revert: #6349 durable task owner manager (breaks dev shard-8 + edit benchmark) by Yeachan-Heo · Pull Request #6351 · Yeachan-Heo/gajae-code

Yeachan-Heo · 2026-10-05T05:37:01Z

Reverts #6349 (merge bb00f4c8), which I merged last round. It breaks dev in suites its CI didn't run:

coding-agent:shard-8-of-8: SessionManager session ids (5 tests, e.g. forks managed artifacts above the recovery-state size cap, rejects a byte-identical whole-root fork artifact replacement, fails closed when retained artifact capture rejects a substituted tree). Passed on 4f1c2add, fails on bb00f4c8.
typescript-edit-benchmark: writeConversationDump > writes benchmark conversations as session dumps and copies artifacts throws task_artifact_owner_session_manager_closed. Passed on 4f1c2add, fails on bb00f4c8.
Locally on test/session-manager/session-id.test.ts: 4f1c2add 15 pass / 5 fail (host baseline), bb00f4c8 6 pass / 14 fail.

This is a fallback: if snowykr fix-forwards first, I'll close it. The 3 sdk-broker failures (#6339) are separate and tracked in #6347.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]

…uction-installed-f4b3" This reverts commit bb00f4c, reversing changes made to 4f1c2ad.

…ot deleted)

chatgpt-codex-connector · 2026-10-05T05:37:18Z

Codex Review Summary

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

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-05T05:41:19.065931Z	`f639b94`	PR opened

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.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f639b94893

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".

chatgpt-codex-connector · 2026-10-05T05:41:16Z

 	const keys = Object.keys(record.patch);
-	if (
-		!keys.every(
-			key =>
-				key === "cwd" ||
-				key === "title" ||
-				key === "titleSource" ||
-				key === "starred" ||
-				key === "taskArtifactOwner",
-		)
-	)
-		return false;
-	const { cwd, title, titleSource, starred, taskArtifactOwner } = record.patch;
-	try {
-		parseTaskArtifactOwnerLocator(taskArtifactOwner);
-	} catch {
-		return false;
-	}
+	if (!keys.every(key => key === "cwd" || key === "title" || key === "titleSource" || key === "starred")) return false;
+	const { cwd, title, titleSource, starred } = record.patch;


Preserve owner locators written by the reverted release

When a managed session was persisted by the immediate predecessor bb00f4c8, its durable artifact owner was recorded in a header_patch containing taskArtifactOwner. This whitelist now rejects that patch, and buildFileEntriesFromRecords silently skips it; after restart, the session instead derives a transcript-basename artifact store, so existing artifact:// and agent:// references into .task-artifact-owners stop resolving and subsequent output is split into a new store. Keep read-side parsing and restoration for already-written owner locators even if provisioning new durable owners is reverted.

Useful? React with 👍 / 👎.

Yeachan-Heo · 2026-10-05T06:06:53Z

CI on bd36a2b: 38 passed, 3 failed. The only real failure is Windows dev:doctor + session-path regression (session-state-lock.test.ts: reclaims a base regular-file lock left by a dead owner instead of faulting, reclaims a malformed owner file only once it is stale, with owner_mismatch). The other 2 are the summary checks. This revert doesn't touch the lock code: it only changes sdk/session.ts, session-manager.ts, task/index.ts and 3 tests. I've re-run the failed jobs. @probepark @snowykr please review; this restores shard-8 and the edit benchmark on dev.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]

probepark

Code review (CI still running on bd36a2b; approval decision follows once CI finishes).

Review (head bd36a2b, gajae-reviewer on behalf of probepark)

CI: still running. 38 checks pass. The Windows dev:doctor + session-path regression rerun now passes. Affected path validation / evidence producer is pending, and the Affected path validation aggregate has not reported yet. Planned coverage includes check:@gajae-code/coding-agent plus the 3 touched test files (artifact-tree-authorization, move-to, no-session-output-refs).
Scope: +82 / -791, 6 files. Three are source files in packages/coding-agent (sdk/session.ts, session/session-manager.ts, task/index.ts), +43/-334 reviewable. The other 3 are tests.
Conventions: changelog fragment changelog.d/task-owner-durable-manager.md is kept. No generated files. No labels.

