request-adversarial-review
Pass
Audited by Gen Agent Trust Hub on Jul 23, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the calling agent to execute a command using the
agentCLI (e.g.,agent <workdir> "/adversarial-review ..."). This is a functional requirement to launch a sub-agent for the review process. - [PROMPT_INJECTION]: The workflow involves an agent reading and potentially acting on findings from a sub-agent that has reviewed arbitrary files in a worktree. This establishes a surface for indirect prompt injection if the files being reviewed contain adversarial instructions. However, the skill explicitly includes a directive to 'Read the artifact before acting on findings,' which encourages validation of the review output.
Audit Metadata