how
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external codebase files and user-provided questions through a sub-agent architecture (explorers and explainers). This creates a surface for indirect prompt injection where adversarial content within the codebase could attempt to influence the agent's behavior. However, the skill explicitly enforces a
read-onlyconstraint for all execution lanes, effectively neutralizing the risk of unauthorized file modifications or command execution resulting from such injections. - Ingestion points: Codebase files (accessed via Read, Grep, and Glob tools) and user-supplied
{QUESTION}strings interpolated into prompt templates. - Boundary markers: Uses Markdown block delimiters and templated prompt structures to separate instructions from external data.
- Capability inventory: Limited to read-only file system operations (Read, Grep, Glob) and synthesis of findings.
- Sanitization: Relies on structured prompt synthesis and read-only environment constraints.
Audit Metadata