The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the local repository via git status and git diff to generate commit messages, which presents a surface for indirect prompt injection from untrusted code content. However, the skill includes robust mitigations: ingestion points are limited to git metadata and stats in SKILL.md Step 1; the capability inventory includes standard git operations like add, commit, and push; boundary markers are established through a required summary report and user confirmation on safety failures in Step 3; and sanitization is performed using a shell heredoc (cat <<'EOF') in Step 6 to prevent the generated message from being interpreted as shell commands.
[SAFE]: The skill incorporates a mandatory safety scan in Step 2 of SKILL.md that specifically detects and blocks the pushing of sensitive files such as .env, private keys (.pem, id_rsa), and real API key strings, while allowing common placeholders. It also follows secure synchronization practices by fetching and rebasing before pushing changes.

auto-push