[MonaDevAI]

Problem

Dashboard summaries retained short stage messages, not the actual bug-to-code-fix explanation. Retries and restarts could leave only statements such as "retained committed fix."

Change

Capture a bounded, sanitized engineering explanation independently of stage messages. Show the problem, expected/actual behavior, root cause, concrete code fix, relative changed files, commit reference, regression details, risks and workarounds in View summary. Preserve the explanation in history and provide it to editing-job continuation prompts.

Explicitly confirmed, exact-instance/repository/job local clients can add reviewed notes for an older stopped job. Notes never start execution, change validation results or grant permissions. Agent reports and reviewed notes are labeled separately; missing evidence remains unverified.

Validation

• Focused tests pass: eight unit cases and both summary browser cases.
• npm pack contains the required files; installed changed files match the committed source.
• Documentation is updated.
• No credentials, private endpoints, customer-specific notes or test data are included.

Final source: d659544952c8a8914e9588920814226a8a000c02. Lint, type-check, formatting, documentation, agent-rule governance, repository policy, self-healing verification and 492/492 unit tests pass. The final local npm run validate attempt exited 1: 47/48 browser tests passed, while the previously observed linked-application case exceeded its 30-second timeout under three workers. That exact case subsequently passed 1/1 with one worker, and the package dry-run passed separately with archive SHA1 3310779de9d7239542e142fbfd72cd1445b33cf0. The failed monolithic attempt is not reported as a pass.

All exact-head GitHub checks pass: test, windows-containment, analyze and CodeQL. Final automated review recommends approval with no open issues. No cases were skipped. Earlier intermediate runs were deliberately stopped when review fixes were added; an initial restart-persistence fixture failure was corrected by using a real Git fixture.

Installed locally before opening this PR. A live headless Edge check verified six detailed-summary sections, zero POSTs when opening the summary, preserved job history and no resumed held execution. Reviewed notes were independently checked in persistent history.

Addressed all seven review observations with regressions in existing tests: normalized relative paths, fresh-job evidence reset, executor settlement before note imports, queue serialization, existing runtime credential screening for every field/list item, intact rejection of malformed arrays/trailing payloads, and explicit persistence failure when the target falls outside the bounded history window.

Risk

Engineering explanations are recorded reports, not independent runner proof or a claim of complete regression coverage. Older jobs without a retained explanation remain explicitly incomplete unless reviewed notes are supplied. Imported notes reject browser Origins, active/unknown jobs, incorrect identities, malformed/oversized records and unsafe paths; persistence failure is explicit.

No dependencies or product-specific logic were added. The additive snapshot fields remain compatible with older jobs.

[Copilot]

🟡 Changes recommended

Critical issues allow restarted jobs to inherit stale evidence, unsafe paths to bypass validation, and unsettled executors to overwrite reviewed notes.

4 open issues

• Validate trimmed summary paths to prevent traversal · New
• Clear copied engineering summaries when creating a restarted job · New
• Reject reviewed notes while the job executor is still unsettled · New
• Include retained summaries in the public queue serializer · New

What changed in this PR

This PR adds detailed bug-to-code-fix records to FixLab’s dashboard summaries, separate from short stage messages, so engineering explanations remain available through continuation and history.

Changes:

• Capture bounded, sanitized engineering summaries and include them in editing-job continuation prompts.
• Add a local endpoint for importing confirmed reviewed notes for stopped jobs.
• Display detailed explanations, evidence and risks, with focused tests and documentation.

File	Description
test/​dashboard-server.test.js	Tests summary validation, persistence, reviewed notes and continuation prompts.
e2e/​dashboard-readiness.spec.js	Checks detailed summary rendering and missing-evidence messaging.
docs/​commands.md	Explains retained summaries and reviewed notes.
docs/​architecture.md	Documents summary records and local note-import constraints.
dashboard/​server.js	Adds summary parsing, serialization, prompt guidance and note imports.
dashboard/​public/​app.js	Displays engineering explanations, file references, regression details and risks.

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

Incomplete credential screening and incorrect splitting of non-object summaries can expose secrets through output logs, history, API responses, summary views, and continuation prompts.

2 open issues

• Screen summary fields for credentials before retaining them · New
• Reject whole arrays when splitting malformed summary lines · New

4 resolved since last review

• Reject reviewed notes while the job executor is still unsettled
• Clear copied engineering summaries when creating a restarted job
• Validate trimmed summary paths to prevent traversal
• Include retained summaries in the public queue serializer

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

The endpoint can report success even when persistence drops the target completed job, causing its reviewed notes to disappear after restart without warning.

No open issues 🎉

1 previously missed issue

In code that hasn't changed since last review

Report when the selected job is dropped from persisted history

dashboard/​server.js:4986

With a current job and 20 completed jobs, the oldest completed job is still visible and accepted here, but persistDashboardState() truncates the snapshots to 20 and drops that job without setting dashboardHistoryWarning. This endpoint consequently returns 200 even though the reviewed notes will disappear after restart. Check that the target survived the persistence window, or retain it explicitly, before reporting success.

2 resolved since last review

• Reject whole arrays when splitting malformed summary lines
• Screen summary fields for credentials before retaining them

_{Review effort: Balanced}

[MonaDevAI]

Addressed the final bounded-history observation in d659544. The stopped-job notes endpoint now reports HTTP 503 unless the selected job survived the persisted snapshot window, instead of falsely reporting durable success. Added a focused existing-suite regression with a visible completed job excluded by current/held snapshots; all eight focused cases pass. Full final validation is running again. Latest package is installed locally; job execution remains held.

[Copilot]

🟢 Approval recommended

The review identifies no active issues with the engineering-summary retention and dashboard changes.

No open issues 🎉

_{Review effort: Balanced}