learn
Pass
Audited by Gen Agent Trust Hub on Mar 5, 2026
Risk Level: SAFEPROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it is designed to ingest and summarize untrusted data from the current session.
- Ingestion points: The 'Analysis Process' section instructs the agent to review all files read/modified and the entire session history (SKILL.md).
- Boundary markers: There are no defined delimiters or instructions to ignore malicious commands embedded within the analyzed files.
- Capability inventory: The skill utilizes the 'Edit' tool to write extracted content into the local filesystem (CLAUDE.md).
- Sanitization: No sanitization or validation of the discovered 'strategic knowledge' is performed before it is suggested for storage.
- [CREDENTIALS_UNSAFE]: The skill encourages the capture of sensitive operational data which may lead to credential exposure in documentation.
- Evidence: The 'What Qualifies for Local CLAUDE.md' section explicitly suggests including 'Database locations and connection details per environment'. If the agent extracts active connection strings containing passwords from configuration files, they would be persisted in a plain-text Markdown file.
Audit Metadata