memory-hygiene
Pass
Audited by Gen Agent Trust Hub on Oct 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes instructions to use a shell command (
rm -rf) targeting the local directory~/.clawdbot/memory/lancedb/to manually wipe the agent's vector memory database. - [PERSISTENCE]: The skill documents the use of a
crontool to schedule a recurring monthly task. This task automates the process of clearing and rebuilding the agent's memory, ensuring the maintenance behavior persists across sessions. - [INDIRECT_PROMPT_INJECTION]: The workflow involves parsing a local file (
MEMORY.md) to reseed the vector database. This creates a surface where instructions embedded in the file could be stored as memories and influence the agent's behavior during future recalls. 1. Ingestion points: Parsing ofMEMORY.mdfor memory reseeding inSKILL.md. 2. Boundary markers: None present in the described maintenance workflow. 3. Capability inventory:rm -rf(shell),crontool,memory_storetool. 4. Sanitization: No validation or filtering is specified for the content read from the local file.
Audit Metadata