pstack-blast-radius
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to read and analyze arbitrary code changes, diffs, and project artifacts (SKILL.md, Steps 1-3). This provides an ingestion point for untrusted data that could contain malicious instructions designed to influence the agent's reporting or proof generation.
- Ingestion points: The skill reads diffs, source files, library code, and API returns (SKILL.md, Step 3).
- Boundary markers: No explicit boundary markers or 'ignore embedded instructions' warnings are provided when processing external files.
- Capability inventory: The skill calls for executing code, running grep, and searching symbols (SKILL.md, Step 5).
- Sanitization: The skill mentions a '/pstack-unslop' tool to strip private information but does not detail sanitization of incoming instructions from the code being analyzed.
- [EXTERNAL_DOWNLOADS]: The skill references source code and plugins from a repository on GitHub (UPSTREAM.md). These references target a well-known service (GitHub) and are documented for transparency and maintenance, which is a safe practice.
- [COMMAND_EXECUTION]: The workflow encourages the agent to write and run scripts or tests against the real codebase to prove safety assumptions (SKILL.md, Step 5). While this is the primary purpose of the skill for verification, it involves executing potentially complex code generated or interpreted by the agent.
Audit Metadata