NODELAY`/`TCP_NOTSENT_LOWAT`
+are set under `suppress(OSError)`. A TIPC `SOCK_STREAM` sock should
+therefore drop straight into `trio.serve_listeners()` with the
+existing `MsgpackTransport` framing, making TIPC mostly
+table-registration boilerplate w/ zero new deps.
+
+### the `wg` maddr grammar — `/wg/` is infix, not suffix
+
+Installed `baudco/py-multiaddr@wg_support` (PR
+multiformats/py-multiaddr#108) into a throwaway venv and
+round-tripped every candidate form:
+
+| maddr | `[p.name for p in m.protocols()]` |
+| --- | --- |
+| `/ip4/1.2.3.4/udp/51820/wg/u<k>` | `['ip4','udp','wg']` |
+| `/ip4/../udp/../wg/u<k>/ip4/../tcp/..` | `['ip4','udp','wg','ip4','tcp']` |
+| `/ip4/10.0.11.1/tcp/1616/wg/u<k>` | `['ip4','tcp','wg']` |
+
+```
+/ip4/192.168.1.50/udp/51820/wg/u<A_pub>/ip4/10.0.11.1/tcp/1616
+\_______ bearer __________/\__ key __/\______ overlay ______/
+```
+
+Segments *before* `/wg/` are the bearer — the underlay
+`(ip, udp-port)` that `wg(8)` itself listens on (`ListenPort`).
+Segments *after* are the overlay endpoint, the only part tractor
+binds. The third row above is #482's original suffix form: it
+parses, but is semantically inverted.
+
+Three parts, three owners — and only one is an `Endpoint`:
+
+| part | bound by | in the runtime? |
+| --- | --- | --- |
+| bearer | kernel, via `wg-quick`/`pyroute2` | no |
+| `/wg/u<key>` | nothing — an identity | no, verified out-of-band |
+| overlay | `tractor`'s `IPCServer` | yes, as `.inner` |
+
+### proto-key-tagged `UnwrappedAddress`
+
+Shape-matching in `wrap_address()` does not survive four backends.
+TIPC's natural unwrapped form is a `(str, int)`, indistinguishable
+from `TCPAddress`; iroh's is a `(str, str)`, already swallowed by
+the existing UDS case (`case (_, filename) if type(filename) is
+str`). Ordering hacks and prefix-tagging only paper over it.
+
+Recommended prerequisite for all three backends: carry an explicit
+proto-key spelled with the `multiaddr` protocol name —
+`('tcp', host, port)`, `('unix', path)`,
+`('tipc', stype, inst, scope)` — so `wrap_address()` collapses to
+`_address_types[addr[0]]` and the collision class stops existing.
+This also makes the on-wire form agree with
+`mk_maddr()`/`parse_maddr()` instead of being an independent
+invention. It is a wire-format change (`SpawnSpec`,
+`_root_mailbox`, `_registry_addrs`) plus every fixture and
+downstream config, so it wants its own migration commit landed
+before any new backend — and it is the moment to stop handing raw
+tuples to users at all, making `Address` the public currency and
+`UnwrappedAddress` an internal serialization detail (the
+discipline `ipaddress` uses).
+
+### netns is a runtime-level config API
+
+`setns(2)` affects the calling thread only and does not move
+already-created sockets. So a netns is a spawn/boot-time input
+alongside `enable_transports`/`tpt_bind_addrs`, and there is
+deliberately no `await actor.enter_netns(...)` — a mid-life API
+would silently leave the IPC server bound in the old namespace.
+Corollary for layer B: pass `netns=` down to `pyroute2` rather
+than assuming a `trio.to_thread` worker inherits it.
+
+### `examples/` collection would have failed CI
+
+`tests/test_docs_examples.py` walks `examples/` recursively and
+subproc-runs every collected file asserting `rc == 0`. Its filter
+never checks the extension, so all four `wg_lan` files were
+collected — including `README.md`, which would have been run as
+`python README.md`. `'multihost' not in p[0]` was already in the
+exclusion list with no directory using it. Moving the set under
+`examples/multihost/wg_lan/` drops collection 24 → 20 with zero
+test changes; confirmed via `pytest --collect-only`.
+
+## Corrections applied during the session
+
+The human corrected two claims that had been asserted without
+verification, both since retracted in-place in the docs and in the
+posted issue comments:
+
+1. that `wg` has "nothing to bind at the tunnel layer, exactly one
+   bind" — wrong; a wg stack is genuinely composed, and the real
+   axis is *who owns* each layer's endpoint.
+2. that `/ip4/../udp/443/quic-v1` was "also composed" — wrong;
+   that is one endpoint with a protocol qualifier, not a tunnel.
+
+A self-review before publication also downgraded two
+overconfident claims to explicitly-flagged assumptions: the
+`uniffi`-uses-asyncio-only-as-executor thesis (contradicted that
+plan's own "do not guess from memory" step 0) and TIPC's
+duplicate-binder round-robin behaviour (unverified).

ai/prompt-io/claude/README.md

@@ -0,0 +1,27 @@
+# AI Prompt I/O Log — claude
+
+This directory tracks prompt inputs and model
+outputs for AI-assisted development using
+`claude` (Claude Code).
+
+## Policy
+
+Prompt logging follows the
+[NLNet generative AI policy][nlnet-ai].
+All substantive AI contributions are logged
+with:
+- Model name and version
+- Timestamps
+- The prompts that produced the output
+- Unedited model output (`.raw.md` files)
+
+[nlnet-ai]: https://nlnet.nl/foundation/policies/generativeAI/
+
+## Usage
+
+Entries are created by the `/prompt-io` skill
+or automatically via `/commit-msg` integration.
+
+Human contributors remain accountable for all
+code decisions. AI-generated content is never
+presented as human-authored work.

ai/prompt-io/opencode/20260804T030309Z_65bf9df5_prompt_io.md

@@ -0,0 +1,39 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: moc-teardown-completion-20260804
+timestamp: 2026-08-04T03:03:09Z
+git_ref: 65bf9df5
+scope: code
+substantive: true
+raw_file: 20260804T030309Z_65bf9df5_prompt_io.raw.md
+---
+
+## Prompt
+
+Patch Tractor's `maybe_open_context()` so the final consumer waits for
+resource `__aexit__()` completion and receives cleanup errors. Reuse
+`outcome.Outcome` for the exit result; use the smaller mutable
+`_CtxExit` holder if that makes the implementation simpler. Add and run
+the relevant existing unit tests, but do not commit or push the patch.
+After reviewing the result, simplify `_CtxExit` back to an optional
+exception because the success outcome carries no useful value.
+
+## Response summary
+
+Added an exception-backed completion handshake between
+`_Cache.run_ctx()` and the final `maybe_open_context()` consumer.
+Serialized consumer registration and final teardown under the per-key
+lock, preserving that lock for queued entrants. Added deterministic
+regressions for normal exit, cleanup errors, cancellation interactions,
+service-nursery cancellation, and teardown re-entry.
+
+## Files changed
+
+- `tractor/trionics/_mngrs.py` - publish and unwrap cached exit outcomes.
+- `tests/test_resource_cache.py` - cover completion and cancellation.
+
+## Human edits
+
+The user directed the final simplification from `outcome.Outcome` to an
+optional exception field. The patch remains uncommitted.

ai/prompt-io/opencode/20260804T030309Z_65bf9df5_prompt_io.raw.md

@@ -0,0 +1,36 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-04T03:03:09Z
+git_ref: 65bf9df5
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Implemented cached-context exit completion in
+`tractor.trionics.maybe_open_context()`.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_mngrs.py`
+
+The generated implementation adds `_CtxExit`, whose `done` event
+publishes an `outcome.Outcome[None]`. `_Cache.run_ctx()` records either
+`Value(None)` or `Error(exc)` after the resource exit attempt. The final
+MOC consumer signals `no_more_users`, waits for completion under a
+shielded cancel scope, removes the per-key lock, and unwraps the outcome
+so ordinary cleanup failures are raised at the consumer boundary.
+`trio.Cancelled`, `KeyboardInterrupt`, and `SystemExit` continue through
+the service task rather than being converted into regular cleanup
+errors.
+
+> `git diff HEAD~1..HEAD -- tests/test_resource_cache.py`
+
+The generated regressions cover successful exit blocking, cleanup-error
+delivery, final-user cancellation, cancellation combined with a cleanup
+error, and service-nursery cancellation. The existing teardown re-entry
+test now uses explicit events instead of a ten-second cleanup sleep and
+asserts that the replacement resource is a fresh cache miss.
+
+Verification:
+
+`env PYTHONPATH="$PWD" /home/goodboy/repos/tractor/py313/bin/python -m pytest tests/test_resource_cache.py`
+
+Result: `16 passed in 8.37s`.

