audit-context-building
Pass
Audited by Gen Agent Trust Hub on Apr 13, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is designed for professional security analysis and explicitly limits its scope to context-building, which prevents the AI from rushing to conclusions or generating hallucinated vulnerabilities.
- [PROMPT_INJECTION]: The instructions are task-oriented and focus on improving analysis depth and accuracy. There are no patterns suggesting attempts to bypass safety filters or override system instructions. While the skill processes untrusted code (ingestion points: Grep, Glob, Read in function-analyzer.md), it incorporates strong boundary markers (structured sections in SKILL.md and OUTPUT_REQUIREMENTS.md), a limited capability inventory (filesystem read only), and process-based sanitization (mandatory line citations and completeness checks) to mitigate indirect injection risks.
- [COMMAND_EXECUTION]: The subagent uses standard filesystem tools (Read, Grep, Glob) required for its function. These are used in a controlled manner to analyze provided source code and do not involve arbitrary shell command execution or system modification.
Audit Metadata