fix(agent-core-v2): harden wire restore and background task lifecycle by sailist · Pull Request #1635 · MoonshotAI/kimi-code

sailist · 2026-07-13T16:05:02Z

Related Issue

No linked issue. This fixes three related agent-core-v2 durability, restore-compatibility, and task-lifecycle problems described below.

Problem

Wire-log replacement could lose or silently fail live records. Restore did not await migration rewrites, live appends could be overwritten by a stale replacement snapshot, and an auto-drain failure let a later flush() report success. Non-empty unversioned logs were also stamped as current instead of being rejected.
Session close did not stop task-owned processes. Agent removal aborted the turn/activity kernel but did not apply the task manager's graceful TERM → configured grace → force-stop policy. Disposal paths that bypassed normal removal could leave a process that ignored SIGTERM alive.
Per-agent task persistence broke previous v2 restores. Moving task records to the owning agent fixed v1 layout compatibility and cross-agent isolation, but the main agent could no longer see records written by the previous v2 session-level root.

What changed

1. Wire and AppendLog durability

AgentWireRecordService.restore() awaits migration rewrites and propagates rewrite failures. Agent create and session fork reject non-empty logs without a protocol metadata envelope; empty logs still receive normal metadata.
AppendLogStore.rewrite(records) now has an explicit cutover contract: records replaces history already durable before the call, while queued or in-flight appends remain an ordered live tail after the atomic replacement.
A cutover epoch prevents an old in-flight drain from deleting or repeatedly draining the tail after rewrite ownership changes. flush() remains pending through the post-rewrite durable append.
Ambiguous append and atomic-rewrite failures are sticky for that acquired buffer generation. Later flush, read, or rewrite calls return the same error instead of guessing whether storage committed and risking duplicates.
Releasing the final acquired owner retires the keyed buffer only after in-flight work settles. A replacement owner waits for that retirement, and identity checks prevent an old owner from deleting the replacement state.

2. Background process teardown

AgentLifecycleService.remove() awaits AgentTaskService.stopAllOnExit() before activity drain and scope disposal.
Normal stop sends SIGTERM/cancel, waits killGracePeriodMs, then calls the task's force-stop callback when needed. The keepAliveOnExit opt-out skips task-manager teardown for independently living work such as processes.
Scope disposal is a synchronous safety net: it cancels/aborts active work and immediately starts best-effort force-stop, covering bootstrap failure, rollback, and root-scope disposal paths that cannot await graceful shutdown.
Legacy [background] values are merged beneath [task]; the environment-only keepAliveOnExit overlay no longer hides maxRunningTasks or killGracePeriodMs.

3. Task persistence compatibility

New writes remain rooted at the owning agent, matching v1's runtime <sessionDir>/agents/<id>/tasks/ layout.
The main agent reads the previous v2 session-level task root only when its local record/output is absent. Local empty, corrupt, or unrecognized entries remain authoritative; output snapshots report the path of the root actually used.
Subagents never read the shared fallback, so they cannot claim another agent's legacy task.

Review follow-up

Added controlled rewrite gates for snapshot → pending append, old drain already in flight, post-rewrite flush durability, ambiguous append failure, atomic rewrite failure, and acquired-owner retirement.
Added an ignore-SIGTERM disposal contract that observes SIGTERM followed by immediate SIGKILL, plus a real process keep-alive contract.
Added persistence and public restore contracts for local-first main-agent fallback, real legacy output paths, corrupt/empty boundaries, and subagent isolation.
Moved implementation rationale into domain headers and documented background.kill_grace_period_ms in the mirrored English and Chinese configuration references.
Session materialization rollback and same-ID initialization admission are handled separately in PR fix(agent-core-v2): roll back failed session initialization #1642.

keepAliveOnExit in this fix is intentionally scoped to task-owned work that can live independently of the Session scope, such as a detached process. Session-scoped subagents still end with their Session lifecycle; this PR does not claim to detach a DI child scope from its owner.

Verification

agent-core-v2: 236 files / 3288 tests passed with one worker.
Targeted lifecycle/durability/restore suite: 6 files / 110 tests passed.
agent-core-v2 build, typecheck, and domain lint passed (848 files).
kap-server typecheck passed.
Targeted type-aware lint completed with 0 errors (existing warnings remain).
VitePress docs build, changeset status, and git diff --check passed.
Independent fixed-diff review reported no actionable findings for cutover, concurrent rewrite, sticky recovery, retirement handoff, and multi-key close.

