error-analysislisted

# Error Analysis **Use this when:** Enough corrections have accumulated that you want to know which ones are a pattern rather than a one-off, and which of them are about the toolkit versus this project. **Don't use this when:** You want to record a correction (that happens automatically at `/document`), or you want to review code (use `/review`). ## What this is `/document` captures **open codes**: free-text notes, in your words, describing what went wrong each time you stepped in. This command does the **axial coding**: reading those open codes across the whole corpus, grouping them into categories, and ranking the categories by how often they occur. The split exists because a category invented at n=1 is a guess. You cannot see a pattern in the first instance of it, so naming patterns is deliberately deferred until there are instances to look at. ## CRITICAL RULES <rules> 1. **Never publish, and never send anything off this machine.** No HTML artifact, no hosted-page publish, never fed to `/ask-gpt`, `/ask-gemini`, or any other external model. This is a flat rule with no consent path and no "ask the user first" exception. The ledger is a private per-machine record of things the user said, and a consent prompt would be an invitation to leak it. If a future session finds a reason this rule should not apply, the reason is wrong. The no-ask publish rule in `.claude/rules/html-outputs.md` does not apply here: this skill renders nothing. 2. **Print only the shareable layer