You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
fix(deep-scan): include accepted worker coverage in reports - #1593
Retain accepted worker coverage and its original evidence when publishing or recovering Deep Scan results.
Changes
Preserve worker ownership, scoped IDs, receipts, and deferred links across ID collisions and selected terminal checkpoints. Parent and worker tasks follow the retained surface through final ID assignment.
Restore the parent's surface copy when a surviving task still needs it and a selected worker checkpoint omits it. Keep its worker ownership, and match named or anonymous worker tasks by their verified source identity while preserving parent-only task associations.
Reserve receipt spellings for retained selected surfaces and task-required restored surfaces before repairing historical evidence. Include anonymous rows and duplicate legacy IDs without reserving discarded fallback surfaces. Reject receipt references outside the worker artifacts directory.
Recover retained receipts from archived attempts, restore missing copies, and preserve available modified copies. Keep unselected partial-checkpoint receipts out of accepted evidence repairs.
Prefer the selected parent snapshot, then historical source chronology, when linking or restoring a surface needed by a surviving task; historical bindings cannot overwrite the selected parent association.
Return receipt-repair warnings alongside workbench warnings from parent writes.
Keep saved coverage projection helpers in the internal workbench package and register them in the plugin file manifest for standalone and installed builds. This also keeps source files within the existing portable-source size limit.
Validate retained coverage before checkpoints and preserve valid findings when recovery skips malformed coverage records. Include canceled-worker checkpoint coverage during zero-accepted deadline recovery.
Align worker-overflow test diagnostics with the earlier coverage validator while retaining rejection and empty-artifact assertions.
Testing
Final Python regression run: 567 passed, six platform skips across 13 CLI, coverage, checkpoint, source-order, publication, and deferred-identity modules.
Recovery regressions cover named and anonymous worker-task links, restored worker ownership through collisions, receipt-case conflicts with duplicate legacy IDs, and selected parent evidence taking precedence over older partial snapshots. Fallback-receipt regressions pass in both case orders; controls verify that no-task and existing-worker-replacement cases do not reserve discarded surfaces. Tests verify frozen replay, retained receipts, and unchanged source artifacts.
Standard and Diff parent-write regressions fail without warning forwarding and pass with the fix; both the receipt-repair warning and the workbench warning are returned.
Four public worker-write/admission cases verify anonymous current receipt precedence, repeated writes, and unchanged archived receipts. These already passed with unchanged TypeScript production code because both public writers normalize surface IDs before reconciliation.
All five required portable source checks passed in order, including nine compatibility-checker tests. MCP typechecking, plugin bundle build, formatting, and diff checks passed.
Standalone plugin build and installed-plugin integrity tests passed (three tests), including loading the registered internal helper from packaged artifacts.
Earlier full local MCP suite: 510 passed, one failed, five skipped. The sole failure is the existing raw-byte filename fixture, which this macOS filesystem rejects with EILSEQ; full local green status is not claimed. After the final recovery fixes and helper extraction, the complete Deep Scan store integration test file passed again.
Includes the Windows fixture lifecycle repair from test(ci): wait for Windows sandbox fixture setup #1799, which passed full hosted CI and five repeated Windows configuration runs. Its resulting main commit also passed full CI.
Full hosted CI passed on 4fce8e3bf574f08ec847b0783834589bee92dbe9: 48 jobs succeeded and two conditional jobs were skipped. Codex and Copilot completed current-head reviews, all review discussions are resolved, and maintainer approval is present.
Risk and rollout
Coverage recovery determines which saved review records and receipts appear in a Deep report. Accepted result snapshots remain authoritative; recovery tests verify that original artifacts and parent associations are preserved. Full passing hosted CI and completed reviews remain required before merge.
Public disclosure review
No customer, partner, prospect, or user identities, data, or identifying details are included.
No credentials, personal data, private source, scan findings, or nonpublic links or tickets are included.
I reviewed the branch name, title, description, commits, changes, comments, logs, screenshots, attachments, and links for public disclosure.
mldangelo-oai
changed the title
fix: retain worker coverage in Deep Scan reports
fix(deep-scan): include accepted worker coverage in reports
Oct 12, 2026
The reason will be displayed to describe this comment to others. Learn more.
Thanks—the earlier dangling-link reproduction and prior collision/anonymous-receipt probes now pass. This owner-level merge-safety pass found a remaining parent-history regression, detailed inline: recovery can select stale surface metadata and receipts by checkpoint hash instead of retaining the current parent copy.
I independently confirmed the existing legacy duplicate-ID receipt issue. I also reproduced the latest review's updated-task link failure, including an interrupted real worker write: after candidate resolution, the retained task references a missing surface. With a parent ID collision, the related ownership issue instead redirects the updated task to an unrelated surface. The unchanged-task control passes. The parent warning issue is confirmed specifically at the immediate writer API response; valid receipts remain retained. These existing findings are not duplicated inline.
Validation: 105 focused Python cases, 205 Node test entries, 16 writer/admission receipt cases, builds/typechecks, and all five portable checks passed. Additional recovery/finalizer probes reproduce the history and duplicate-receipt regressions against the exact base, plus the updated-task defects with frozen replay and unchanged source bytes. The base itself has a separate raw updated-task link defect, so that comparison is not wholly green. Full local store/deadline integration remains limited by the unchanged trusted-owner guard; full suites and pending hosted jobs are not claimed green.
