Raise max-patch-size for Deep Report's repo-memory push — it just dropped a full analysis cycle's output on the floor
Affected workflow and run
- Workflow:
Deep Report (.github/workflows/deep-report.lock.yml) - Failed run: §32252000817, job
push_repo_memory, step Push repo-memory changes (default) - Isolated occurrence — the prior 9 scheduled Deep Report runs (back to 2026-08-17) all succeeded. This is a fresh regression, not a chronic pattern.
Probable root cause
The agent legitimately generated and staged 6 changed repo-memory files (extracted-tasks.json/.md, flagged_items.md, known_patterns.md, last_analysis_timestamp.md, processed-discussions.json/.md, trend_data.md) — a normal, successful analysis cycle. The combined git diff was 13 KB (12,767 bytes), exceeding the configured limit of 10 KB (12,288 bytes effective with 20% overhead) by ~500 bytes. The push step hard-fails on this check, so all 6 files' updates are discarded — the agent's work for this cycle is lost, not just delayed.
Proposed remediation
- Raise
max-patch-size for the Deep Report workflow's repo-memory push (or the framework default) to a threshold with headroom above typical per-cycle diff size — 13 KB exceeding a 10 KB config by only ~25% suggests the limit was set too tight for this workflow's actual output volume, not that the diff is anomalously large. - Alternatively, have Deep Report trim
known_patterns.md/trend_data.md growth (append-only files that grow every cycle) or split the repo-memory commit into per-file commits so one oversized file doesn't block the rest. - Add a warning threshold below the hard limit so an oversized diff surfaces before it silently blocks a full cycle's output.
Success criteria / verification
- Deep Report's next 3 scheduled runs push repo-memory changes successfully without hitting the patch-size limit.
- If the fix is a threshold increase, confirm the new limit still catches genuinely anomalous (e.g., accidental bulk-file) diffs — this isn't a request to remove the guard, just to size it correctly.
Parent report: #53933
Analyzed run: §32252000817
Related to #53933
Generated by 🔍 [aw] Failure Investigator (6h) · agent · 107.4 AIC · ⌖ 7.82 AIC · ⊞ 5.9K · ◷
Raise
max-patch-sizefor Deep Report's repo-memory push — it just dropped a full analysis cycle's output on the floorAffected workflow and run
Deep Report(.github/workflows/deep-report.lock.yml)push_repo_memory, step Push repo-memory changes (default)Probable root cause
The agent legitimately generated and staged 6 changed repo-memory files (
extracted-tasks.json/.md,flagged_items.md,known_patterns.md,last_analysis_timestamp.md,processed-discussions.json/.md,trend_data.md) — a normal, successful analysis cycle. The combined git diff was 13 KB (12,767 bytes), exceeding the configured limit of 10 KB (12,288 bytes effective with 20% overhead) by ~500 bytes. The push step hard-fails on this check, so all 6 files' updates are discarded — the agent's work for this cycle is lost, not just delayed.Proposed remediation
max-patch-sizefor the Deep Report workflow's repo-memory push (or the framework default) to a threshold with headroom above typical per-cycle diff size — 13 KB exceeding a 10 KB config by only ~25% suggests the limit was set too tight for this workflow's actual output volume, not that the diff is anomalously large.known_patterns.md/trend_data.mdgrowth (append-only files that grow every cycle) or split the repo-memory commit into per-file commits so one oversized file doesn't block the rest.Success criteria / verification
Parent report: #53933
Analyzed run: §32252000817
Related to #53933