evaluate-skill

Pass

Audited by Gen Agent Trust Hub on Mar 9, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes external data (SKILL.md and evals.json files) and interpolates them into the context of 'general-purpose' and 'eval-grader' subagents. If a skill being evaluated contains malicious instructions, it could attempt to influence the grader or the evaluation environment.
  • Ingestion points: The skill reads SKILL.md (Step 1, Step 3) and evals.json (Step 3) from arbitrary plugin paths provided in arguments.
  • Boundary markers: The execution logic does not specify the use of clear delimiters or 'ignore embedded instructions' blocks when passing the content of the tested skill to subagents.
  • Capability inventory: The skill uses Bash (ls, cat, jq, mkdir, find), Read, Write, and Task (subagent spawning) across its execution steps.
  • Sanitization: No explicit sanitization or validation of the content of the skill files is mentioned before they are processed by the LLM or used in the evaluation pipeline.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 9, 2026, 08:11 PM