skill-optimizer

Pass

Audited by Gen Agent Trust Hub on Jun 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes the gbrain skillopt CLI tool to run benchmarks and optimize instruction sets.
  • [PROMPT_INJECTION]: The skill optimization process involves processing external benchmark data which could present an indirect prompt injection surface.
  • Ingestion points: Data is read from skills//skillopt-benchmark.jsonl and routing-eval.jsonl files.
  • Boundary markers: The system requires the manual removal of a # BOOTSTRAP_PENDING_REVIEW sentinel line before optimization can proceed, ensuring human oversight of generated tasks.
  • Capability inventory: The skill performs rolling edits to SKILL.md and related history files via the gbrain CLI.
  • Sanitization: It implements validation gating using median-of-3 scoring and explicitly forbids mutation of the skill's routing frontmatter.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 13, 2026, 04:43 AM
Security Audit — agent-trust-hub — skill-optimizer