2. Content Cleanup Pass
Purpose: Rewrite fact titles and context for voice consistency, with audit-before and verify-after.
Prerequisites:
- Supabase credentials
- AI model API key
Cost / Duration: ~$0.004/fact | varies by volume
Prompt
Run a content cleanup pass to fix voice inconsistencies in fact titles and context.
Step 1 — Audit current state (read-only):
```bash
bun scripts/seed/cleanup-content.ts --audit
```
Step 2 — Run the cleanup with validation:
```bash
bun scripts/seed/cleanup-content.ts --fix --validate --concurrency 3
```
Adjust `--concurrency` based on your rate limits (default: 3).
Step 3 — Verify results:
```bash
bun scripts/seed/cleanup-content.ts --audit
```
Compare the before/after audit results. The cleanup should reduce voice
inconsistencies without changing factual content.
Verification
- Pre-cleanup audit captured baseline metrics
- Cleanup ran without errors
- Post-cleanup audit shows improved voice consistency
- No factual content was changed (titles/context only)
- Cost reported and within expected range
Related Prompts
- Audit Fact Quality — Broader corpus health check
- CQ-002 Compliance Patch — Fix challenge engagement language