🐛🧑‍💻 - Progress partition rejects per file implemented text with schema string long and no file identity #830

Is there an existing issue for this?

• I have searched the existing issues.

The exact diagnostic search for schema-string-long returned no existing issue. Related #828 concerns scope-clause-limit; #829 concerns assessment-text-limit at $.requirements. This observation is a different phase and output field. Please consolidate if investigation establishes a common cause.

Which actions are affected?

Commits

Which platforms are affected?

Linux

Description

A normal push to the existing Action-owned Domus feature branch reaches bounded progress partition assessment. Part 7 of 9 completes agent execution, then Copilot rejects its returned per-file implemented description with schema-string-long at $.files[0].implemented. It fails progress verification and retains the previous progress state. The diagnostic contains no requirement/source identity and does not identify which file or what generated length exceeded which bound.

Expected: preserve fail-closed output validation and old progress, but obtain an output satisfying the unchanged contract or provide safely bounded diagnostics identifying the failing part/file and expected/observed bound. A legitimate consumer should not need to weaken scope, coverage, acceptance or output validation.

Actual: phase partition / reason schema-string-long / field $.files[0].implemented / requirement none / source none. No final percentage is published. The previous scope and assessment failures are not assumed resolved merely because this run reached another phase. The raw model output is not visible in the inspected logs, so its contents, exact length and root cause remain unknown.

Reproducing the issue

1. Consumer: ✨🧑‍💻 - Implement Home Assistant identity and SQLite platform persistence domus#6 and Implement HA identity, SQLite persistence and verified restoration domus#7.
2. Existing Action-owned branch: feature/6-implement-home-assistant-identity-and-sqlite-platform-persistence; comparison base 0012a172492515e282be09c37921f4014a451e64; exact source head ead7dde44f57d32d5a42797a811e798d92b36a89.
3. Push-triggered run: https://github.com/vypdev/domus/actions/runs/41227309965; job https://github.com/vypdev/domus/actions/runs/41227309965/job/124614034440; generated failing verification check https://github.com/vypdev/domus/runs/124614821407.
4. Evidence collection completes: 204 required paths, 9 complete bounded source parts, 55 recovered files, 916285 bytes, 78 requests, 188 tree entries in one frame.
5. Scope execution is admitted at 18:02:19.9691591Z and completes at 18:02:35.9141049Z. Parts 1 through 6 finish; part 7 is admitted at 18:04:00.0863036Z and completes at 18:04:11.8700218Z.
6. At 18:04:11.8735892Z, validation rejects the partition result. Check publication retains the exact same diagnostic/reference. This was an ordinary implementation/test commit, not a manual job retry.
7. Independent project CI Check for this exact head passes: https://github.com/vypdev/domus/actions/runs/41227316511. A second ordinary ready-for-review event also runs CI; this report does not conflate its status with the completed synchronize run.

This is an observed exact-run reproduction, not a minimized deterministic fixture or an alleged queue-delay defect.

copilot Version

vypdev/copilot@v3 resolved to immutable revision 3e7ceea. Agent runtime: codex codex-cli 0.166.0; configured model gpt-6-luna. No configuration or model change was made to work around the result.

Relevant Log Output

2026-10-31T18:01:07.7703369Z Download action repository 'vypdev/copilot@v3' (SHA:3e7ceea08e6131360e026937d1c0cc1b120472e0)
2026-10-31T18:02:19.8737474Z Complete progress evidence: 204 required paths; recovery usage {"frames":1,"current":{"files":55,"bytes":916285,"requests":78,"treeEntries":188},"total":{"files":55,"bytes":916285,"requests":78,"treeEntries":188}}; 9 source parts. Previous progress remains until final validation.
2026-10-31T18:02:35.9901237Z Progress comparison requires 9 complete bounded parts; no percentage is published until all parts and final assessment finish.
2026-10-31T18:04:00.0602549Z Assessing progress evidence part 7 of 9, round 1 of 1.
2026-10-31T18:04:00.0863036Z Agent execution admitted (codex codex-cli 0.166.0).
2026-10-31T18:04:11.8700218Z Agent execution completed.
2026-10-31T18:04:11.8735892Z Progress partition result was rejected (schema-string-long; field $.files[0].implemented; requirement none; source none). Existing progress was retained.
Error code: agent.failed
Details: phase partition, reason schema-string-long, field $.files[0].implemented, requirement none, original issue line none.
Reference: 87231e03-e663-4ba5-b142-ecc326db50e8

Additional context and comments

• This real product change fixes missing membership explanations, independently tests all role/permission pairs and traces 61 identity plus 18 restoration cases to exact semantic tests or bounded observations.
• Backend: 1387 tests, 4692 supported statements and 1304 branches at 100%. Tooling: 66 tests, 73 statements and 34 branches independently at 100%. Architecture: 41 negative fixtures; actual 103-file/542-dependency application graph with all three contracts kept. Node: 11 policy tests. Both actual native image jobs pass in the completed exact-head CI run.
• Original twenty-section SDD and catalog validators pass without changes to their source. All 29 protected installed Copilot resources remain unchanged. No requirement, output bound, coverage scope or verification gate was relaxed.
• The PR was subsequently marked ready to obtain complete Bugbot review; earlier skipped draft checks are not counted as acceptance. It remains unmerged pending full current-head review and coverage. The progress partition failure is distinct from that review lifecycle.
• Useful regression coverage could include long but legitimate per-file implemented summaries, exact bounded partition provenance, safely attributable length diagnostics and bounded recovery preserving all source parts and old progress. This is a request to investigate, not a prescription to raise limits or silently truncate output.
• No credentials, owner HA account IDs, session cookies, private Ingress paths or raw identity payloads are included. Product work continues independently without editing Copilot.