research-worker
Pass
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill explicitly defines itself as a "read-only investigation agent" and includes a clear list of prohibited actions, such as preventing file creation, modification, deletion, and the execution of destructive commands.
- [INDIRECT_PROMPT_INJECTION]: While the skill ingests external data from web searches and package registries, which represents a potential surface for indirect prompt injection, the risk is mitigated by the agent's strictly limited capabilities.
- Ingestion points: Processes data from the local codebase (via grep and glob), external documentation (via Context7), and web search results.
- Boundary markers: The skill mandates a structured "Research Report" output contract, although it does not specify character-level delimiters for external data.
- Capability inventory: Restricted to read-only operations. It is explicitly barred from file system writes, diff generation, and package installation.
- Sanitization: No specific sanitization or filtering of external content is described in the instructions.
Audit Metadata