code-hidden-failures
Warn
Audited by Gen Agent Trust Hub on Jul 31, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill employs dynamic context injection (!command syntax) in the SKILL.md file to automatically execute shell commands like 'find' when the skill is loaded by the agent.
- [DATA_EXFILTRATION]: One of the dynamic context commands specifically searches for sensitive environment files (.env*). While it only lists the files rather than reading their content, this structural information is automatically disclosed to the model context.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted source code through the PATH argument. Malicious instructions within scanned files could potentially influence the agent's behavior as the skill lacks explicit boundary markers to isolate processed code from instructions.
- [COMMAND_EXECUTION]: The skill invokes tools like 'ast-grep' and custom bash scripts (scripts/scan-shell.sh) to analyze code, which involves executing logic against potentially untrusted user-supplied content.
Audit Metadata