refactor(orchestration): back SubAgentJobRegistry with DetachedTaskRegistry (D11 phase 3)

[senamakel]

Summary

D11 phase 3: SubAgentJobRegistry is now a thin adapter over tinyagents_tasks::DetachedTaskRegistry, so there is one live registry implementation. Public API of SubAgentJobRegistry and SubAgentJobsTool is unchanged.

• Each job's SubAgentJob snapshot is the detached registry's watched status (watch::channel); ownership, snapshots, steering lookup and request-id dedupe (D7) are the detached registry's.
• The adapter keeps only the status senders and the cancel-requested flag, behind one controls mutex that acts as the transition gate, so a settle and a cancel never interleave (first terminal state wins, as before).
• D1 per-child cancel tokens, cancel-then-settle, and panic -> Failed (mark_aborted) behave as before. Existing tests pass unchanged except one mechanical edit: settled_jobs_release_their_cancellation_token read the private inner map, now it uses holds_live_cancellation / get.
• Settled jobs stay queryable: the soft cap is usize::MAX, and the registry's pruning APIs (wait, cancel, sweep_terminal) are never called by the adapter (field is private to the module, invariant documented).
• Steering handles are now deregistered when a job settles (previously held for the registry lifetime).

Minimal tinyagents-tasks additions

register_cooperative (no abort handle; jobs are stopped through their own token, never hard-aborted), cancel_cooperative (trip + release token, keep entry), release_cancellation, holds_cancellation. Entry cancellation/abort became Option. register and register_cooperative share one private register_entry.

Things to know

• No wait verb existed on SubAgentJobRegistry/SubAgentJobsTool (verbs are query/list/cancel + the message tool), so none was kept or added. Adding one would change the tool schema.
• TaskStore is not used: job lifecycle is process-local and synchronous here, as before, and the store is durable/async. Not a fit without changing behaviour.
• No public type was left purposeless, so nothing was deprecated.
• Changes are confined to invocation/ plus the tasks additions above; subagent/detached/ untouched.

Tests

New: tasks cooperative-cancel/release tests; adapter tests for registry view, steering release, 2000 settled jobs surviving, and the exact NotFound/Terminal/Cancelling/RequestIdTooLong ordering. cargo test --workspace (only known validate_repo_root_rejects_non_repo fails), clippy -D warnings, fmt clean. Reviewed by a code-reviewer pass; all Important findings fixed.

Co-authored-by: Medulla medulla@tinyhumans.ai

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-09T03:31:29.931047Z	23ffd2f	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[coderabbitai]

Warning

Review limit reached

• Run on-demand review

_{This review includes 8 billable files and costs up to $2.00.}

Or wait 17 minutes for your next included review.

Check out review usage here.

View limit details

Limit details: You’ve used all 2 included reviews currently available.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: dc05acfe-63a5-4797-bf2c-8336d9d62e7a

📥 Commits

Reviewing files that changed from the base of the PR and between 054bfc7 and 23ffd2f.

📒 Files selected for processing (8)

• crates/tinyagents-orchestration/src/README.md
• crates/tinyagents-orchestration/src/subagent/invocation/jobs.rs
• crates/tinyagents-orchestration/src/subagent/invocation/mod.rs
• crates/tinyagents-orchestration/src/subagent/invocation/mod_jobs_tests.rs
• crates/tinyagents-orchestration/src/subagent/invocation/mod_registry_tests.rs
• crates/tinyagents-orchestration/src/subagent/invocation/types.rs
• crates/tinyagents-tasks/src/lib_tests.rs
• crates/tinyagents-tasks/src/runtime.rs

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

_{Comment @coderabbitai help to get the list of available commands.}

[tinysweeper]

Tiny Sweeper review

Tiny Sweeper reviewed this change across 6 lane(s) and found 7 active actionable finding(s). Detailed lane evidence and any incomplete work are listed below.

State: Changes requested
Priority: high
Reviewed head: 9e76d28fd226
Updated: 1791516383 (Unix time)

Review snapshot

Change surface	Files	Review signal	Count
Production	4	Active findings	10
Tests	3	Noted findings	0
Documentation	1	Resolved findings	42
Configuration	0	Pending checks/questions	0

Completeness: Complete
Test assessment: No supported feature-to-test mapping was available; this does not mean tests are absent or passed.

What changed

The review could not produce a supported behavioral summary; inspect the cited changed surface and lane details below.

Features

None identified with supported citations.

