The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill invokes pre-commit run in Step 3. This tool executes hooks defined in a repository's .pre-commit-config.yaml file. If the repository is untrusted or the configuration file has been compromised, it can lead to the execution of arbitrary scripts and binaries during the commit process.
[PROMPT_INJECTION]: The instructions contain a strong directive: 'IMPORTANT: Execute all steps continuously without pausing to ask for confirmation between steps.' This instruction specifically encourages high autonomy and bypasses standard human-in-the-loop safeguards for side-effect heavy operations like pushing code to remote repositories and creating PRs.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute several powerful CLI commands, including git push, gh pr create, and gh issue create. These commands modify remote repository states and create new resources on GitHub.
[DATA_EXFILTRATION]: The skill gathers repository metadata via dynamic context commands such as git remote -v, git log, and git diff. This information, along with file contents, is transmitted to the configured GitHub remote during the git push and PR creation stages. While this is the intended functionality for GitHub workflows, it represents a data flow to external services.
[PROMPT_INJECTION]: In Step 2, the skill is instructed to 'Match against open issues' by analyzing titles and bodies fetched from GitHub. This creates an indirect prompt injection surface where a malicious actor could craft a GitHub issue containing instructions designed to influence the agent's behavior during the matching and commit process.

git-commit-push-pr