fde-engagement
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements a comprehensive security architecture. It includes 'scripts/preflight-sources.mjs', a utility designed to scan untrusted input for prompt injections, credential harvesting attempts, and shell commands before the agent processes the content.
- [SAFE]: The 'Operating rules' and 'Evidence and safety' references provide explicit instructions for the agent to treat external source material as untrusted data, ignore embedded instructions, and require manual user authorization for any consequential actions, adhering to the principle of least privilege.
- [SAFE]: While the skill contains strings associated with prompt injection (e.g., 'SYSTEM OVERRIDE' in 'evals/files/injected-interview.txt'), these are explicitly identified as synthetic evaluation fixtures used to verify the agent's defensive capabilities in controlled tests.
- [SAFE]: The utility scripts included with the skill ('validate-case-file.mjs', 'render-readout.mjs', etc.) perform localized structural validation and content rendering using built-in Node.js modules without external dependencies or unauthorized network access.
Audit Metadata