| name | audit-problem-corpus |
| description | Use when scanning `problem/normal` for duplicate titles, sparse testcases, missing files, or uneven code coverage across many generated bundles before cleanup or migration work. |
Audit Problem Corpus
Use this skill for a broad health pass across the generated problem library. It is best before large
cleanup work, parser migrations, or reporting on corpus quality.
Quick Start
- Default scan:
python .agents/skills/audit-problem-corpus/scripts/audit_problem_corpus.py problem/normal
- Stricter testcase threshold:
python .agents/skills/audit-problem-corpus/scripts/audit_problem_corpus.py problem/normal --min-testcases 10
- JSON output:
python .agents/skills/audit-problem-corpus/scripts/audit_problem_corpus.py problem/normal --json
Workflow
- Run the audit across
problem/normal.
- Group the findings by issue type instead of fixing folders in random order.
- If many folders share the same defect pattern, trace it back to prompt or parser behavior.
- Use
$validate-problem-bundle on any folder you are about to modify.
What The Audit Highlights
- Missing required files
- Zero or low testcase coverage
- Empty code directories
- Duplicate metadata titles
- Folder-level issue counts that help prioritize cleanup
Guardrails
- Report corpus-wide patterns before bulk rewriting legacy content.
- Keep audit output separate from repair work so the user can choose the cleanup scope.
- Treat this as structural triage, not proof that the generated algorithms are correct.
References
- Read
references/health-signals.md when you need triage rules.