feature
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes external inputs to drive its development workflow.
- Ingestion points: The skill ingests user goals during the 'Grill' phase (Step 1) and processes external PR review findings and check failures (Steps 6 and 7).
- Boundary markers: The skill requires an explicit manual approval gate ('Iterate to approval' in Step 4) where the user must review and approve the design plan before any code is built or executed.
- Capability inventory: The skill possesses the authority to write to the filesystem (SKILL.md), execute local build and test commands (Step 5), and interact with remote repositories using the GitHub CLI (Step 7).
- Sanitization: No specific sanitization or filtering of external review comments or PR feedback is defined in the workflow.
- [DYNAMIC_EXECUTION]: The skill performs dynamic execution of local code as part of its verification process.
- Evidence: Step 5 explicitly requires the agent to "run build and tests" and "run the deployed check on any local or preview environment." This involves executing code that may have been generated or modified by the agent based on the provided goal.
Audit Metadata