grill
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill defines a structured interview process to resolve plan ambiguities. It is entirely instructional and does not include any scripts, network operations, or persistence mechanisms.
- [INDIRECT_PROMPT_INJECTION]: The skill involves a surface for indirect prompt injection by ingesting user-provided plans and referencing external project documents. It also allows for environment exploration such as filesystem lookups based on these inputs. The instructions attempt to mitigate risk by requiring human confirmation before plan execution.
- Ingestion points: User-provided plans, conversation context, and referenced files (e.g., specs, design docs, issues).
- Boundary markers: None defined in the skill instructions to separate untrusted content from the system prompt.
- Capability inventory: Filesystem and tool access via subagents to retrieve facts from the codebase.
- Sanitization: None defined.
Audit Metadata