The Agent Skills Directory

[SAFE]: The skill implements a robust 'live-truth' verification protocol, requiring the agent to cross-reference any recalled data against the actual project state (e.g., state.json, history.jsonl, or the live filesystem) before acting on it. This mitigates the risk of stale or manipulated data influencing the agent's behavior.
[SAFE]: The skill documentation and method include explicit prohibitions against persisting sensitive data, such as secrets, API tokens, credentials, or personally identifiable information (PII), within the memory files.
[SAFE]: The utility script scripts/checklist.mjs is a lightweight, zero-dependency Node.js tool that only performs local file read operations within the project directory. It does not use network resources, dynamic code execution (like eval), or external packages.
[SAFE]: The skill handles data ingestion (Category 8) from .refactor-chain/memory/sessions.jsonl with clear boundary markers and verification rules. Ingestion point: sessions.jsonl. Boundary markers: Explicit instructions to treat remembered facts as 'hints'. Capability inventory: Local file reading via scripts/checklist.mjs. Sanitization: Standard JSON parsing with error handling.

refactor-memory