ai/prompt-io/opencode/20260811T233833Z_7cbd64ee_prompt_io.md

@@ -0,0 +1,30 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_0799212ebffe42arY96czXn89F
+timestamp: 2026-08-11T23:38:33Z
+git_ref: 7cbd64ee
+scope: code
+substantive: true
+raw_file: 20260811T233833Z_7cbd64ee_prompt_io.raw.md
+---
+
+## Prompt
+
+Open a new isolated worktree in the local Tractor repository and draft a fix
+for `BroadcastReceiver` reporting that a lagged one-slot consumer dropped
+zero values when one value had actually been displaced.
+
+## Response summary
+
+Corrected the off-by-one lag count while preserving cursor recovery and added
+deterministic narrow- and wider-window regressions for exact loss reporting.
+
+## Files changed
+
+- `tractor/trionics/_broadcast.py` - exact broadcast overrun count.
+- `tests/test_task_broadcasting.py` - lag count and recovery regression.
+
+## Human edits
+
+None - generated output follows the user's diagnosed edge case.

ai/prompt-io/opencode/20260811T233833Z_7cbd64ee_prompt_io.raw.md

@@ -0,0 +1,50 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-11T23:38:33Z
+git_ref: 7cbd64ee
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+The user asked for a Tractor fix in a new isolated worktree after a live piker
+failure reported:
+
+```text
+tractor.trionics._broadcast.Lagged:
+Task `piker.brokers.ib.broker.handle_order_requests` overrun and
+dropped `0` values
+```
+
+Inspection showed the lag exception was valid but its count was off by one.
+`BroadcastReceiver.receive_nowait()` treats `seq` as a deque index. With a
+one-entry queue, index zero is the only retained value and `seq == 1` already
+means one value was displaced. The old `seq - maxlen` calculation therefore
+reported zero instead of one.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_broadcast.py`
+
+Adjusted the lag count to `seq - maxlen + 1` and documented why the first
+invalid deque index must be included. The existing Tokio-style cursor reset
+remains unchanged.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added a deterministic parameterized regression covering a one-slot queue
+with one dropped value and a three-slot queue with two dropped values. The
+test keeps the root receiver idle while a child subscriber synchronously
+drains each produced value, asserts the exact `Lagged` message, and proves the
+next receive resumes at the oldest retained item.
+
+Verification output:
+
+```text
+..                                                                       [100%]
+2 passed in 0.04s
+
+..........                                                               [100%]
+10 passed in 4.65s
+```
+
+The targeted import resolved to the new Tractor worktree. Python compilation
+and `git diff --check` passed. Adversarial review found no actionable issues;
+zero-capacity channels remain a pre-existing untested edge outside this fix.

ai/prompt-io/opencode/20260812T012324Z_06c4af17_prompt_io.md

@@ -0,0 +1,31 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_0799212ebffe42arY96czXn89F
+timestamp: 2026-08-12T01:23:24Z
+git_ref: 06c4af17
+scope: code
+substantive: true
+raw_file: 20260812T012324Z_06c4af17_prompt_io.raw.md
+---
+
+## Prompt
+
+Iteratively refine Tractor PR 490. For item one, correct broadcast statistics
+queue counts and Trio event checks, verify and review the exact change, then
+return a complete commit plan before proceeding.
+
+## Response summary
+
+Converted subscriber cursor indexes into clamped retained queue counts,
+removed deprecated event truthiness, and added deterministic state and
+deprecation regressions.
+
+## Files changed
+
+- `tractor/trionics/_broadcast.py` - accurate queue and waiter statistics.
+- `tests/test_task_broadcasting.py` - retained-count and event regression.
+
+## Human edits
+
+None - generated output follows the requested first iterative item.

ai/prompt-io/opencode/20260812T012324Z_06c4af17_prompt_io.raw.md

@@ -0,0 +1,37 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-12T01:23:24Z
+git_ref: 06c4af17
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+After opening draft Tractor PR 490, the user requested an iterative pass over
+additional broadcast subsystem findings. The first item was to correct
+`BroadcastState.statistics()` queued counts and its deprecated Trio event
+truthiness check, then stop for a complete commit plan.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_broadcast.py`
+
+Changed `queued_len_by_task` from raw deque cursor indexes to retained,
+receivable counts. Caught-up `-1` reports zero, valid indexes report index plus
+one, and lagged cursors clamp to the current retained queue length. Replaced
+`trio.Event` truthiness with an explicit `is not None` branch.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added a deterministic statistics regression using actual sends and receives.
+It verifies caught-up and one-queued states, drives a root receiver beyond a
+three-slot retention window to prove clamping, and installs a real
+`trio.Event` while treating deprecations as errors.
+
+Verification output:
+
+```text
+...........                                                              [100%]
+11 passed in 5.76s
+```
+
+Python compilation and `git diff --check` passed. Initial adversarial review
+caught unclamped lagged cursors and an ineffective event test; both were
+fixed. Final review found no actionable issues.

ai/prompt-io/opencode/20260812T030608Z_1095e7f7_prompt_io.md

@@ -0,0 +1,33 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_0799212ebffe42arY96czXn89F
+timestamp: 2026-08-12T03:06:08Z
+git_ref: 1095e7f7
+scope: code
+substantive: true
+raw_file: 20260812T030608Z_1095e7f7_prompt_io.raw.md
+---
+
+## Prompt
+
+For Tractor PR 490 item two, make shared underlying receive failures wake and
+terminate every broadcast subscriber without losing retained values. Review,
+verify and return a complete commit plan before continuing.
+
+## Response summary
+
+Published ordinary receive failures as terminal broadcast state, introduced
+a public chained peer exception, kept control-flow exits transient while
+waking peers, documented the contract, and added deterministic regressions.
+
+## Files changed
+
+- `tractor/trionics/_broadcast.py` - terminal failure and peer wake protocol.
+- `tractor/trionics/__init__.py` - public peer exception export.
+- `docs/api/trionics.rst` - failure-delivery API contract.
+- `tests/test_task_broadcasting.py` - terminal and transient failure tests.
+
+## Human edits
+
+None - generated output follows the requested second iterative item.

ai/prompt-io/opencode/20260812T030608Z_1095e7f7_prompt_io.raw.md

@@ -0,0 +1,49 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-12T03:06:08Z
+git_ref: 1095e7f7
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+The user requested the second iterative refinement for Tractor PR 490: ensure
+non-EOC failures from a shared underlying broadcast receiver do not leave peer
+subscribers blocked forever, then stop for a complete commit plan.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_broadcast.py`
+
+Added shared terminal failure publication for ordinary `Exception` values.
+The receive owner gets the original exception; peers may drain retained
+values and then get a fresh `BroadcastReceiveError` chained from the original.
+Late subscribers observe the same terminal state without retrying the failed
+underlying receiver. Process-control and cancellation-like `BaseException`
+values wake peers but are re-raised without becoming durable channel state.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/__init__.py`
+
+Exported `BroadcastReceiveError` as the public peer-delivery exception.
+
+> `git diff HEAD~1..HEAD -- docs/api/trionics.rst`
+
+Documented `BroadcastReceiveError` and the owner-versus-peer delivery
+contract, including retained-value draining and late subscribers.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added deterministic bounded regressions. One scripts a successful receive
+followed by `RuntimeError`, proving the root drains retained data, all current
+and late receivers observe terminal failure, and the source is not retried.
+The second scripts a custom `BaseException`, proving peers wake and take over
+the next source receive without retaining control-flow state.
+
+Verification output:
+
+```text
+.............                                                            [100%]
+13 passed in 5.62s
+```
+
+Python compilation and `git diff --check` passed. Iterative adversarial review
+drove independent peer exception wrappers, ordinary-versus-control-flow
+classification, bounded test completion, public docs, and the final catch-all
+peer wake. Final review found no issues.

ai/prompt-io/opencode/20260812T150027Z_c2a6ccef_prompt_io.md

@@ -0,0 +1,33 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_0799212ebffe42arY96czXn89F
+timestamp: 2026-08-12T15:00:27Z
+git_ref: c2a6ccef
+scope: code
+substantive: true
+raw_file: 20260812T150027Z_c2a6ccef_prompt_io.raw.md
+---
+
+## Prompt
+
+For Tractor PR 490 item three, prevent subscriber closure from waking another
+receiver's shared event or stranding peers. Preserve close-safe source-read
+ownership handoff, then review, verify and return a complete commit plan.
+
+## Response summary
+
+Introduced receiver-local wait/source cancellation, owner-specific handoff,
+and close precedence over shielded source values/errors/EOC. Clarified that
+private scope cancellation means explicit close while outer task cancellation
+remains `trio.Cancelled` for both source-read and peer-wait scopes, with
+deterministic receiver-close regressions.
+
+## Files changed
+
+- `tractor/trionics/_broadcast.py` - receiver-local close and ownership scopes.
+- `tests/test_task_broadcasting.py` - peer close and owner handoff regressions.
+
+## Human edits
+
+None - generated output follows the requested third iterative item.

