🤖 tests: validate persistent sub-agent compaction - #3832
Conversation
|
@codex review Please review the complete stacked diff, focusing on persistent child lifecycle across compaction, workspace-turn correlation/agent attribution preservation, and deterministic no-live-LLM coverage. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2e8b0a9248
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
Addressed the mid-stream correlation finding: @codex review |
|
Codex Review: Didn't find any major issues. Already looking forward to the next diff. Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
Keep active and reawakened children seamless across mechanical compaction turns, preserve internal attribution and workspace-turn correlation in follow-ups, and add deterministic no-provider coverage from unit through backend integration layers. --- _Generated with `mux` • Model: `openai:gpt-5.6-sol` • Thinking: `xhigh` • Cost: `$41.38`_ <!-- mux-attribution: model=openai:gpt-5.6-sol thinking=xhigh costs=41.38 -->
Carry the resolved workspace-turn correlation through automatic mid-stream compaction so the post-compaction continuation settles the existing reawakened execution. --- _Generated with `mux` • Model: `openai:gpt-5.6-sol` • Thinking: `xhigh` • Cost: `$41.38`_ <!-- mux-attribution: model=openai:gpt-5.6-sol thinking=xhigh costs=41.38 -->
60e70d7 to
362acf3
Compare
|
Rebased the two PR-only commits onto current @codex review |
|
Codex Review: Didn't find any major issues. Chef's kiss. Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
Replace the unused generic archive/unarchive/worktree lifecycle subsystem with the one supported persistent-child removal operation, and retain focused scope, active-state, ordering, race, and idempotency coverage. --- _Generated with `mux` • Model: `openai:gpt-5.6-sol` • Thinking: `xhigh` • Cost: `$54.08`_ <!-- mux-attribution: model=openai:gpt-5.6-sol thinking=xhigh costs=54.08 -->
|
Refactored the PR to net negative LoC through maintainability improvements: removed the unused generic archive/unarchive/delete-worktree TaskService backend and its lifecycle-only tests, while keeping historical transcript schemas/rendering and the supported direct @codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c59b84dd9b
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Only defer task settlement for compact turns that durably stage a follow-up; bare /compact continues normal child completion recovery. --- _Generated with `mux` • Model: `openai:gpt-5.6-sol` • Thinking: `xhigh` • Cost: `$88.81`_ <!-- mux-attribution: model=openai:gpt-5.6-sol thinking=xhigh costs=88.81 -->
|
Addressed the standalone compaction finding: TaskService now defers compact stream-end settlement only when the persisted compaction request actually contains a follow-up. Bare @codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2a4260cf1b
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Coordinate TaskService settlement with AgentSession's durable compaction outcome so successful continued compactions defer to their follow-up while rejected compactions resume normal child recovery. --- _Generated with `mux` • Model: `openai:gpt-5.6-sol` • Thinking: `xhigh` • Cost: `$99.28`_ <!-- mux-attribution: model=openai:gpt-5.6-sol thinking=xhigh costs=99.28 -->
|
Addressed the failed-compaction finding with an explicit AgentSession → TaskService outcome handshake keyed by the compact assistant message. TaskService now waits for persistence handling: successful continued compactions defer to the correlated follow-up; empty/invalid/persistence-failed compactions take normal child recovery. Coverage exercises successful continuation, bare @codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d868600bee
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Wait for AgentSession's compaction outcome, then inspect the durable compaction boundary to decide whether child settlement should defer to a real follow-up. --- _Generated with `mux` • Model: `openai:gpt-5.6-sol` • Thinking: `xhigh` • Cost: `$109.89`_ <!-- mux-attribution: model=openai:gpt-5.6-sol thinking=xhigh costs=109.89 -->
|
Addressed the real-stream metadata finding: TaskService no longer expects @codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 688444553e
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Resolve TaskService's compaction decision only after the durable continuation is accepted, so failed dispatches and non-continuing compactions take normal child recovery. --- _Generated with `mux` • Model: `openai:gpt-5.6-sol` • Thinking: `xhigh` • Cost: `$117.50`_ <!-- mux-attribution: model=openai:gpt-5.6-sol thinking=xhigh costs=117.50 -->
|
Addressed both handshake findings. AgentSession now resolves the compaction decision in @codex review |
|
Codex Review: Didn't find any major issues. Hooray! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
2 participants
Summary
Adds deterministic, no-live-LLM validation that persistent sub-agents remain seamless across chat compaction. Mechanical compact turns no longer advance child lifecycle state, reawakened continuation correlation and agent-initiated attribution survive compaction, and the mock AI runtime now mirrors real stream metadata closely enough for a true backend integration test.
The PR is also a net maintainability reduction: it removes the unused generic archive/unarchive/delete-worktree TaskService backend left behind after persistent sub-agents adopted the public active → inactive → removed lifecycle. The supported
task_removepath now uses one direct, parent-scoped persistent-child removal implementation.Background
The persistent sub-agent foundation from #3825 is merged into
main. A child compaction stream was visible directly toTaskServiceand could be mistaken for the end of delegated work, moving an active child toward completion recovery. Mock streams also dropped workspace-turn correlation, preventing high-confidence end-to-end coverage for a reawakened persistent child.The old generic lifecycle methods had no production callers; only TaskService tests exercised them. Historical
task_workspace_lifecycleschemas and transcript rendering remain intact for old chats.Implementation
TaskService; only the post-compaction delegated follow-up may settle the child or continuation handle.agentInitiatedand resolved workspace-turn correlation on persisted compaction follow-ups and their crash-safe dispatch/retry path.Validation
make static-checkbun test src/node/services/taskService.test.tsbun test src/node/services/agentSession.autoCompaction.test.ts src/node/services/agentSession.workspaceTurnInheritance.test.ts src/node/services/agentSession.continueMessageAgentId.test.ts src/node/services/mock/mockAiStreamPlayer.test.tsbun test src/node/services/tools/task_remove.test.tsTEST_INTEGRATION=1 bun x jest tests/ipc/tasks/persistentSubagentCompaction.test.ts --runInBandgit diff --numstat origin/main...HEAD: 790 additions, 1,239 deletions, net −449 LoCNo test invokes a live LLM API.
Risks
Low-to-moderate product risk. Compaction changes are narrowly scoped and covered end-to-end. Lifecycle deletion removes backend methods with no production callers; the supported
task_removepath retains scope, activity, descendant, race, and idempotency protections. Historical transcript rendering remains compatible.Generated with
mux• Model:openai:gpt-5.6-sol• Thinking:xhigh• Cost:$88.81