memory-hygiene

Warn

Audited by Gen Agent Trust Hub on Feb 25, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides instructions to execute destructive shell commands (rm -rf ~/.clawdbot/memory/lancedb/) to wipe memory and commands to restart the gateway service. It also includes instructions to create a persistent monthly task via a cron tool to automate these actions.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) through its memory maintenance automation. 1. Ingestion points: Data is ingested from the MEMORY.md file during the reseed phase of the monthly maintenance job. 2. Boundary markers: No delimiters or instructions to ignore embedded commands are present when parsing the external file. 3. Capability inventory: The skill can execute filesystem commands (rm -rf), modify service configuration (gateway action=config.patch), and schedule persistent jobs (cron action=add). 4. Sanitization: The skill lacks logic to sanitize or validate content before storing it in vector memory or using it in automated jobs.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 25, 2026, 08:59 PM