grill-me
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user input during the 'Grill' phase and subsequently interpolates that information into 'briefs' for subagents in the 'Dispatch' phase. If a user provides malicious instructions during the interview, those instructions could be passed to subagents.
- Ingestion points: User responses to interview questions processed in Phase 1 and Phase 2 (SKILL.md).
- Boundary markers: The skill instructions lack explicit delimiters to separate user-provided content from agent instructions in the generated briefs.
- Capability inventory: The system is capable of writing code to the repository, performing git operations (git-worktree), and executing tests via the referenced 'dispatch' skill.
- Sanitization: The skill includes a 'verify-before-merge' protocol, requiring the agent to read subagent diffs against their briefs and run tests before merging, which acts as a manual verification layer to mitigate unauthorized actions.
Audit Metadata