Tests

No supported feature-to-test mapping was produced. Test execution is not inferred.

Findings

• medium · critique · Report unknown tasks when releasing cancellation — `release_cancellation_trusted` returns `Ok(())` when `task_id` is absent, so callers cannot distinguish releasing a registered task from releasing nothing. A stale or mistyped ID i (crates/tinyagents\-tasks/src/runtime\.rs:194)
• medium · critique · Do not hide owned-job registry lookup failures — Unlike `get`, this path converts every `DetachedTaskRegistryError` into `None`. That hides failures such as a poisoned registry lock and makes them indistinguishable from an unknow (crates/tinyagents\-orchestration/src/subagent/invocation/jobs\.rs:311)
• medium · critique · Handle task registration errors without panicking — A registration failure is converted into a panic. The registry API can reject the operation when its lock is poisoned, and the previous implementation recovered poisoned state inst (crates/tinyagents\-orchestration/src/subagent/invocation/jobs\.rs:116)
• medium · critique · Require ownership before releasing a cancellation token — This public release operation accepts only a task ID and performs no ownership check. Any caller that can obtain a registry reference can clear another owner's live cancellation to (crates/tinyagents\-tasks/src/runtime\.rs:190)
• medium · critique · Reject cooperative cancellation after token release — A task with a non-terminal status but `cancellation == None` passes the ownership and terminal checks, then this block skips cancellation and still returns `Ok`. This can occur aft (crates/tinyagents\-tasks/src/runtime\.rs:181)
• high · security · Add the referenced registry test module — The revision adds cooperative registration, cooperative cancellation, optional abort handles, and cancellation-token release behavior, but the crate has no sibling registry test mo (crates/tinyagents\-tasks/src/runtime\.rs:108)
• medium · security · Return an error for unknown cancellation entries — This method returns `Ok(())` when `task_id` is not registered, hiding a detached-registry lookup failure from its caller. A stale or mistyped task ID is indistinguishable from a su (crates/tinyagents\-tasks/src/runtime\.rs:194)
• medium · security · Make cancellation state update atomic with task cancellation — `cancel_cooperative` can cancel the task and allow the child to finish before `control.cancellation_requested` is set below. In that window, `settle` observes the request flag as f (crates/tinyagents\-orchestration/src/subagent/invocation/jobs\.rs:292)
• medium · security · Require ownership before releasing cancellation tokens — This public release operation accepts only a task ID and performs no ownership check, unlike the owner-aware control methods. Any caller that can access the registry can drop anoth (crates/tinyagents\-tasks/src/runtime\.rs:190)
• medium · tests · Log registry failures in get_owned instead of returning None — `get` and the `sorted` listing now log non-`Unknown` detached-registry errors before degrading, but `get_owned` still collapses every error — including a poisoned internal lock — i (crates/tinyagents\-orchestration/src/subagent/invocation/jobs\.rs:311)

Resolved this pass

• Add the referenced registry test module
• Preserve stable ordering when listing all jobs
• Sort owned jobs by job ID before returning them
• Require ownership before releasing a cancellation token
• Preserve stable ordering when listing jobs
• Add the referenced registry test module
• Do not hide detached-registry lookup failures
• Preserve stable ordering when listing all jobs
• Sort owned jobs by job ID before returning them
• Require ownership before releasing a cancellation token
• Preserve stable ordering when listing jobs
• Add the referenced registry test module
• Do not hide detached-registry lookup failures
• Preserve stable ordering when listing all jobs
• Sort owned jobs by job ID before returning them
• Require ownership before releasing a cancellation token
• Preserve stable ordering when listing jobs
• Do not hide detached-registry lookup failures
• Preserve stable ordering when listing all jobs
• Sort owned jobs by job ID before returning them
• Require ownership before releasing a cancellation token
• Preserve stable ordering when listing jobs
• Add the referenced registry test module
• Do not hide detached-registry lookup failures
• Preserve stable ordering when listing all jobs
• Sort owned jobs by job ID before returning them
• Require ownership before releasing a cancellation token
• Preserve stable ordering when listing jobs
• Add the referenced registry test module
• Preserve stable ordering when listing all jobs
• Sort owned jobs by job ID before returning them
• Preserve stable ordering when listing jobs
• Add the referenced registry test module
• Preserve stable ordering when listing all jobs
• Sort owned jobs by job ID before returning them
• Preserve stable ordering when listing jobs
• Add the referenced registry test module
• Do not hide detached-registry lookup failures
• Preserve stable ordering when listing all jobs
• Sort owned jobs by job ID before returning them
• Require ownership before releasing a cancellation token
• Preserve stable ordering when listing jobs

Before merge

• Address Add the referenced registry test module (crates/tinyagents\-tasks/src/runtime\.rs).