Revert fidelity: I checked this mechanically. git diff --stat bb00f4c8^1 bd36a2b is exactly 1 file, packages/coding-agent/changelog.d/task-owner-durable-manager.md (+3). The head tree is byte-identical to the pre-merge dev tip 4f1c2add everywhere except that fragment. The PR body says shard-8 and typescript-edit-benchmark passed on that tree. No hand-edited partial revert is hidden in the source.

Notable (non-blocking):

packages/coding-agent/src/session/session-manager.ts header-patch whitelist (codex P1, f639b94). This is correct mechanically: a session persisted by bb00f4c8 carries a header_patch with taskArtifactOwner, and after the revert that patch is skipped, so its .task-artifact-owners artifacts stop resolving. The exposure is narrow. bb00f4c8 landed on dev at 04:21Z today and is not on main or in any release (compare bb00f4c8...main ahead_by=0; latest tag v0.18.7 predates it). Only dev-tip sessions from the last ~2 h are affected. The fix-forward that relands #6349 restores the read side. I am not counting this as a merge blocker for an owner-directed dev revert.
changelog.d/task-owner-durable-manager.md still advertises "Persist managed task artifacts under a durable logical-session owner…". After this merge, that fix is not in the tree. If a release is cut before the reland, the notes will claim a fix that is not shipped. Before the release cut, either edit the fragment text (keep the file) to say it was reverted, or make sure the fix-forward lands first.

Blocking: none found in step 3.

Approval is held until Affected path validation finishes on this exact head.

probepark · 2026-10-05T06:29:08Z

Additional evidence for this revert from the bb00f4c8 Dev CI run (https://github.com/Yeachan-Heo/gajae-code/actions/runs/37263181896): #6349 also breaks coding-agent:shard-1-of-8. That is in addition to shard-8 and the edit benchmark listed above.

task-autorouting-preflight.test.ts: 3 tests (T1m, T2m, T3m) fail at :194 with managed parent artifact manager unavailable.
managed staged publication certainty: 5 tests fail with the same error.
multi-session session runner > runs the complete workload and compares reproducibly normalized evidence also fails.

None of these failed on 4f1c2add (https://github.com/Yeachan-Heo/gajae-code/actions/runs/37263170666). The range 4f1c2add..bb00f4c8 contains only #6349. The test blobs did not change (task-autorouting-preflight.test.ts 285e3bbaa976, session-id.test.ts 9f387c13a4b5, bench-multisession-runner.test.ts fd046fec21b3). The source blobs did: session-manager.ts 4b6da5e7c341 -> 0262d056c56e, task/index.ts d3358d82900a -> 3679a614714f, sdk/session.ts bffbf2e94ebd -> c39566a14f5e.

The 3 sdk-broker session.delete failures on shard-1 come from #6339 and are tracked in #6347, as noted above. A fix-forward for #6349 should also turn the shard-1 tests above green.

probepark · 2026-10-05T06:44:32Z

gajae-merge-routine 2026-10-05T06:44Z: dev is RED again at bb00f4c8 (https://github.com/Yeachan-Heo/gajae-code/actions/runs/37263181896). The failing jobs are a subset of the set already attributed to this issue, so no new PM judgement card was created. Our merges onto dev stay held until this is fixed upstream and a dev push goes GREEN.

Yeachan-Heo · 2026-10-05T06:52:09Z

@probepark Affected path validation has finished on bd36a2b: CI passed 42/0, including the re-run Windows session-path job. Please review again.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]

probepark

Review (head bd36a2b, gajae-reviewer on behalf of probepark)

The code review for this head is the earlier review #6351 (review), which found 0 blocking issues. CI has now finished, and I found no blocking issues.

CI: green. The approve gate returned ALLOW on bd36a2b with 0 pending and 0 failed, and check:@gajae-code/coding-agent was in the plan. This matches the owner's comment that 42/0 passed, including the re-run Windows session-path job. Only the merge-approval gate checks are still waiting.
Scope: +82 / -791, 6 files. This is a mechanical revert of #6349 (bb00f4c8). Apart from the kept changelog fragment, the head tree is byte-identical to bb00f4c8^1.
Conventions: the changelog fragment is kept, no generated files are touched, and there are no labels.
Notable (non-blocking, carried over from the code review): the session-manager header-patch whitelist only affects dev-tip sessions created after bb00f4c8. The text in changelog.d/task-owner-durable-manager.md must be fixed before the release cut, unless the reland of #6349 lands first.
Blocking: none.

