blast-radius
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process external, untrusted data such as code diffs, pull requests, and commit histories to determine the impact of changes. This creates an attack surface where instructions embedded in code comments or metadata could attempt to influence the agent's analysis or actions.
- Ingestion points: Step 1 of the instructions specifies reading diffs, symbols, PRs, and commits.
- Boundary markers: The instructions lack explicit boundary markers or directives to treat external code as untrusted data.
- Capability inventory: The agent is authorized to read local source files and execute scripts to verify its findings.
- Sanitization: No sanitization or filtering logic is prescribed for the ingested code content.
- [DYNAMIC_EXECUTION]: The skill requires the agent to generate and execute local scripts or tests to provide proof of code safety.
- Evidence: Step 5 ("Write a script or test that runs the real code, run it") and the safety fact hierarchy (Level 4: "You ran it. A script or test that calls the real code") mandate the runtime generation and execution of code.
- Logic: While this is intended for legitimate testing, it involves the dynamic assembly and execution of code based on external inputs.
Audit Metadata