ai/prompt-io/opencode/20260812T150027Z_c2a6ccef_prompt_io.raw.md

@@ -0,0 +1,51 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-12T15:00:27Z
+git_ref: c2a6ccef
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+The user requested the third iterative refinement for Tractor PR 490: closing
+one broadcast subscriber must not set another receiver owner's shared event
+and create a runnable hot loop, then stop for a complete commit plan.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_broadcast.py`
+
+Added receiver-local wait cancellation and source-read ownership scopes.
+Closing a non-owner waiting behind another source reader cancels only that
+receiver's private wait and maps it to `ClosedResourceError`; the shared event
+remains untouched. Closing the active source owner cancels only its private
+source-read scope, wakes peers after cleanup, and lets one peer take ownership.
+
+Source outcomes are captured inside the owner scope and classified only after
+checking close/cancel state. A cancellation-shielding source therefore cannot
+publish a returned value, ordinary error, or EOC after its owner was closed.
+The private scope's `cancel_called` bit is asserted to imply receiver closure;
+outer task cancellation remains `trio.Cancelled` and is not translated into
+`ClosedResourceError`. Owner-key comments document that only the receiver
+identified by `recv_ready[0]` may cancel the shared source-read scope. The
+same explicit-close invariant is enforced symmetrically for private peer-wait
+scope cancellation.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added deterministic bounded regressions for both close positions. The
+non-owner test places two peers behind an active source read, closes one and
+proves only that peer exits while the shared event stays unset. The owner test
+closes a source owner whose receive shields cancellation and parameterizes a
+returned value, `RuntimeError`, and `EndOfChannel`; each discarded outcome
+hands the next source receive to the waiting root without terminal-state or
+EOC publication.
+
+Verification output:
+
+```text
+.................                                                        [100%]
+17 passed in 5.84s
+```
+
+Python compilation and `git diff --check` passed. Iterative adversarial review
+caught a waiting non-owner hang, cancellation-shielded source returns, and
+shielded source exceptions. All were fixed. Final review found no actionable
+issues.

ai/prompt-io/opencode/20260812T213117Z_51185487_prompt_io.md

@@ -0,0 +1,34 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_0799212ebffe42arY96czXn89F
+timestamp: 2026-08-12T21:31:17Z
+git_ref: 51185487
+scope: code
+substantive: true
+raw_file: 20260812T213117Z_51185487_prompt_io.raw.md
+---
+
+## Prompt
+
+For Tractor PR 490 item four, expose `raise_on_lag` through the public IPC and
+asyncio linked-channel subscription wrappers. Review, verify and return a
+complete commit plan before applying the API downstream in piker.
+
+## Response summary
+
+Added public per-subscription lag policy to both wrappers, preserved first-call
+root policy, documented the semantics, and covered forwarding plus real fan-out
+paths.
+
+## Files changed
+
+- `tractor/_streaming.py` - `MsgStream` lag policy forwarding.
+- `tractor/to_asyncio.py` - linked-channel lag policy forwarding.
+- `docs/guide/streaming.rst` - IPC fan-out policy docs.
+- `docs/guide/asyncio.rst` - linked-channel fan-out policy docs.
+- `tests/test_task_broadcasting.py` - wrapper policy regression.
+
+## Human edits
+
+None - generated output follows the requested fourth iterative item.

ai/prompt-io/opencode/20260812T213117Z_51185487_prompt_io.raw.md

@@ -0,0 +1,53 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-12T21:31:17Z
+git_ref: 51185487
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+The user requested the fourth iterative refinement for Tractor PR 490: expose
+subscriber lag policy through the public `MsgStream.subscribe()` and
+`LinkedTaskChannel.subscribe()` wrappers, then stop for a complete commit
+plan. This enables piker to replace private receiver mutation.
+
+> `git diff HEAD~1..HEAD -- tractor/_streaming.py`
+
+Added `raise_on_lag: bool = True` to `MsgStream.subscribe()`. The first call
+passes the policy to the irreversibly allocated root broadcaster and its
+child; later calls configure each child independently while retaining the
+root's first-call policy.
+
+> `git diff HEAD~1..HEAD -- tractor/to_asyncio.py`
+
+Added equivalent lag-policy forwarding to `LinkedTaskChannel.subscribe()`.
+
+> `git diff HEAD~1..HEAD -- docs/guide/streaming.rst`
+
+> `git diff HEAD~1..HEAD -- docs/guide/asyncio.rst`
+
+Documented strict versus warn/drop/resume behavior, independent child policy,
+and first-call root policy for both wrapper types.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added a parameterized wrapper-level regression using minimal receive-compatible
+handles. It verifies a first non-raising subscription configures root and
+child, then a later strict child does not mutate the sticky root policy.
+
+Verification output:
+
+```text
+...................                                                      [100%]
+19 passed in 5.71s
+
+....................                                                     [100%]
+20 passed in 6.88s
+
+.                                                                        [100%]
+1 passed in 0.86s
+```
+
+The second and third runs cover actual `MsgStream` and infected-asyncio
+`LinkedTaskChannel` fan-out respectively. Python compilation and
+`git diff --check` passed. Adversarial review found no actionable issues.

ai/prompt-io/opencode/20260813T181901Z_a2e0df4b_prompt_io.md

@@ -0,0 +1,36 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: unavailable
+timestamp: 2026-08-13T18:19:01Z
+git_ref: a2e0df4b
+scope: code
+substantive: true
+raw_file: 20260813T181901Z_a2e0df4b_prompt_io.raw.md
+---
+
+## Prompt
+
+Continue Tractor PR 490 after the paired piker EMS consumer commit. Clean
+cancelled-task diagnostics, define or reject zero-buffer broadcast behavior,
+review and verify the change, then stop at a complete commit plan.
+
+## Response summary
+
+Bound cancellation diagnostics to receiver progress, terminal state and
+resource lifetime; made EOC durable across peers; released wrapper-owned root
+broadcasters without breaking graceful EOC or subclass overrides; and rejected
+non-positive fan-out retention capacity.
+
+## Files changed
+
+- `tractor/trionics/_broadcast.py` - diagnostic lifecycle, durable EOC and
+  buffer validation.
+- `tractor/_streaming.py` - safe `MsgStream` root broadcaster cleanup.
+- `tractor/to_asyncio.py` - linked-channel root broadcaster cleanup.
+- `tests/test_task_broadcasting.py` - cancellation, EOC, wrapper and capacity
+  regressions.
+
+## Human edits
+
+None - generated output follows the requested fifth iterative item.

ai/prompt-io/opencode/20260813T181901Z_a2e0df4b_prompt_io.raw.md

@@ -0,0 +1,58 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-13T18:19:01Z
+git_ref: a2e0df4b
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+The user asked to continue after committing the paired piker EMS consumer
+fix. The next isolated Tractor PR 490 item was to clean cancelled-task
+diagnostics and define zero-buffer broadcast behavior, then review, test and
+stop at a complete commit plan.
+
+> `git diff HEAD~1..HEAD -- tractor/trionics/_broadcast.py`
+
+Made `BroadcastState.cancelled` transient: receiver progress and close clear
+that receiver's diagnostic, terminal EOC and shared receive failure clear all
+stale cancelled tasks, and durable EOC prevents peers from re-entering the
+closed source. `broadcast_receiver()` now rejects non-positive retention
+capacity before creating an unusable zero-length deque.
+
+> `git diff HEAD~1..HEAD -- tractor/_streaming.py`
+
+Made explicit `MsgStream.aclose()` release its internally allocated root
+broadcaster while preserving graceful receive-internal EOC teardown. Used a
+task-local marker so the public zero-argument `aclose()` signature and valid
+subclass overrides remain compatible.
+
+> `git diff HEAD~1..HEAD -- tractor/to_asyncio.py`
+
+Made `LinkedTaskChannel.aclose()` release its internally allocated root
+broadcaster before closing the underlying Trio receive channel.
+
+> `git diff HEAD~1..HEAD -- tests/test_task_broadcasting.py`
+
+Added synchronized regressions for transient child cancellation diagnostics,
+cross-receiver terminal cleanup, durable EOC peer wakeups, root broadcaster
+cleanup through both public wrappers, `MsgStream.aclose()` subclass
+compatibility, and zero-buffer rejection.
+
+Verification output:
+
+```text
+...........................                                              [100%]
+27 passed in 5.89s
+
+.                                                                        [100%]
+1 passed in 1.22s
+
+.                                                                        [100%]
+1 passed in 0.88s
+```
+
+The integration runs cover real `MsgStream` actor fan-out and infected-asyncio
+`LinkedTaskChannel` fan-out. Ruff, Python compilation and `git diff --check`
+passed. Repeated adversarial review found and resolved root close re-entrancy,
+cross-receiver terminal retention, durable-EOC and subclass-compatibility
+issues; final review reported no findings.

