← ClaudeAtlas

ai-agent-error-memory-registrylisted

Use when an agent or team keeps repeating the same class of mistake, when you want a durable record of failures instead of one-off apologies, or when setting up a project's error memory. Defines an append-only scar registry — symptom, consequence, guard, rule, and the voice that produced it — plus the ladder from remembered rule to enforced mechanism, and the recidivism counter that tells you when a rule is unenforceable rather than merely forgotten.
nowespojrzenie/HEXAGRAM · ★ 1 · AI & Automation · score 80
Install: claude install-skill nowespojrzenie/HEXAGRAM
# Scar Registry **The problem this solves:** an agent apologises, corrects itself, and makes the same mistake nine days later. Not because it is careless — because the correction lived in a conversation that ended. Nothing carried. A scar registry is error memory that outlives the session. Not a changelog (that records what changed), not a postmortem archive (that records incidents). A registry of **failure modes**, written so the next occurrence is cheaper than the last. The name matters. A scar is not a wound and not shame — it is tissue that grew back tougher, and it stays visible. ## When to Use Load this when: - the same class of error has now happened twice - you are setting up how a project or agent will remember its own failures - a postmortem is about to be written and you want it to change behaviour, not just describe the past - someone says: *you did this before · we already fixed this · why does this keep happening* Do **not** load this for one-off user-facing errors that need an immediate fix and nothing more. Not every mistake deserves an entry; a registry that records everything gets read by nobody. ## The entry Five fields. Fewer, and the entry cannot act; more, and it will not be written. | field | question it answers | failure if skipped | |---|---|---| | **SYMPTOM** | what was observably wrong — concrete, quotable | entry becomes a mood | | **CONSEQUENCE** | what it cost, or would have cost | no one prioritises it | | **GUARD** | what would have c