methodology-pattern-extractor
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from meeting logs and session transcripts to extract collaboration patterns, creating a vulnerability surface where malicious instructions in the logs could influence the agent.
- Ingestion points: The agent is instructed to scan all files in
meetings/*.mdand search for session JSONL files in~/.claude/projects/*/. - Boundary markers: No specific boundary markers or instructions to ignore embedded commands are provided to the agent when processing these external files.
- Capability inventory: The skill possesses the ability to read files from the
meetings/directory and internal Claude project directories, and append generated content todocs/methodology.md. - Sanitization: There is no mention of sanitizing or filtering the content extracted from meeting logs before it is processed or written to the methodology document.
- [DATA_EXFILTRATION]: The skill procedure involves accessing internal Claude project session logs stored at
~/.claude/projects/*/<sessionId>.jsonl. While these files are accessed locally to enrich methodology patterns with specific details, they contain the history of interactions between the user and the agent.
Audit Metadata