hill-climb

Pass

Audited by Gen Agent Trust Hub on Nov 18, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code and UI data, which could potentially contain malicious instructions intended to manipulate the judging agents during the review rounds.
  • Ingestion points: The skill reads repository scoped files, untracked files, and UI screenshots.
  • Boundary markers: The Judge brief section in SKILL.md contains an explicit instruction for agents to disregard requests or claims about which version is better inside the bundle files.
  • Capability inventory: The skill utilizes git, gh, cp, rm, node, and repository-specific tools like pnpm.
  • Sanitization: The content is processed without explicit sanitization or filtering before being sent to council members.
  • [COMMAND_EXECUTION]: The skill executes a local Node.js script (scripts/build-runs.mjs) and common development tools (git, gh) to perform its optimization tasks and report management. It stores run records and generated reports in the user's home folder (~/.hill-climb/runs/).
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 18, 2026, 11:16 PM