deep-grill
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process user-supplied "raw ideas," "formed plans," or "implied approaches," which are untrusted data sources.
- Ingestion points: Untrusted data enters the agent context through user input provided during "discovery mode" or "audit mode" as defined in
SKILL.md. - Boundary markers: The instructions do not specify any delimiters (such as triple quotes or specific tags) or explicit safety warnings to the agent to disregard instructions embedded within the user-provided ideas.
- Capability inventory: The skill directs the agent to "Investigate factual branches with the available environment, sources, tests, and bounded experiments." Across all files, this gives the agent broad authority to utilize its available toolset (e.g., shell access, file operations, or network tools) based on content derived from the user input.
- Sanitization: There are no mechanisms described or implemented to sanitize or validate the user-provided content before the agent performs its investigation or experiments.
Audit Metadata