grill-me
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates by analyzing external plan files and GitHub issues to drive an interactive questioning session and verification pass.
- Ingestion points: Input data enters the agent context via user-supplied plan file paths and referenced GitHub issues (via URL or issue body).
- Boundary markers: The prompt instructions lack explicit structural delimiters or instructions to prevent the agent from executing commands or instructions embedded within the ingested plans.
- Capability inventory: The skill possesses code exploration capabilities (Grep, Glob, Read, Agent) and can write to local files or invoke external commands like
gh issue commentto post tracking updates. - Sanitization: No sanitization or escaping mechanisms are specified to handle untrusted input text before it is analyzed or integrated into the tracking comments.
Audit Metadata