ai/prompt-io/opencode/20260817T231825Z_359fe75c_prompt_io.md

@@ -0,0 +1,39 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: pr475-review-fixes-20260817
+timestamp: 2026-08-17T23:18:25Z
+git_ref: 359fe75c
+scope: code
+substantive: true
+raw_file: 20260817T231825Z_359fe75c_prompt_io.raw.md
+---
+
+## Prompt
+
+Continue the `/code-review-changes` pass for PR #475 in its isolated
+worktree. Address the seven accepted manual-review findings in
+`tractor/ipc/_types.py` and `tractor/ipc/_uds.py`, preserve the existing
+Windows capability behavior, verify the result, and prepare the work for
+human-controlled commit and review-reply steps. Do not publish replies,
+stage, commit, or push without the required explicit authorization.
+
+## Response summary
+
+Restored project quote, docstring, multiline-expression, and
+`match/case` conventions while retaining the Windows-safe UDS guard.
+Removed unnecessary structural and comment churn, then verified the
+focused transport, discovery, and lazy-import paths plus the missing
+`AF_UNIX` behavior.
+
+## Files changed
+
+- `tractor/ipc/_types.py` - restore project style and guarded
+  socket-family dispatch.
+- `tractor/ipc/_uds.py` - format the UDS capability gate
+  consistently.
+
+## Human edits
+
+None - the generated patch remains uncommitted and awaits human
+review.

ai/prompt-io/opencode/20260817T231825Z_359fe75c_prompt_io.raw.md

@@ -0,0 +1,45 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-17T23:18:25Z
+git_ref: 359fe75c
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Applied the seven accepted manual-review fixes for PR #475 while
+preserving the Windows transport capability behavior.
+
+> `git diff HEAD~1..HEAD -- tractor/ipc/_types.py`
+
+The generated changes restore the project's single-quote docstring and
+string conventions, remove the unnecessary helper divider, simplify the
+transport-registry comments, and restore `match/case` socket-family
+dispatch. The UDS case retains a `HAS_UDS` guard that short-circuits
+before `socket.AF_UNIX` is evaluated on unsupported hosts. Nearby error
+messages are wrapped without changing their content.
+
+> `git diff HEAD~1..HEAD -- tractor/ipc/_uds.py`
+
+The generated change reformats the `HAS_UDS` conjunction according to
+the project's multiline boolean-expression convention and simplifies
+the adjacent capability comment.
+
+Verification:
+
+`/home/goodboy/repos/tractor/py313/bin/pytest -q tests/test_lazy_imports.py tests/discovery tests/ipc/test_server.py`
+
+Result: `66 passed, 2 xpassed in 60.62s`.
+
+`ruff check --no-cache --output-format=json tractor/ipc/_types.py tractor/ipc/_uds.py`
+
+Result: no findings.
+
+`git diff --check`
+
+Result: no whitespace errors.
+
+An explicit missing-`AF_UNIX` probe set `HAS_UDS = False`, removed the
+socket constant, and exercised an unsupported socket family. It raised
+the expected `NotImplementedError` instead of `AttributeError`.
+
+No review replies, commits, or pushes were published.

ai/prompt-io/opencode/20260818T021729Z_d9a6e2e9_prompt_io.md

@@ -0,0 +1,69 @@
+---
+model: gpt-5.6-sol
+service: opencode
+session: intercepted-claude-7b9c97c4-fff7-4ac4-97fb-35720453308e
+timestamp: 2026-08-18T02:17:29Z
+git_ref: d9a6e2e9
+scope: code
+substantive: true
+raw_file: 20260818T021729Z_d9a6e2e9_prompt_io.raw.md
+---
+
+## Prompt
+
+> Intercept Claude session
+> `7b9c97c4-fff7-4ac4-97fb-35720453308e`, pick up where it
+> stopped in its open worktree, finish the tunnelled-address change
+> and commit plan, and prepare any outstanding context for another
+> provider.
+
+The recovered final prompt specifically called out the `uds` versus
+`unix` boundary decision, absent concrete `Address.namespace`
+implementations, the corrected namespace test, and the passing
+focused and wider suites.
+
+During review, the human further required that tunnelled-maddr work
+delegate to `multiaddr`'s encapsulation APIs, challenged the premature
+transport-shaped listener hooks and directed their removal, corrected
+the long-term bearer provisioning model, and confirmed the intended
+split between discovery metadata and bindspace lifecycle code.
+
+## Response summary
+
+Recovered the transcript and matched it to `wkts/addr_unpacking`,
+audited the staged implementation, and completed the interrupted
+verification and commit-plan work. The audit removed premature
+transport-shaped listener hooks, widened the namespace identifier
+type, updated stale import documentation, and removed an
+invalid-escape warning from the maddr diagram. It also preserved the
+layer-C design where tractor provisions the kernel-owned tunnel
+bearer without treating it as a message transport.
+
+## Files changed
+
+- `tractor/discovery/_tunnel.py` - tunnel specs, address wrapper, and
+  peeling helpers.
+- `tractor/discovery/_addr.py` - wrapped-address recognition and
+  namespace typing.
+- `tractor/discovery/__init__.py` - public tunnel API exports.
+- `tests/discovery/test_tunnelled_addr.py` - delegation and boundary
+  regression coverage.
+- `ai/tpt-backends/03_wg_tunnel_bindspace.md` - distinguish
+  tractor-owned bindspace provisioning from kernel socket ownership.
+
+## Human edits
+
+Substantial human-directed editing occurred over several review turns:
+
+- required use of `multiaddr`'s `.encapsulate()`/`.decapsulate()`
+  family rather than a hand-rolled tunnel peeler
+- rejected the premature `start_listener()`/`close_listener()` hooks
+  and directed their removal from this foundational change
+- corrected the documentation so tractor retains ownership of future
+  bindspace provisioning while the kernel owns the bearer socket
+- reviewed and accepted the placement of declarative tunnel metadata
+  under `tractor.discovery`, with lifecycle code kept separate
+
+The final source lines were applied through the coding agents, but
+these design corrections and deletion decisions came from the human
+review and materially shaped the patch.

ai/prompt-io/opencode/20260818T021729Z_d9a6e2e9_prompt_io.raw.md

@@ -0,0 +1,66 @@
+---
+model: gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-18T02:17:29Z
+git_ref: d9a6e2e9
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+# Raw output - tunnelled-address handoff completion
+
+Recovered Claude Code session
+`7b9c97c4-fff7-4ac4-97fb-35720453308e` and continued its
+interrupted `wkts/addr_unpacking` changes.
+
+## Generated code
+
+> `git diff HEAD~1..HEAD -- tractor/discovery/_tunnel.py`
+
+Added frozen `WGTunnelSpec` and `TunnelledAddress` structs. The
+wrapper delegates transport identity, validity, bindspace, and wire
+serialization to its overlay while retaining tunnel metadata locally.
+Added pure helpers to peel nested wrappers and enumerate their tunnel
+specs. The module documents why wrappers must be peeled before
+`Endpoint` selects the overlay transport backend.
+
+> `git diff HEAD~1..HEAD -- tractor/discovery/_addr.py`
+
+Extended `is_wrapped_addr()` to recognize `TunnelledAddress` without
+registering tunnels as message transports, and widened the namespace
+identifier type to cover named network namespaces.
+
+> `git diff HEAD~1..HEAD -- tractor/discovery/__init__.py`
+
+Exported the tunnel address API from `tractor.discovery` and updated
+the eager-import documentation.
+
+> `git diff HEAD~1..HEAD -- tests/discovery/test_tunnelled_addr.py`
+
+Added focused coverage for delegation, serialization, rewrapping,
+namespace fallback, nested peeling order, and frozen structs.
+
+> `git diff HEAD~1..HEAD -- ai/tpt-backends/03_wg_tunnel_bindspace.md`
+
+Clarified that tractor owns the eventual bindspace lifecycle and may
+provision the WireGuard iface, routes, and kernel UDP listener through
+netlink/`pyroute2`. Kernel socket ownership does not make the bearer an
+application `MsgTransport` endpoint.
+
+## Verification
+
+Focused tests:
+
+```text
+9 passed in 0.03s
+```
+
+Discovery and IPC suites:
+
+```text
+67 passed, 2 xpassed
+```
+
+The audit removed premature module-level listener hooks. Runtime
+integration must peel the wrapper explicitly at bind and dial
+boundaries rather than make `._tunnel` impersonate a transport
+backend.

