managing-git-workflow

Pass

Audited by Gen Agent Trust Hub on May 11, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a potential surface for indirect prompt injection as it ingests and processes untrusted data from git history and file changes to generate natural language outputs.\n
  • Ingestion points: The skill reads external data using git status, git log, and git diff as described in reference/commit.md, reference/pr.md, and reference/push.md.\n
  • Boundary markers: While the agent is instructed to match project styles and extract specific ticket patterns, the instructions do not include explicit boundary markers or directives to ignore instructions that might be embedded within the files or commit logs being analyzed.\n
  • Capability inventory: The skill has access to write-enabled tools including git commit, git push, and gh pr create.\n
  • Sanitization: The implementation consistently uses HEREDOC syntax (cat <<'EOF') within shell commands to prevent the generated message or PR content from being executed as shell code, which mitigates direct command injection risks at the execution layer.
Audit Metadata
Risk Level
SAFE
Analyzed
May 11, 2026, 03:02 AM
Security Audit — agent-trust-hub — managing-git-workflow