Checklist

I have read the CONTRIBUTING document.
I have linked a related issue, or explained the problem above.
I have added tests that prove my feature works.
Ran gen-changesets skill. (3 focused patch changesets)
Ran gen-docs skill. (Updated mirrored EN/ZH config references)

…ioned logs (8 files) - await the migration rewrite in wireRecordService.restore() so restore only resolves once the migrated log is durable and rewrite failures surface - serialize AppendLogStore.rewrite() behind the log flush so appends arriving during the whole-file replace land after the new content - reject wire logs missing the metadata envelope (missingWireMetadataError) in restore, session fork, and agent create instead of fabricating a current-version envelope over legacy records

- add IAgentTaskService.stopAllOnExit: suppress terminal notifications, then SIGTERM → grace → SIGKILL for every active task (v1 stopBackgroundTasksOnExit parity, gated by keepAliveOnExit) - call stopAllOnExit from agentLifecycle.remove before scope disposal, and abort live tasks in AgentTaskService.dispose as a last resort for disposal paths that bypass the graceful close - honor [task] killGracePeriodMs in the stop grace window and bind the v1 KIMI_CODE_BACKGROUND_KEEP_ALIVE_ON_EXIT env override for keepAliveOnExit

- persist task records under <sessionScope>/agents/<agentId>/tasks/ (v1's per-agent layout) so tasks written by older versions are found on resume - stop one agent's restore from loading, marking lost, or re-notifying another agent's tasks - add a per-agent isolation test covering loadFromDisk + reconcile

changeset-bot · 2026-07-13T16:05:08Z

🦋 Changeset detected

Latest commit: 9fcca23

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@moonshot-ai/kimi-code	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

pkg-pr-new · 2026-07-13T16:07:15Z

pnpm dlx https://pkg.pr.new/@moonshot-ai/kimi-code@9fcca23

npx https://pkg.pr.new/@moonshot-ai/kimi-code@9fcca23

commit: 9fcca23

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 942ad7c0f8

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

wbxl2000 · 2026-07-13T20:37:15Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ddbc96d52f

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

wbxl2000 · 2026-07-13T20:59:14Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 96cc1e4cde

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

wbxl2000 · 2026-07-13T23:07:33Z

@codex review

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9bee9d83e8

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

wbxl2000 · 2026-07-13T23:27:52Z

@codex review

chatgpt-codex-connector · 2026-07-13T23:36:17Z

Codex Review: Didn't find any major issues. What shall we delve into next?

Reviewed commit: 9fcca23f28

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

sailist added 3 commits July 14, 2026 00:03

chatgpt-codex-connector Bot reviewed Jul 13, 2026

View reviewed changes

Comment thread packages/agent-core-v2/src/agent/task/configSection.ts

Comment thread packages/agent-core-v2/src/persistence/backends/node-fs/appendLogStore.ts Outdated

Comment thread packages/agent-core-v2/src/persistence/backends/node-fs/appendLogStore.ts Outdated

fix(agent-core-v2): address task lifecycle review findings

ddbc96d

chatgpt-codex-connector Bot reviewed Jul 13, 2026

View reviewed changes

Comment thread packages/agent-core-v2/src/persistence/backends/node-fs/appendLogStore.ts Outdated

wbxl2000 added 2 commits July 14, 2026 04:51

fix(agent-core-v2): preserve queued appends across rewrites

44e58a7

test(agent-core-v2): cover rewrite cutover flush

96cc1e4

chatgpt-codex-connector Bot reviewed Jul 13, 2026

View reviewed changes

Comment thread packages/agent-core-v2/src/session/agentLifecycle/agentLifecycleService.ts

Comment thread packages/agent-core-v2/src/agent/task/taskService.ts

fix(agent-core-v2): surface append durability failures

1e7028d

wbxl2000 mentioned this pull request Jul 13, 2026

fix: stop individual swarm members from the web UI #1645

Open

5 tasks

wbxl2000 added 4 commits July 14, 2026 05:30

fix(agent-core-v2): retire released append log buffers

3cea6bb

fix(agent-core-v2): preserve session-level task records

8713fe3

Merge remote-tracking branch 'origin/main' into codex/pr-1635-followup

278ef5a

fix(agent-core-v2): harden append log cutover recovery

9bee9d8

chatgpt-codex-connector Bot reviewed Jul 13, 2026

View reviewed changes

Comment thread packages/agent-core-v2/src/agent/task/taskService.ts Outdated

fix(agent-core-v2): suppress only background exit notifications

9fcca23

wbxl2000 mentioned this pull request Jul 14, 2026

fix(agent-core-v2): roll back failed session initialization #1642

Draft

5 tasks

sailist merged commit e49b3b8 into MoonshotAI:main Jul 14, 2026
15 checks passed

github-actions Bot mentioned this pull request Jul 14, 2026

ci: release packages #1583

Merged

Uh oh!

Conversation

sailist commented Jul 13, 2026 • edited by wbxl2000 Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Related Issue

Problem

What changed

1. Wire and AppendLog durability

2. Background process teardown

3. Task persistence compatibility

Review follow-up

Verification

Checklist

Uh oh!

changeset-bot Bot commented Jul 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🦋 Changeset detected

Uh oh!

pkg-pr-new Bot commented Jul 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

wbxl2000 commented Jul 13, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

wbxl2000 commented Jul 13, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

wbxl2000 commented Jul 13, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

wbxl2000 commented Jul 13, 2026

Uh oh!

chatgpt-codex-connector Bot commented Jul 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

sailist commented Jul 13, 2026 •

edited by wbxl2000

Loading

changeset-bot Bot commented Jul 13, 2026 •

edited

Loading

pkg-pr-new Bot commented Jul 13, 2026 •

edited

Loading