adversarial-diff-review
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions (SKILL.md) explicitly direct the agent to execute code within the repository to verify its analysis, including running CLI tools, loading modules, and executing test suites. This poses a risk if the repository or the diff under review contains malicious code designed to execute during the review process.
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes untrusted content from task definitions and code patches. It lacks defined boundary markers or sanitization logic to prevent the agent from following instructions embedded in the analyzed data.
- Ingestion points: Reads task definitions, Git status, diffs, and worktree changes into the agent's context (SKILL.md).
- Boundary markers: The skill suggests 'establishing the boundary' conceptually but does not implement technical delimiters or safety prompts to isolate untrusted data.
- Capability inventory: The agent is authorized to execute shell commands, CLI tools, and test suites as part of the review process (SKILL.md).
- Sanitization: No evidence of input sanitization or filtering for the data being analyzed.
Audit Metadata