ai/prompt-io/opencode/20260818T031532Z_4151b956_prompt_io.md

@@ -0,0 +1,43 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: pr481-review-fixes-p1-20260818
+timestamp: 2026-08-18T03:15:32Z
+git_ref: 4151b956
+scope: code
+substantive: true
+raw_file: 20260818T031532Z_4151b956_prompt_io.raw.md
+---
+
+## Prompt
+
+Address the approved review findings on PR #481, but work
+iteratively: implement and verify one finding at a time, prepare a
+separate `/commit-plan` after each fix, and stop for the human commit
+before starting the next finding. Begin with the P1 per-child
+lifecycle issue. Also publish the already-approved review findings
+against the reviewed PR head before editing.
+
+## Response summary
+
+Published the approved non-approving review at head `4151b956`, then
+implemented only the P1 lifecycle fix. Owned one-shot actors now use
+a child-specific cancellation and process-reap handshake, including
+hard escalation for unacknowledged cancellation and deterministic
+bookkeeping removal before `to_actor.run()` returns.
+
+## Files changed
+
+- `tractor/runtime/_supervise.py` - coordinate child-specific cancel
+  and reap.
+- `tractor/spawn/_trio.py` - wait on the Trio child's reap request.
+- `tractor/spawn/_mp.py` - wait on the multiprocessing child's reap
+  request.
+- `tractor/spawn/_spawn.py` - publish monitor completion centrally.
+- `tractor/to_actor/_api.py` - await owned-child process reaping.
+- `tests/test_to_actor.py` - cover cleanup, escalation, and startup
+  ordering.
+
+## Human edits
+
+None - the generated P1 patch remains uncommitted for human review.

ai/prompt-io/opencode/20260818T031532Z_4151b956_prompt_io.raw.md

@@ -0,0 +1,74 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-18T03:15:32Z
+git_ref: 4151b956
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Implemented only the P1 lifecycle finding from the approved PR #481
+review, preserving the requested one-fix-at-a-time commit boundary.
+
+> `git diff HEAD~1..HEAD -- tractor/runtime/_supervise.py`
+
+Added per-child reap request/completion events to `ActorNursery`, a
+shielded child-specific cancel-and-reap operation, late-registration
+latching for nursery teardown, and cancellation escalation that waits
+for debugger release before using non-ignorable process termination.
+The nursery-wide cancellation path snapshots child records before
+checkpointing so concurrent one-shot cleanup cannot invalidate its
+iteration.
+
+> `git diff HEAD~1..HEAD -- tractor/spawn/_trio.py`
+
+Changed Trio child monitors to wait on their per-child reap requests.
+
+> `git diff HEAD~1..HEAD -- tractor/spawn/_mp.py`
+
+Changed multiprocessing child monitors to wait on their per-child reap
+requests.
+
+> `git diff HEAD~1..HEAD -- tractor/spawn/_spawn.py`
+
+Ensured every backend publishes child-reap completion after its process
+monitor exits.
+
+> `git diff HEAD~1..HEAD -- tractor/to_actor/_api.py`
+
+Changed owned one-shot cleanup to await child-specific process joining
+and bookkeeping removal instead of treating the cancel RPC as reaping.
+
+> `git diff HEAD~1..HEAD -- tests/test_to_actor.py`
+
+Added regressions for immediate caller-managed nursery cleanup, failed
+cancel acknowledgement escalation, and child registration after a
+latched nursery-wide teardown request.
+
+Verification:
+
+`pytest -q tests/test_to_actor.py tests/test_cancellation.py tests/test_spawning.py tests/discovery/test_multi_program.py`
+
+Result: `46 passed, 1 xfailed, 3 xpassed`.
+
+`pytest -q tests/test_to_actor.py --tpt-proto uds`
+
+Result: `13 passed`.
+
+`pytest -q tests/test_to_actor.py --spawn-backend mp_spawn --tpt-proto tcp`
+
+Result: `13 passed`.
+
+One broad verification run was mistakenly launched in parallel with
+the UDS and `mp_spawn` actor suites. It timed out
+`test_remote_error_from_caller_nursery`; the node passed immediately
+in isolation and the complete broad selection then passed serially.
+The failure was classified as concurrent test-session interference,
+not accepted as a passing boundary result.
+
+Ruff, Python compilation, and `git diff --check` passed for the changed
+boundary. Ruff's existing `_trio.py` F401 finding was reproduced at the
+unmodified PR head and excluded from attribution to this patch.
+
+No source files were staged, committed, pushed, or used for review
+replies. The previously approved top-level review was published before
+the fix at reviewed head `4151b956`.

ai/prompt-io/opencode/20260818T075031Z_dd02c7c0_prompt_io.md

@@ -0,0 +1,50 @@
+---
+model: gpt-5.6-sol
+service: opencode
+session: tractor-addr-unpacking-followup
+timestamp: 2026-08-18T07:50:31Z
+git_ref: dd02c7c0
+scope: code
+substantive: true
+raw_file: 20260818T075031Z_dd02c7c0_prompt_io.raw.md
+---
+
+## Prompt
+
+The human requested the complete tunnelled-maddr parsing/composition
+slice as an unattended batch, with every numbered requirement finished
+and atomic commit plans prepared at the end. Existing human decisions
+required native `multiaddr` encapsulation APIs, no hand-written peeler,
+and preservation of tractor's future bindspace lifecycle ownership.
+
+## Response summary
+
+Implemented strict WG key codecs and native single/nested tunnel maddr
+parsing and composition, integrated them into discovery APIs, migrated
+the multihost example off its duplicate parser, corrected package
+dependency metadata, and added focused and end-to-end parser
+regressions. Verified the complete tractor suite and built both package
+artifacts.
+
+## Files changed
+
+- `tractor/discovery/` - WG codecs, parser/composer, wrapper typing,
+  public exports, and discovery dispatch.
+- `tests/discovery/` - key, grammar, nesting, round-trip, and public
+  boundary regressions.
+- `examples/multihost/wg_lan/` - production parser migration and
+  updated usage documentation.
+- `pyproject.toml`, `uv.lock` - reproducible WG codec and multibase
+  dependencies for checkout and package installs.
+- `ai/tpt-backends/03_wg_tunnel_bindspace.md` - current layer-A state
+  and future bindspace ownership.
+
+## Human edits
+
+The human selected the five-step scope and batch execution model,
+required delegation to `multiaddr`'s encapsulation APIs, rejected
+transport-shaped listener placeholders in the prerequisite commit, and
+clarified that tractor will eventually provision the kernel-owned
+bearer through its bindspace layer. The agent implemented and tested
+those decisions; no direct manual source edits were observed during
+this batch.

ai/prompt-io/opencode/20260818T075031Z_dd02c7c0_prompt_io.raw.md

@@ -0,0 +1,61 @@
+---
+model: gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-18T07:50:31Z
+git_ref: dd02c7c0
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+# Raw output - native WireGuard maddr integration
+
+The human requested completion of the five-step tunnelled-maddr slice:
+port the proven WireGuard parser, delegate to `py-multiaddr`'s native
+tunnel APIs, integrate public parse and composition entry points, add
+regressions, and return atomic commit plans after completing the batch.
+
+## Generated code
+
+> `git diff HEAD~1..HEAD -- tractor/discovery/_tunnel.py tractor/discovery/_multiaddr.py tractor/discovery/_addr.py tractor/discovery/__init__.py`
+
+Added strict WireGuard standard-base64/multibase key codecs and native
+WG maddr parsing/composition. Nested stacks peel the last `/wg/`
+repeatedly with `.decapsulate_code()`, isolate segments through
+`.split()`/`.join()`, and compose recursively with `.encapsulate()`.
+Public discovery parsing, wrapping, endpoint-table parsing, and maddr
+composition now preserve `TunnelledAddress` metadata locally.
+
+> `git diff HEAD~1..HEAD -- pyproject.toml uv.lock`
+
+Declared `py-multibase` directly and pinned the unreleased
+py-multiaddr#108 WG codec revision in distribution metadata. Enabled
+Hatch direct-reference metadata so editable, wheel, and sdist builds
+use the same dependency contract.
+
+> `git diff HEAD~1..HEAD -- tests/discovery/test_multiaddr.py tests/discovery/test_tunnelled_addr.py`
+
+Added key-codec, single/nested WG round-trip, malformed key, invalid
+bearer/overlay, missing codec, missing bearer, `wrap_address()`, and
+`parse_endpoints()` coverage while retaining plain TCP/UDS regressions.
+
+> `git diff HEAD~1..HEAD -- examples/multihost/wg_lan ai/tpt-backends/03_wg_tunnel_bindspace.md`
+
+Removed the duplicate example parser, migrated callers to production
+`TunnelledAddress`, retained only explicit `wg(8)` verification, and
+updated dependency and future bindspace ownership documentation.
+
+## Verification
+
+- focused discovery files: `45 passed`
+- discovery and IPC suites: `81 passed, 2 xpassed`
+- full suite: `443 passed, 9 skipped, 7 xfailed, 4 xpassed`
+- collection: `462 tests`
+- Ruff: clean
+- sdist and wheel builds: successful
+
+## Review corrections
+
+A review caught that `[tool.uv.sources]` alone would not constrain
+built-package consumers to a WG-capable py-multiaddr revision. The
+dependency was moved into PEP 621 metadata and the parser now converts
+missing-codec failures into an actionable tractor error without
+misclassifying Unix paths containing a `wg` directory.

