blast-radius
Pass
Audited by Gen Agent Trust Hub on Aug 9, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted external code changes (diffs, pull requests, and commits), which serves as a vector for indirect prompt injection. Malicious instructions within the code could manipulate agent logic or bypass safety constraints during analysis.\n
- Ingestion points:
SKILL.mdStep 1 (reading diffs, pull requests, and commits).\n - Boundary markers: Absent; no delimiters are specified to separate analysis instructions from external code content.\n
- Capability inventory: Access to file system, network fetching, and command execution.\n
- Sanitization: Absent; no explicit instructions for filtering analyzed code content.\n- [COMMAND_EXECUTION]: The skill instructs the agent to verify safety by writing and running scripts that execute the "real code" being analyzed (Steps 4 and 5), creating a risk if that code contains malicious logic that triggers during execution.\n- [DATA_EXFILTRATION]: The instruction to "strip anything private before it goes anywhere public" indicates a workflow where analysis results are shared externally, posing a risk of sensitive data exposure if the agent fails to correctly identify and remove private information.
Audit Metadata