Body verdict line count=0, not updated. Suggested verdict line: gajae.pr-review-verdict.v1 merge-approved sha256:e20099646a8dffac93cc0d0f4d3919209e4cd5736bef8962ae210ab2b9a21cfb reviewer:human reviewer-id:probepark evidence:ci-green;approve-gate-allow;mechanical-revert-of-bb00f4c8-verified

Verdict: gajae.pr-review-verdict.v1 merge-approved sha256:e20099646a8dffac93cc0d0f4d3919209e4cd5736bef8962ae210ab2b9a21cfb reviewer:human reviewer-id:probepark evidence:ci-green;approve-gate-allow;mechanical-revert-of-bb00f4c8-verified

Yeachan-Heo · 2026-10-05T07:21:03Z

Merged into dev as a8ece169.

probepark approved the exact head bd36a2b.
CI on that head: 42 passed, 0 failed.
I merged it locally with current dev (which now includes fix: insert paths and task input verbatim at placeholder substitutions #6352 and perf(natives): retain exact guards while reducing reaper overhead #6353) and ran test/session-manager/: 369 passed / 5 failed on the merged tree, vs 365 / 15 on unmodified dev. The 5 remaining failures are this host's baseline from before fix(session): persist and restore logical task artifact owners #6349.

Note for 형님 before the next release cut: changelog.d/task-owner-durable-manager.md was kept, so it still describes the reverted #6349 fix. Either the reland lands first, or that fragment needs editing.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]

Revert "Merge pull request #6349 from Yeachan-Heo/fix/task-owner-prod…

f639b94

…uction-installed-f4b3" This reverts commit bb00f4c, reversing changes made to 4f1c2ad.

Yeachan-Heo requested review from probepark and snowykr as code owners October 5, 2026 05:37

chore: keep #6349 changelog note (CI requires changelog entries are n…

bd36a2b

…ot deleted)

Yeachan-Heo mentioned this pull request Oct 5, 2026

fix(session): persist and restore logical task artifact owners #6349

Merged

chatgpt-codex-connector Bot reviewed Oct 5, 2026

View reviewed changes

probepark reviewed Oct 5, 2026

View reviewed changes

Yeachan-Heo mentioned this pull request Oct 5, 2026

fix: insert paths and task input verbatim at placeholder substitutions #6352

Merged

4 tasks done

Yeachan-Heo requested a review from probepark October 5, 2026 06:52

probepark approved these changes Oct 5, 2026

View reviewed changes

Yeachan-Heo merged commit a8ece16 into dev Oct 5, 2026
91 of 94 checks passed

Yeachan-Heo deleted the revert/6349-durable-owner-manager branch October 5, 2026 07:21

Yeachan-Heo mentioned this pull request Oct 5, 2026

test(sdk): pin correlated agent_end across todo-reminder continuation #6344

Merged

1 of 3 tasks

Navigation Menu

Repository navigation

revert: #6349 durable task owner manager (breaks dev shard-8 + edit benchmark) - #6351

revert: #6349 durable task owner manager (breaks dev shard-8 + edit benchmark)#6351
Yeachan-Heo merged 2 commits into
devfrom
revert/6349-durable-owner-manager

Yeachan-Heo commented Oct 5, 2026

Uh oh!

chatgpt-codex-connector Bot commented Oct 5, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Oct 5, 2026

Uh oh!

Yeachan-Heo commented Oct 5, 2026

Uh oh!

probepark left a comment

Uh oh!

probepark commented Oct 5, 2026

Uh oh!

probepark commented Oct 5, 2026

Uh oh!

Yeachan-Heo commented Oct 5, 2026

Uh oh!

probepark left a comment

Uh oh!

Uh oh!

Yeachan-Heo commented Oct 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

Yeachan-Heo commented Oct 5, 2026

Uh oh!

chatgpt-codex-connector Bot commented Oct 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Oct 5, 2026

Choose a reason for hiding this comment

Uh oh!

Yeachan-Heo commented Oct 5, 2026

Uh oh!

probepark left a comment

Choose a reason for hiding this comment

Uh oh!

probepark commented Oct 5, 2026

Uh oh!

probepark commented Oct 5, 2026

Uh oh!

Yeachan-Heo commented Oct 5, 2026

Uh oh!

probepark left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Yeachan-Heo commented Oct 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

chatgpt-codex-connector Bot commented Oct 5, 2026 •

edited

Loading