ai/prompt-io/opencode/20260818T193001Z_bf06b4f8_prompt_io.md

@@ -0,0 +1,36 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_3e4c90d3eafeqHEtRYSIHgHhpA
+timestamp: 2026-08-18T19:30:01Z
+git_ref: bf06b4f8
+scope: code
+substantive: true
+raw_file: 20260818T193001Z_bf06b4f8_prompt_io.raw.md
+---
+
+## Prompt
+
+Fix cancellation during an indeterminate transport write without
+allowing a second RPC msg to corrupt the stream. Keep the transport
+fix separate from the higher-level `to_actor` remediation and explain
+the length-prefixed protocol rather than using unexplained "framed
+send" terminology.
+
+## Response summary
+
+Close a msgpack transport stream when cancellation interrupts its
+length-prefixed `send_all()` operation. The stream can no longer be
+safely reused because Trio cannot report how many bytes were written.
+
+## Files changed
+
+- `tractor/ipc/_transport.py` - close an interrupted send stream.
+- `tests/ipc/test_each_tpt.py` - cover cancellation during the write.
+
+## Human edits
+
+The human required this transport edge-case fix to land as its own
+behavioral commit with a detailed message. During staged review, the
+human also rejected the unexplained "framed send" wording and asked
+for terminology tied directly to the actual transport operation.

ai/prompt-io/opencode/20260818T193001Z_bf06b4f8_prompt_io.raw.md

@@ -0,0 +1,19 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-18T19:30:01Z
+git_ref: bf06b4f8
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Prospective review found that cancellation can interrupt
+`MsgpackTransport.send()` after `send_all()` writes only part of its
+length-prefixed msg. Sending a cancellation request afterward can
+append another msg to the indeterminate stream and desynchronize the
+peer decoder.
+
+> `git diff HEAD~1..HEAD -- tractor/ipc/_transport.py tests/ipc/test_each_tpt.py`
+
+Close the stream under a cancellation shield when `send_all()` is
+cancelled. Cover the behavior with a fake stream that checkpoints
+inside the write and records forced closure.

ai/prompt-io/opencode/20260818T193002Z_bf06b4f8_prompt_io.md

@@ -0,0 +1,37 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_3e4c90d3eafeqHEtRYSIHgHhpA
+timestamp: 2026-08-18T19:30:02Z
+git_ref: bf06b4f8
+scope: code
+substantive: true
+raw_file: 20260818T193002Z_bf06b4f8_prompt_io.raw.md
+---
+
+## Prompt
+
+Distill repeated `Actor._contexts.pop()` machinery into a wrapper like
+the RPC-task registration helper so future teardown sites do not keep
+reconstructing the context-registry key independently. Preserve the
+existing lifecycle-specific cleanup behavior.
+
+## Response summary
+
+Add idempotent `Actor._drop_context()` registry removal keyed from the
+context's own channel and CID. Use it for caller context teardown and
+the strict callee-side RPC deregistration path.
+
+## Files changed
+
+- `tractor/runtime/_runtime.py` - own context-registry removal.
+- `tractor/runtime/_rpc.py` - use the helper for callee teardown.
+- `tractor/_context.py` - use the helper after caller teardown.
+
+## Human edits
+
+The human identified the repeated registry-pop code and requested a
+central primitive analogous to `_register_rpc_task()`. The agent first
+suggested an async helper that also closed receive channels; the final
+design was narrowed to registry removal only so each lifecycle owner
+retains its existing closure, debugger, shielding, and error policy.

ai/prompt-io/opencode/20260818T193002Z_bf06b4f8_prompt_io.raw.md

@@ -0,0 +1,18 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-18T19:30:02Z
+git_ref: bf06b4f8
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Repeated teardown sites reconstruct the `Actor._contexts` registry
+key from a portal channel and context ID before popping it. Add an
+idempotent actor-owned helper deriving the key from the context itself,
+then route caller and callee context teardown through that helper.
+
+> `git diff HEAD~1..HEAD -- tractor/runtime/_runtime.py tractor/runtime/_rpc.py tractor/_context.py`
+
+Keep receive-channel closure and cancellation shielding in each
+lifecycle owner so the helper centralizes registry machinery without
+changing their teardown ordering.

ai/prompt-io/opencode/20260818T193003Z_bf06b4f8_prompt_io.md

@@ -0,0 +1,38 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_3e4c90d3eafeqHEtRYSIHgHhpA
+timestamp: 2026-08-18T19:30:03Z
+git_ref: bf06b4f8
+scope: code
+substantive: true
+raw_file: 20260818T193003Z_bf06b4f8_prompt_io.raw.md
+---
+
+## Prompt
+
+Cancel a remote task when its caller is cancelled after `Start`
+publication but before startup acknowledgement. Keep cancellation
+bounded, prevent its private `_cancel_task` RPC from recursively
+cancelling itself and preserve public target kwargs unchanged.
+
+## Response summary
+
+Add private portal startup policy, use it for non-recursive context
+cancellation and clean caller-side startup state under a shield.
+
+## Files changed
+
+- `tractor/runtime/_portal.py` - separate private startup policy.
+- `tractor/_context.py` - disable recursion for cancellation RPCs.
+- `tractor/runtime/_runtime.py` - clean cancelled task startup.
+- `tests/test_context_stream_semantics.py` - control cancellation
+  between `Start` publication and acknowledgement.
+
+## Human edits
+
+The human required this cancellation behavior to remain a distinct
+commit from general startup failures and from the public `to_actor`
+API. The human also requested that its runtime comment describe the
+actual length-prefixed transport guarantee and concrete `_cancel_task`
+operation rather than referring to an unnamed wrapper.

ai/prompt-io/opencode/20260818T193003Z_bf06b4f8_prompt_io.raw.md

@@ -0,0 +1,20 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-18T19:30:03Z
+git_ref: bf06b4f8
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Cancellation while `Actor.start_remote_task()` waits for `StartAck`
+can strand its caller-side context and leave the remote task running.
+Make one bounded cleanup request, remove local startup state and close
+its receive channel.
+
+> `git diff HEAD~1..HEAD -- tractor/runtime/_runtime.py tractor/runtime/_portal.py tractor/_context.py tests/test_context_stream_semantics.py`
+
+Separate private startup-cancellation policy from public target kwargs
+using `Portal._run_from_ns()`. Have `Context.cancel()` disable recursive
+startup cancellation for its own `_cancel_task` RPC. Exercise
+cancellation after `Start` publication and prove the caller-owned actor
+remains reusable without leaked contexts.

ai/prompt-io/opencode/20260818T193004Z_bf06b4f8_prompt_io.md

@@ -0,0 +1,36 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_3e4c90d3eafeqHEtRYSIHgHhpA
+timestamp: 2026-08-18T19:30:04Z
+git_ref: bf06b4f8
+scope: code
+substantive: true
+raw_file: 20260818T193004Z_bf06b4f8_prompt_io.raw.md
+---
+
+## Prompt
+
+Release caller-side context state for every remote-task startup failure,
+not only local cancellation. Preserve the remote error, avoid unsafe
+follow-up sends and prove pre-publication serialization failures leave
+a reused portal healthy.
+
+## Response summary
+
+Extend remote-task startup cleanup across send, acknowledgement and
+validation errors. Track completed publication, perform only safe
+best-effort cancellation and deterministically remove local state.
+
+## Files changed
+
+- `tractor/runtime/_runtime.py` - clean every startup failure path.
+- `tests/test_context_stream_semantics.py` - cover authorization and
+  serialization failures before context entry.
+
+## Human edits
+
+The human accepted the discovered edge-case fixes but required general
+startup cleanup to land separately from cancellation cleanup, transport
+integrity and the public API. This boundary preserves that behavioral
+distinction and its dedicated commit-message rationale.

ai/prompt-io/opencode/20260818T193004Z_bf06b4f8_prompt_io.raw.md

@@ -0,0 +1,19 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-18T19:30:04Z
+git_ref: bf06b4f8
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+`Actor.start_remote_task()` inserts a context before sending `Start`,
+but startup errors other than cancellation escape without removing or
+closing that caller state. Serialization errors, acknowledgement
+timeouts, malformed acknowledgements and remote authorization errors
+can therefore leak context-registry entries.
+
+> `git diff HEAD~1..HEAD -- tractor/runtime/_runtime.py tests/test_context_stream_semantics.py`
+
+Cover the complete send, acknowledgement and validation phase with
+exceptional cleanup. Attempt remote cancellation only when publication
+is known complete or protocol-safe, and always release local state.

