What problem does it solve?
When an AI agent states a wrong fact, simply noting the correction leaves the bad data in place to resurface later. This Skill traces every user-reported factual error back to its true source β a brain page, memory file, rendered identity file, facts table row, or an LLM hallucination β fixes it at the correct surface, and checks for propagation.
Core Features & Use Cases
- Structured root-cause chain: Searches the gbrain knowledge base, memory files, SOUL.md/USER.md, and the facts table in order to locate the contamination source before fixing anything.
- Seven error classifications: Distinguishes BRAIN_ERROR, MEMORY_ERROR, SOUL_USER_ERROR, FACTS_TABLE_ERROR, HALLUCINATION, STALE_DATA, and CROSS_CONTAMINATION, each with its own fix procedure (page edit + commit + re-sync, forget/remember, answer-bank fix + re-render).
- Propagation checking and severity tiers: Greps the whole brain for other instances of the wrong fact and prioritizes identity-level errors (S1) over minor factual ones (S4).
- Use Case: A user says "That's wrong β Alice never worked at WidgetCo." The Skill searches the brain, finds the false claim on Alice's page, edits it with a dated source note, commits, re-syncs, checks for spread, and reports the root cause.
Quick Start
When I tell you a fact you stated is wrong, run the correction pipeline to root-cause the error, fix it at the source, and report what changed.