Please review 5ae3da9. This update addresses restored worker ownership, updated worker-task surface links, and parent receipt-repair warning forwarding. It also fixes the duplicate-ID receipt reservation issue from Copilot's previous overview: every retained row at the latest selected source order now reserves its receipts before historical repair. The new recovery and warning regressions failed before their fixes and now pass; the broader Python run passed 540 tests with six platform skips.
A legacy worker task without an id receives an aggregate ID such as first/0, but recovery generates first/saved-…. When a selected checkpoint retains that task but omits its surface, this ID comparison cannot match the recovered task to the parent copy. The restoration loop then leaves surfaceIds: ["first/api"] pointing to a missing surface. Track aggregate tasks through their source and retained output objects, as for surfaces, rather than comparing differently generated IDs. Extend the omitted-surface regression to cover anonymous worker tasks and frozen replay.
Please review 458ed55. It adds the parent-snapshot chronology fix to the previous worker-link, receipt-reservation, and warning-forwarding fixes: selected parent evidence now wins over older matching snapshots during fallback recovery. All four chronology reproductions failed before the fix and now pass; the broader Python run passed 544 tests with six platform skips, and the complete Deep Scan store integration file passed. Please also verify the earlier fixes against this current head.
Task-dependent parent surfaces are restored after _generic_surface_updates reserves receipt spellings and merges historical receipts. If a restored surface references Review.md while another surface's repair retains historical review.md in the same directory, the finalizer discards the restored surface's receipt and changes no_issue_found to needs_follow_up, even when that receipt exists. Include surviving task-dependent fallback surfaces in receipt reservation before historical repair. Add a regression combining parent fallback with a case-colliding historical receipt.
The reason will be displayed to describe this comment to others. Learn more.
Thanks—the earlier fallback chronology, duplicate receipt, updated-task/ownership, and immediate warning reproductions now pass. This review still confirms three issues already reported:
Anonymous persisted worker tasks: admission preserves a missing task ID, aggregation assigns first/0, and recovery assigns first/saved-…. When the selected checkpoint omits the surface, the association is lost and publication seals a dangling first/api link. Named-task and typed-writer controls pass. The base retains the original scoped association but also has a separate dangling raw task, so it is not a wholly clean baseline.
Selected parent association: with retained partial worker histories, the later historical binding overwrites the selected parent's binding. The sealed task points to old.md; the preceding head and the no-parent-history control keep new.md. This concerns partial checkpoint recovery, not completed discovery admission.
Fallback receipt precedence: historical review.md is repaired before the restored current Review.md is reserved. Selected-checkpoint recovery drops the current receipt reference and downgrades no_issue_found; the distinct-name control preserves both. Source files remain unchanged.
Fresh validation: 123 Python tests, three MCP-to-Python recovery cases, eight combined receipt/link controls, targeted publication/frozen-replay probes, builds/typechecks, formatting, and all five portable checks. Another 204 MCP cases, 72 receipt cases, and 16 writer/admission cases were reused with verified unchanged inputs and scope. Full local store/deadline integration remains limited by the unchanged ownership guard; no full typed-writer transaction or all-platform suite pass is claimed. Hosted CI is still pending. Existing findings are not duplicated inline.
Please review current head 614393613bdb653a19ed3bb7a282beadfceceb53 after the latest recovery fixes. Anonymous persisted worker tasks now retain their source association through ID normalization. Historical partial-worker bindings cannot overwrite the selected parent binding. Receipt repair reserves evidence for genuine surviving fallback surfaces, with controls for discarded surfaces and existing worker replacements. The coverage projection helpers were extracted to satisfy the existing portable-source file-size limit.
The 12-module Python run passed 558 tests with six platform skips. All five required portable checks passed in order, as did MCP typechecking, plugin build, and the complete Deep Scan store integration test file. Full hosted CI is pending on this head.
Please review current head 4fce8e3bf574f08ec847b0783834589bee92dbe9. This follow-up packages the saved coverage helper under the existing internal workbench package and registers it in the plugin file manifest. Installed and standalone plugins can now import it, and the repository's script help contract remains intact. The recovery logic from the previous reviewed head is unchanged.
The 13-module Python run passed 567 tests with six platform skips. All five required portable checks passed in order, plugin build and manifest formatting passed, and the standalone-build/installed-plugin integrity tests passed. Full hosted CI is pending on this head.
The reason will be displayed to describe this comment to others. Learn more.
Thanks—no blocking findings in this re-review. Accepted worker coverage, receipt ownership and deferred links survive publication and frozen recovery. The previous association/precedence cases and the helper packaging/help issues are fixed.
Fresh validation passed 146 focused Python tests, three MCP-to-Python cases, nine publication/replay scenarios using the fresh bundle, 39 isolated startup/help checks, builds, types, formatting and all five portable checks. Another 204 unchanged TypeScript cases were reused after verifying input identity. Full local store/deadline integration remains limited by the existing filesystem ownership guard; hosted CI is still running.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
alandelong-oai
left a comment
Summary
Retain accepted worker coverage and its original evidence when publishing or recovering Deep Scan results.
Changes
Testing
Full hosted CI passed on
4fce8e3bf574f08ec847b0783834589bee92dbe9: 48 jobs succeeded and two conditional jobs were skipped. Codex and Copilot completed current-head reviews, all review discussions are resolved, and maintainer approval is present.Risk and rollout
Coverage recovery determines which saved review records and receipts appear in a Deep report. Accepted result snapshots remain authoritative; recovery tests verify that original artifacts and parent associations are preserved. Full passing hosted CI and completed reviews remain required before merge.
Public disclosure review