How this fits together

flowchart LR
  n0["new"]:::impacted
  n1["invoke_in_parent_context"]:::impacted
  n2["child_config"]:::impacted
  n3["run_child"]:::impacted
  n4["RunContext"]:::impacted
  n5["AgentRun"]:::impacted
  n1 -->|calls| n2
  n1 -->|uses| n4
  n2 -->|calls| n0
  n3 -->|uses| n4
  n3 -->|uses| n5
  classDef changed fill:#0d4429,stroke:#238636,color:#e6edf3
  classDef impacted fill:#161b22,stroke:#6e7681,color:#c9d1d9
  classDef flagged fill:#5a1e02,stroke:#d93f0b,color:#ffffff
  classDef blocking fill:#67060c,stroke:#f85149,color:#ffffff

Loading

Agent review details

critique

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: Reviewed 4 files; 5 findings. (1 observation(s) grouped into shared inline comments) _Code retrieval was unavailable (model: ladder embeddings returned 400 Bad Request: {"error":{"message":"unknown ladder vectors; known ladders are flash (also chat-v1, flash-v1), instant (also no-think, instant-v1), reasoning (also deepseek), max-reasoning (also max-reasoning-v1), deepseek-flash (also reasoning-v1, agentic-v1), deep (also luna), scribe, uncensored, vectors-oai3 (also embeddings-oai3-v1), vision (also vision-v1, multimodal-v1), image (also images-v1, image-v1), vi), so this review saw the diff alone._ _Memory was unavailable (model: cortex: v1/recall: error sending request for url (http://cortexdb:3141/v1/recall\)\), so this review ran without it._
• Evidence: crates/tinyagents\-tasks/src/runtime\.rs — Report unknown tasks when releasing cancellation
• Evidence: crates/tinyagents\-orchestration/src/subagent/invocation/jobs\.rs — Do not hide owned-job registry lookup failures
• Evidence: crates/tinyagents\-orchestration/src/subagent/invocation/jobs\.rs — Handle task registration errors without panicking
• Evidence: crates/tinyagents\-tasks/src/runtime\.rs — Require ownership before releasing a cancellation token
• Evidence: crates/tinyagents\-tasks/src/runtime\.rs — Reject cooperative cancellation after token release

security

• Conclusion: Failure
• Scope reviewed: all assigned evidence
• Lane summary: Reviewed 4 files; 4 findings. (1 observation(s) grouped into shared inline comments) _Code retrieval was unavailable (model: ladder embeddings returned 400 Bad Request: {"error":{"message":"unknown ladder vectors; known ladders are flash (also chat-v1, flash-v1), instant (also no-think, instant-v1), reasoning (also deepseek), max-reasoning (also max-reasoning-v1), deepseek-flash (also reasoning-v1, agentic-v1), deep (also luna), scribe, uncensored, vectors-oai3 (also embeddings-oai3-v1), vision (also vision-v1, multimodal-v1), image (also images-v1, image-v1), vi), so this review saw the diff alone._ _Memory was unavailable (model: cortex: v1/recall: error sending request for url (http://cortexdb:3141/v1/recall\)\), so this review ran without it._
• Evidence: crates/tinyagents\-tasks/src/runtime\.rs — Add the referenced registry test module
• Evidence: crates/tinyagents\-tasks/src/runtime\.rs — Return an error for unknown cancellation entries
• Evidence: crates/tinyagents\-orchestration/src/subagent/invocation/jobs\.rs — Make cancellation state update atomic with task cancellation
• Evidence: crates/tinyagents\-tasks/src/runtime\.rs — Require ownership before releasing cancellation tokens

