skills/a-canary/arc-skills/hard-merge/Gen Agent Trust Hub

hard-merge

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to use the GitHub CLI (gh) for fetching diffs and checking branch status, along with grep for inspecting code content.
  • [EXTERNAL_DOWNLOADS]: Pull request data and repository metadata are retrieved from GitHub. This is standard behavior for a development-focused skill and utilizes a well-known service.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted content from pull request diffs.
  • Ingestion points: Diffs are retrieved via gh pr diff and stored locally for review in SKILL.md.
  • Boundary markers: The instructions enforce an adversarial mindset ("assume the diff is wrong") and use freshly spawned agent contexts to ensure independence.
  • Capability inventory: The agent can execute shell commands via gh, read files, and spawn sub-agents to perform reviews.
  • Sanitization: A critical safety requirement is included where the Author agent must manually verify every claim made by the sub-agent against the actual source code before merging.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 08:52 PM
Security Audit — agent-trust-hub — hard-merge