hillclimb

Pass

Audited by Gen Agent Trust Hub on Nov 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests results from external measurement commands and logs state to files, creating an indirect prompt injection surface.
  • Ingestion points: Results from user-provided baseline measurement commands are ingested into the agent context.
  • Boundary markers: The skill references 'benchmark-checklist' and 'principles' skills to guide manual evidence verification and labeling.
  • Capability inventory: The agent performs file writes to '.omc/state/' and executes git commits for 'verified wins'.
  • Sanitization: No explicit sanitization or strict schema validation is provided for the measurement data.
  • [NO_CODE]: The skill contains no executable scripts, binaries, or automated shell commands, consisting entirely of instructions and metadata.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 8, 2026, 07:21 PM