[aw-failures] [Schema Consistency Checker] Recurring max-turns failures since 2026-05-03

[github-actions]

Summary

The Schema Consistency Checker (schema-consistency-checker.md) has been failing with error_max_turns on 5 of the last 7 scheduled runs. The agent exhausts all 60 turns before completing its analysis and calling a safe-output tool.

Failure Cluster

Date	Run ID	Conclusion	Turns	Duration
2026-05-08	§25539641121	❌ failure	61 (max 60)	7.7m
2026-05-07	§25479424655	❌ failure	61 (max 60)	9.5m
2026-05-06	§25419613447	✅ success	101	10.5m
2026-05-05	§25360601823	❌ failure	—	8.7m
2026-05-04	§25304247413	❌ failure	—	7.2m
2026-05-04	§25297162115	❌ failure	—	—
2026-05-03	§25271731933	❌ failure	—	7.5m

Root Cause

The workflow is configured with max-turns: 60 in schema-consistency-checker.md (line 13). The Claude agent performs a comprehensive schema diff analysis across main_workflow_schema.json, pkg/parser/*.go, pkg/workflow/*.go, and documentation files. With the current schema complexity (50+ top-level fields), 60 turns are insufficient to complete the analysis, generate findings, and call create_discussion.

The May 6 success completed in 101 turns—well above the 60-turn cap—suggesting the limit was not enforced or the strategy cache significantly reduced turns on that run.

Evidence: Max-turns exit from run 25539641121

{
  "type": "result",
  "subtype": "error_max_turns",
  "is_error": true,
  "num_turns": 61,
  "terminal_reason": "max_turns",
  "errors": ["Reached maximum number of turns (60)"]
}

The safe_outputs job was then skipped because detection was skipped (which skips when agent fails), causing the workflow to conclude as failure.

Cascade failure chain

agent: failure (exit code 1, max_turns)
detection: skipped (depends on agent success)
safe_outputs: skipped (requires detection == success)
conclusion: success (always runs)
→ workflow conclusion: failure

Proposed Fix

Increase max-turns in schema-consistency-checker.md from 60 to 120. The May 6 success used 101 turns, so 120 provides a comfortable buffer. The workflow already has a 30-minute timeout, so a higher turn limit won't increase wall-clock cost materially.

-  max-turns: 60
+  max-turns: 120

Alternatively, the workflow prompt could be tightened to front-load the most critical checks and call create_discussion earlier with partial findings if turns are running low.

Success Criteria

• Schema Consistency Checker completes without error_max_turns for 5 consecutive daily runs
• create_discussion is called successfully before the turn limit

References:

• §25539641121 — today's failing run
• §25479424655 — May 7 failing run
• §25419613447 — May 6 successful run (101 turns)

Generated by [aw] Failure Investigator (6h) · ● 628.6K · ◷

• expires on May 15, 2026, 7:22 AM UTC