knowledge-base-audit

name	knowledge-base-audit
description	Validate, maintain, and audit the longevity hallmarks evidence knowledge base. Use when Codex is asked to check data health, find broken references, inspect schema drift, verify candidate changes or evidence reviews, audit taxonomy coverage, prepare release readiness, or maintain repository quality.

Knowledge Base Audit

Workflow

1. Work from the repository root.
2. When running as the interactive coordinator, read AGENTS.md, then read plan.md, docs/audit-and-release.md, and relevant schemas.
3. Run the baseline verification command:

npm run verify:knowledge-base

4. Use rg and structured JSON tools to inspect references, record counts, and scope-specific consistency.

Audit Targets

Check for:

• JSON files without schema coverage.
• Records whose record_type does not match their collection.
• Duplicate IDs across canonical records.
• Broken references between sources, studies, findings, candidate changes, evidence reviews, sessions, agent runs, coverage assessments, and synthesis groups.
• Track IDs or hallmark IDs not present in taxonomies.
• Candidate changes missing required review lanes.
• in_review candidates that lack active evidence-review records for each required lane.
• Accepted or applied candidates that lack promotion metadata.
• Agent runs with canonical writes that lack a linked candidate change.
• Codex CLI worker agent runs whose execution metadata points to missing prompt, schema, or output paths.
• Evidence reviews that refer to missing candidate changes.
• Coverage assessments that cite missing sources or findings.
• Synthesis groups that claim pooling is allowed without required effect fields.
• Placeholder schemas that are being used for production-like records.
• Generated/export files that are stale or not reproducible.
• Extraction-grade records or exports whose provenance lacks source snapshot links.
• Deprecated non-agentic process vocabulary.

Minimal Commands

Use these first:

npm run verify:knowledge-base
find data research taxonomies schemas -name '*.json' | sort
rg -n '"record_type"|"source_id"|"source_ids"|"study_id"|"finding_ids"|"track_ids"|"hallmark_ids"' data research taxonomies

Prefer adding or tightening scripts when the same audit would otherwise be repeated through one-off agent checks.

Maintenance Rules

• Do not delete canonical records just to make an audit pass.
• Do not rewrite scientific content unless the task explicitly includes correction or repair.
• If a problem is structural, improve schemas or validation scripts.
• If a problem is scientific, create a candidate change or evidence review finding.
• Keep audit output concise: findings first, then validation status, then recommended next action.

Handoff

End with:

• validation result
• high-priority findings
• files changed, if any
• checks not yet automated