nuke-eval
Pass
Audited by Gen Agent Trust Hub on Aug 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill resolves and executes evaluation harnesses (such as promptfoo, braintrust, or custom scripts) to run test cases against target prompts and instruction blocks.
- [EXTERNAL_DOWNLOADS]: During the Phase 0 preflight resolution, the agent may attempt to fix environmental blockers by installing missing dependencies or binaries required for the evaluation tooling to function.
- [PROMPT_INJECTION]: The skill utilizes an adversarial testing methodology called 'Perturbation Checks' in Phase 3, where it intentionally degrades prompts by dropping constraints or inverting instructions to verify that the evaluation suite correctly identifies failures.
Audit Metadata