ai/prompt-io/opencode/20260818T193005Z_bf06b4f8_prompt_io.md

@@ -0,0 +1,52 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_3e4c90d3eafeqHEtRYSIHgHhpA
+timestamp: 2026-08-18T19:30:05Z
+git_ref: bf06b4f8
+scope: code
+substantive: true
+raw_file: 20260818T193005Z_bf06b4f8_prompt_io.raw.md
+---
+
+## Prompt
+
+Replace abandoned `Portal.run()` one-shots with a static linked-context
+endpoint. Follow Trio positional-call semantics, use partials for target
+keywords, preserve Python 3.14 Placeholder behavior, keep target lookup
+behind the RPC allowlist and support private, nursery and portal
+placement.
+
+## Response summary
+
+Use `Portal.open_context()` and `Context.wait_for_result()` for one-shot
+tasks. Normalize every partial layer, validate signatures locally and
+send target namespace/function components separately to the authorized
+remote resolver. Retain the client-side function in its `NamespacePath`
+so `to_tuple()` does not re-import it. Owned actors enable the declaring
+`_api.__name__` directly; caller-owned portals opt in through the public
+`to_actor.MODULE` alias.
+
+## Files changed
+
+- `tractor/to_actor/_api.py` - implement linked one-shot calls.
+- `tractor/to_actor/__init__.py` - export `MODULE`.
+- `tractor/msg/ptr.py` - retain refs created by `from_ref()`.
+- `tests/test_to_actor.py` - cover the public API and authorization.
+- `examples/parallelism/to_actor_one_shots.py` - use positional inputs.
+
+## Human edits
+
+The human rejected nested target-kwargs configuration and selected
+Trio-style positional inputs plus `functools.partial()`. During staged
+review the human required a Python 3.14 compatibility comment rather
+than removing Placeholder support, requested separate namespace and
+function inputs, preserved `_get_rpc_func(ns: str, funcname: str)`
+authorization, renamed `RPC_MODULE` to `MODULE`, rejected global module
+exposure and deferred speculative nursery/module-list helpers to the
+`open_taskman()` design line. The human also required this public API
+to land only after its lower-level safety dependencies. In final staged
+review, the human required `_invoke_from_portal()` to use
+`NamespacePath.to_tuple()` with the already-held function ref and
+required internal actor setup to use `_api.__name__` directly, keeping
+`to_actor.MODULE` solely as the public importer-facing alias.

ai/prompt-io/opencode/20260818T193005Z_bf06b4f8_prompt_io.raw.md

@@ -0,0 +1,24 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-18T19:30:05Z
+git_ref: bf06b4f8
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Implement `to_actor.run()` with Trio-style positional target arguments,
+`functools.partial` keyword and Python 3.14 Placeholder binding, and a
+static context endpoint that links remote results, errors and caller
+cancellation.
+
+> `git diff HEAD~1..HEAD -- tractor/to_actor/_api.py tractor/to_actor/__init__.py`
+
+Resolve target functions through `Actor._get_rpc_func()` so module
+authorization remains authoritative. Automatically expose the helper
+module for actors owned by `to_actor.run()` and document explicit
+exposure for a caller-owned portal.
+
+> `git diff HEAD~1..HEAD -- tests/test_to_actor.py examples/parallelism/to_actor_one_shots.py`
+
+Cover placement modes, argument binding, nested partials, caller-linked
+cancellation, remote errors and module authorization.

ai/prompt-io/opencode/20260819T020757Z_b38efed7_prompt_io.md

@@ -0,0 +1,34 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: ses_3e4c90d3eafeqHEtRYSIHgHhpA
+timestamp: 2026-08-19T02:07:57Z
+git_ref: b38efed7
+scope: code
+substantive: true
+raw_file: 20260819T020757Z_b38efed7_prompt_io.raw.md
+---
+
+## Prompt
+
+Resolve the remaining P3 review finding before landing PR #481:
+`runtime_kwargs={}` must not be silently accepted alongside either
+`an=` or `portal=` merely because the dict is falsey. Keep this as its
+own final review-remediation commit.
+
+## Response summary
+
+Treat any non-`None` `runtime_kwargs` value as provided when validating
+placement. Cover both placement APIs with empty and configured dicts,
+proving the error is raised locally before actor startup.
+
+## Files changed
+
+- `tractor/to_actor/_api.py` - validate option presence explicitly.
+- `tests/test_to_actor.py` - cover four invalid option combinations.
+
+## Human edits
+
+No direct line edits. The human accepted the P3 finding, required it to
+remain separate from the five P2 behavioral commits and prioritized it
+before the final PR #484 integration rebase and PR #481 landing steps.

ai/prompt-io/opencode/20260819T020757Z_b38efed7_prompt_io.raw.md

@@ -0,0 +1,25 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-19T02:07:57Z
+git_ref: b38efed7
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Fix the final PR #481 review finding: `runtime_kwargs` is mutually
+exclusive with both caller placement options whenever it is provided,
+including an empty dict.
+
+> `git diff HEAD~1..HEAD -- tractor/to_actor/_api.py tests/test_to_actor.py`
+
+Use an explicit `is not None` check rather than dict truthiness. Expand
+the validation regression across `an=` and `portal=`, each with empty
+and configured runtime kwargs, so every invalid combination fails
+before actor runtime startup.
+
+Verification:
+
+- Trio/TCP: `23 passed`
+- Trio/UDS: `23 passed`
+- `mp_spawn`/TCP: `23 passed`
+- Ruff and `git diff --check`: clean

ai/prompt-io/opencode/20260819T184640Z_481ba003_prompt_io.md

@@ -0,0 +1,56 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: 76c5d31c-5a2f-4503-9b16-410ee7f4fab3
+timestamp: 2026-08-19T18:46:40Z
+git_ref: 481ba003
+scope: code
+substantive: true
+raw_file: 20260819T184640Z_481ba003_prompt_io.raw.md
+---
+
+## Prompt
+
+Rebase PR #484 onto final PR #481, migrate every affected one-shot call
+to the new positional target API and continue through downstream tests,
+examples and documentation review.
+
+## Response summary
+
+Converted stale target keyword calls to target partials so previously
+named inputs remain explicit while placement/runtime controls stay
+direct. Updated error expectations for local signature validation and
+linked remote error propagation, then corrected docs which still
+described the removed one-shot implementation. Linked spawning and
+context lifecycle prose to the corresponding API methods and detailed
+context guide.
+
+## Files changed
+
+- `docs/api/core.rst` - describe linked one-shot context execution.
+- `docs/guide/rpc.rst` - update placement and target call semantics.
+- `docs/guide/spawning.rst` - document positional target inputs.
+- `examples/debugging/multi_nested_subactors_error_up_through_nurseries.py` - migrate nested actor target inputs.
+- `examples/debugging/root_cancelled_but_child_is_in_tty_lock.py` - preserve named recursive target inputs with partials.
+- `tests/test_advanced_streaming.py` - migrate streaming target inputs.
+- `tests/test_cancellation.py` - migrate calls and tighten errors.
+- `tests/test_infected_asyncio.py` - bind asyncio target options.
+- `tests/test_rpc.py` - migrate RPC target argument binding.
+- `tests/test_runtime.py` - preserve named runtime target inputs.
+- `tests/test_spawning.py` - preserve named spawning target inputs.
+
+## Human edits
+
+The human selected the stack order and final PR #481 base, asked the
+agent to continue after each diagnostic step and required a complete
+commit plan after independently force-pushing the rebased history.
+After reviewing the migration, the human required every formerly named
+target input to remain visibly named through `functools.partial()`
+rather than becoming positional. These were human-directed agent edits;
+the human also required plain `start_actor()` and `open_context()`
+references in the spawning and RPC guides to link to their API methods
+and the detailed context guide, then clarified that `to_actor.run()`
+already uses the full context API while `Portal.run()` should share
+linked lifecycle machinery without necessarily delegating through
+`Portal.open_context()` or adding a `Started` message. The human made
+no direct source-line edits.

ai/prompt-io/opencode/20260819T184640Z_481ba003_prompt_io.raw.md

@@ -0,0 +1,30 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-19T18:46:40Z
+git_ref: 481ba003
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Migrate PR #484's downstream one-shot calls to PR #481's final
+`tractor.to_actor.run()` contract after the stack rebase.
+
+> `git diff HEAD~1..HEAD -- docs examples tests`
+
+Pass target arguments positionally and bind target keyword-only inputs
+with `functools.partial()`. Keep placement and runtime controls as
+direct `to_actor.run()` keywords. Update the invalid-target-argument
+test to expect local signature binding before actor startup and require
+direct `RemoteActorError` propagation from linked one-shots.
+
+Update API and guide prose to describe positional target inputs,
+linked `Portal.open_context()` execution and per-child reaping instead
+of the removed `Portal.run()` and target-`**kwargs` conventions.
+
+Verification:
+
+- core and migrated runtime batches: `97 passed`
+- discovery and related lifecycle batch: `33 passed, 1 skipped`
+- changed executable examples: `9 passed`
+- mapped debugger cases: `12 passed, 6 skipped`
+- Ruff, compilation and `git diff --check`: clean

ai/prompt-io/opencode/20260819T213144Z_f81fc5e5_prompt_io.md

@@ -0,0 +1,37 @@
+---
+model: gpt-5.6-sol
+service: opencode
+session: tractor-addr-unpacking-followup
+timestamp: 2026-08-19T21:31:44Z
+git_ref: f81fc5e5
+scope: code
+substantive: true
+raw_file: 20260819T213144Z_f81fc5e5_prompt_io.raw.md
+---
+
+## Prompt
+
+The human requested runtime boundary integration with the established
+per-step implementation and commit-plan workflow.
+
+## Response summary
+
+Kept `TunnelledAddress` available to callers while peeling it at the
+last outbound boundary before transport lookup and dialing. Added a
+regression which captures both transport arguments and confirms plain
+TCP behavior is unchanged.
+
+## Files changed
+
+- `tractor/ipc/_chan.py` - peel tunnel annotations before outbound
+  transport dispatch and connection.
+- `tests/ipc/test_channel_tunnel_boundary.py` - verify plain and
+  tunnelled channel inputs deliver only TCP overlays.
+
+## Human edits
+
+The human chose the runtime-boundary slice, required the existing
+per-step commit-plan flow, and previously established that wrappers
+must retain bindspace metadata without impersonating transports. The
+agent implemented those constraints; no direct manual source edits were
+observed during this step.

