evaluate-improve

Pass

Audited by Gen Agent Trust Hub on Apr 21, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes data from evaluation benchmarks (benchmark.json) which may contain untrusted content. This content is analyzed by a subagent to generate modifications for the skill's instructions or description.
  • Ingestion points: Reads evaluation results from <plugin-name>/skills/<skill-name>/eval-results/benchmark.json in Step 1.
  • Boundary markers: The instructions do not define clear delimiters or include warnings for the subagent to ignore instructions embedded within the benchmark data.
  • Capability inventory: The skill has the ability to modify local files using the Edit and Write tools, and can execute specific shell commands via Bash.
  • Sanitization: No sanitization or validation of the benchmark JSON content is performed before it is passed to the analysis subagent.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 21, 2026, 01:18 AM
Security Audit — agent-trust-hub — evaluate-improve