command-research
Pass
Audited by Gen Agent Trust Hub on Aug 5, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill defines a research workflow that utilizes a sub-agent to perform web searches and fetches (WebSearch, WebFetch). This introduces a risk of indirect prompt injection, where malicious instructions embedded in external web content could attempt to influence the agent's logic or output.
- Ingestion points: External data enters the agent context via the
WebSearchandWebFetchtools used by theapp-researchersub-agent as described inSKILL.md. - Boundary markers: The skill documentation notes that the sub-agent runs in an "isolated context," which is a positive security design, but it lacks explicit prompt delimiters or instructions to ignore instructions found within the fetched data.
- Capability inventory: The sub-agent possesses network access capabilities; the main thread possesses file-writing capabilities to the
research/directory. - Sanitization: No evidence of content sanitization, validation, or filtering of the fetched web data was found in the provided instructions.
Audit Metadata