ai/prompt-io/opencode/20260819T213144Z_f81fc5e5_prompt_io.raw.md

@@ -0,0 +1,25 @@
+---
+model: gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-19T21:31:44Z
+git_ref: f81fc5e5
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+# Raw output - outbound tunnel boundary
+
+The human requested the next tunnelled-address slice using the same
+per-step commit-plan flow. Existing design decisions require retaining
+tunnel metadata until the narrow IPC transport boundary and never
+teaching exact-type transport tables about tunnel wrappers.
+
+> `git diff HEAD~1..HEAD -- tractor/ipc/_chan.py tests/ipc/test_channel_tunnel_boundary.py`
+
+Extended channel address inputs to accept tunnel declarations, then
+called `strip_tunnels()` immediately before exact-type transport lookup
+and `connect_to()`. Added plain/tunnel parameterized coverage proving
+both operations receive the identical TCP overlay while the original
+wrapper retains its tunnel spec.
+
+Verification included focused IPC tests, Ruff, discovery/IPC suites,
+and the full tractor suite.

ai/prompt-io/opencode/20260819T213145Z_f81fc5e5_prompt_io.md

@@ -0,0 +1,40 @@
+---
+model: gpt-5.6-sol
+service: opencode
+session: tractor-addr-unpacking-followup
+timestamp: 2026-08-19T21:31:45Z
+git_ref: f81fc5e5
+scope: code
+substantive: true
+raw_file: 20260819T213145Z_f81fc5e5_prompt_io.raw.md
+---
+
+## Prompt
+
+The human requested completion of inbound runtime peeling using the
+same per-step implementation and commit-plan workflow.
+
+## Response summary
+
+Preserved tunnel declarations through listener configuration, peeled
+them immediately before `Endpoint` construction, and used the overlay
+for backend-specific random listener allocation after registry
+discovery. Added a real listener regression for the reflection and
+exact-type boundary.
+
+## Files changed
+
+- `tractor/ipc/_server.py` - accept wrapper declarations and peel at
+  `Endpoint` construction.
+- `tractor/_root.py` - allocate random transport addresses from the
+  contacted registry's overlay.
+- `tests/ipc/test_server_tunnel_boundary.py` - verify a real listener
+  stores only TCP while preserving the source annotation.
+
+## Human edits
+
+The human selected the runtime-boundary work and previously corrected
+the architecture so tractor retains future bindspace provisioning
+ownership while `Endpoint` sees only application transports. The agent
+implemented and tested that direction; no direct manual source edits
+were observed during this step.

ai/prompt-io/opencode/20260819T213145Z_f81fc5e5_prompt_io.raw.md

@@ -0,0 +1,25 @@
+---
+model: gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-19T21:31:45Z
+git_ref: f81fc5e5
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+# Raw output - inbound tunnel boundary
+
+The human requested runtime boundary integration while preserving the
+future tractor-owned bindspace lifecycle.
+
+> `git diff HEAD~1..HEAD -- tractor/ipc/_server.py tractor/_root.py tests/ipc/test_server_tunnel_boundary.py`
+
+Broadened listener declarations to carry tunnel wrappers until
+`_serve_ipc_eps()` and peeled immediately before `Endpoint`
+construction. Also peeled a contacted tunnelled registry before
+backend-specific random listener allocation. Added a real TCP listener
+regression proving `Endpoint` stores only the resolved overlay while
+the original declaration retains bindspace metadata.
+
+Verification included `465` collected tests, `84` passing
+discovery/IPC tests with two xpasses, Ruff, and the full suite with
+`447` passes.

ai/prompt-io/opencode/20260819T234823Z_557065d8_prompt_io.md

@@ -0,0 +1,37 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: 76c5d31c-5a2f-4503-9b16-410ee7f4fab3
+timestamp: 2026-08-19T23:48:23Z
+git_ref: 557065d8
+scope: tests
+substantive: true
+raw_file: 20260819T234823Z_557065d8_prompt_io.raw.md
+---
+
+## Prompt
+
+Investigate PR #481's red CI run, explain the missing T-800 and
+debugger-output failures, and proceed with fixes in the PR #481
+worktree.
+
+## Response summary
+
+Updated stale teardown assertions to match #481's direct hard-reap
+path and observable process-lifetime invariants. Made nested debugger
+checks consume the complete pexpect transcript rather than only the
+last prompt latch.
+
+## Files changed
+
+- `tests/devx/test_debugger.py` - assert EOF/dead-process teardown and
+  accumulate nested debugger output across prompt boundaries.
+- `tests/devx/test_tooling.py` - assert cancel-timeout hard-reap
+  escalation instead of the bypassed T-800 backend marker.
+
+## Human edits
+
+The human reported the still-red PR #481 CI, supplied a failing job URL,
+required work in `/wkts/pr481_review_fixes` and directed the agent to
+continue immediately. No direct source-line edits were made by the
+human.

ai/prompt-io/opencode/20260819T234823Z_557065d8_prompt_io.raw.md

@@ -0,0 +1,26 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+timestamp: 2026-08-19T23:48:23Z
+git_ref: 557065d8
+diff_cmd: git diff HEAD~1..HEAD
+---
+
+Diagnose and fix the stale debugger and reaper assertions failing PR
+#481's Unix CI jobs.
+
+> `git diff HEAD~1..HEAD -- tests/devx/test_debugger.py tests/devx/test_tooling.py`
+
+Replace the old T-800 backend-log requirement with the new bounded
+cancel-ack escalation evidence. Prove debugger teardown with EOF and a
+dead child process instead of requiring optional `KeyboardInterrupt`
+text. Accumulate all pexpect prompt chunks for nested error propagation
+so expected tracebacks are not lost when `child.before` advances.
+
+Verification:
+
+- exact failed debugger/reaper nodes: `4 passed`
+- debugger/tooling TCP: `39 passed, 6 skipped`
+- debugger/tooling UDS: `39 passed, 6 skipped`
+- full TCP suite: `478 passed, 9 skipped, 7 xfailed, 3 xpassed`
+- full UDS rerun: `476 passed, 11 skipped, 8 xfailed, 2 xpassed`

ai/prompt-io/opencode/20260819T234824Z_557065d8_prompt_io.md

@@ -0,0 +1,43 @@
+---
+model: openai/gpt-5.6-sol
+service: opencode
+session: 76c5d31c-5a2f-4503-9b16-410ee7f4fab3
+timestamp: 2026-08-19T23:48:24Z
+git_ref: 557065d8
+scope: code
+substantive: true
+raw_file: 20260819T234824Z_557065d8_prompt_io.raw.md
+---
+
+## Prompt
+
+Investigate and fix PR #481's macOS TCP clustering and stream-overrun
+failures without sacrificing IPC frame integrity or structured
+concurrency.
+
+## Response summary
+
+Changed cancellation during `send_all()` from actor-wide stream closure
+to shielded complete-frame publication followed by immediate pending
+cancellation. Prevented failed overrun error shipment from promoting a
+secondary transport closure over the context-local primary condition.
+
+## Files changed
+
+- `tractor/ipc/_transport.py` - complete in-flight frames before
+  delivering sender cancellation.
+- `tractor/_context.py` - absorb transport closure while reporting an
+  overrun on an already-closing channel.
+- `tests/ipc/test_each_tpt.py` - prove complete framing, cancellation
+  delivery and channel reuse.
+- `tests/test_context_stream_semantics.py` - prove overrun reporting
+  tolerates a closed transport.
+
+## Human edits
+
+The human reported PR #481's red CI, asked for diagnosis and directed
+the agent to proceed in the dedicated PR #481 worktree. During final
+review, the human required preservation of the original far-end
+cancellation rationale and fuller documentation of frame shielding,
+shared-channel ownership and cancellation-delay tradeoffs. These were
+human-directed agent edits; the human made no direct source-line edits.