debug
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates by reading and analyzing external data from log files and SQLite databases, which represents a surface for indirect prompt injection attacks.
- Ingestion points: The skill reads content from log files at
~/.humanlayer/logs/*.logand queries the SQLite database at~/.humanlayer/daemon.db(specifically thesessionsandconversation_eventstables). - Boundary markers: The instructions do not specify any delimiters or safety markers to help the agent distinguish between its instructions and the data being analyzed.
- Capability inventory: The skill utilizes
sqlite3,git status,git log,git diff,ls, andps. It does not contain network exfiltration or file-writing capabilities in the provided instructions. - Sanitization: There is no evidence of content sanitization or filtering of the log or database data before it is processed by the agent.
Audit Metadata