errors/ stores recurring failures in Agent Pitbook itself.
Current status: the schema is ready and error-book.yaml has no entries yet (entries: []). Entries are added when a recurring Pitbook-process failure is actually observed, not speculatively.
This is inspired by later LLM-Wiki work that treats structural and semantic errors as persistent correction targets.
Examples:
- generated pit lacks source evidence
- duplicate pit created instead of updating canonical record
- stale workaround remains marked verified
- adapter suggests unsafe command
- search misses a known pit
- claim contradicts a newer source