tests

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: The registry-adapter rewrite is now backed by a real test module (mod_registry_tests.rs) that pins registration, ownership, error ordering, steering release, sweep survival and list sorting, and the tasks crate gained cooperative-cancel and release tests — the earlier missing-test findings are resolved. Sorting and logged registry failures are also fixed. One earlier concern remains: `get_owned` still converts every registry failure into `None` without logging, so a poisoned-lock error is indistinguishable from an unknown job. Otherwise this looks safe to merge. (2 earlier finding(s) still open) (1 observation(s) grouped into shared inline comments) _Code retrieval was unavailable (model: ladder embeddings returned 400 Bad Request: {"error":{"message":"unknown ladder vectors; known ladders are flash (also chat-v1, flash-v1), instant (also no-think, instant-v1), reasoning (also deepseek), max-reasoning (also max-reasoning-v1), deepseek-flash (also reasoning-v1, agentic-v1), deep (also luna), scribe, uncensored, vectors-oai3 (also embeddings-oai3-v1), vision (also vision-v1, multimodal-v1), image (also images-v1, image-v1), vi), so this review saw the diff alone._ _Memory was unavailable (model: cortex: v1/recall: error sending request for url (http://cortexdb:3141/v1/recall\)\), so this review ran without it._
• Evidence: crates/tinyagents\-orchestration/src/subagent/invocation/jobs\.rs — Log registry failures in get_owned instead of returning None

commits

• Conclusion: Neutral
• Scope reviewed: all assigned evidence
• Lane summary: Nothing sensitive found in what this pull request commits.

description

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: The revision completes the registry-as-adapter refactor and fixes all earlier findings: the referenced registry test module now exists (mod_registry_tests.rs), listings are sorted via `sorted()`, detached-registry errors are logged or mapped rather than silently dropped, and ownership is enforced through the registry's owner checks. The new cooperative registration/cancel APIs are covered by tests, lock ordering (controls → tasks) is consistent, and the cancel-then-settle protocol preserves the first-terminal-state-wins behaviour. The change looks sound and ready to merge. _Code retrieval was unavailable (model: ladder embeddings returned 400 Bad Request: {"error":{"message":"unknown ladder vectors; known ladders are flash (also chat-v1, flash-v1), instant (also no-think, instant-v1), reasoning (also deepseek), max-reasoning (also max-reasoning-v1), deepseek-flash (also reasoning-v1, agentic-v1), deep (also luna), scribe, uncensored, vectors-oai3 (also embeddings-oai3-v1), vision (also vision-v1, multimodal-v1), image (also images-v1, image-v1), vi), so this review saw the diff alone._ _Memory was unavailable (model: cortex: v1/recall: error sending request for url (http://cortexdb:3141/v1/recall\)\), so this review ran without it._

e2e

• Conclusion: Neutral
• Scope reviewed: all assigned evidence
• Lane summary: No end-to-end harness in this repository: no e2e test files and no e2e workflow.

Evidence and run details

• Models: gpt-5.6-luna, glm-5.3-flash
• Spend: $0.025483
• Tokens: 454543 input · 36749 output · 39261 cached · 0 embedding

Head	State	Pass summary
a878424b279a	changes requested	7 active finding(s), 0 resolved finding(s) (at 1791516029)
9e76d28fd226	changes requested	10 active finding(s), 42 resolved finding(s) (at 1791516383)

[tinysweeper]

Requesting changes: 2 lane(s) blocking, worst finding is critical.

Fix or reply to the findings below and push. The next review clears this automatically once they are gone — you should not need to dismiss anything by hand.

             $0.0302 · 623,973 in / 35,558 out · 66,973 cached (11%) · gpt-5.6-luna, glm-5.3-flash
critique:    $0.0158 · 316,619 in / 15,591 out · 34,570 cached (11%) · gpt-5.6-luna, glm-5.3-flash
security:    $0.0129 · 236,790 in / 11,771 out · 32,211 cached (14%) · gpt-5.6-luna
tests:       $0.0012 · 39,023 in  / 5,960 out  · 64 cached (0%)      · glm-5.3-flash
description: $0.0001 · 15,782 in  / 492 out    · 64 cached (0%)      · glm-5.3-flash

[tinysweeper]

Requesting changes: 1 lane(s) blocking, worst finding is high.

Fix or reply to the findings below and push. The next review clears this automatically once they are gone — you should not need to dismiss anything by hand.

             $0.0255 · 454,543 in / 36,749 out · 39,261 cached (9%)  · gpt-5.6-luna, glm-5.3-flash
critique:    $0.0141 · 232,715 in / 19,100 out · 18,310 cached (8%)  · gpt-5.6-luna, glm-5.3-flash
security:    $0.0110 · 172,342 in / 14,016 out · 17,879 cached (10%) · gpt-5.6-luna
tests:       $0.0001 · 16,589 in  / 1,040 out  · 1,536 cached (9%)   · glm-5.3-flash
description: $0.0001 · 16,484 in  / 801 out    · 1,408 cached (9%)   · glm-5.3-flash