agent-creation
Pass
Audited by Gen Agent Trust Hub on Mar 5, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by defining agents that ingest data from external sources or local files. 1. Ingestion points: Tools like fetch and readFile are used in agent templates. 2. Boundary markers: The templates do not use delimiters to isolate untrusted content. 3. Capability inventory: Created agents are assigned high-privilege tools such as runInTerminal and editFiles. 4. Sanitization: No input validation is recommended.
- [COMMAND_EXECUTION]: The instructions encourage the use of the runInTerminal tool, which allows agents to execute shell commands in the local environment.
- [EXTERNAL_DOWNLOADS]: The skill suggests including the fetch tool in agent configurations, which allows the retrieval of content from arbitrary URLs.
Audit Metadata