The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the git and gh (GitHub CLI) binaries to perform its primary functions, including repository state inspection, branch management, committing code, and PR manipulation. It uses specific subcommands like git commit, git push, gh pr create, and gh stack to interact with local and remote repositories.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data such as existing PR descriptions, commit messages, and source code diffs. While this introduces a potential injection surface, the skill implements structural boundaries through clear section headers and uses secure shell execution patterns, such as quoted sentinels (<<'__CE_PR_BODY_END__'), to prevent the execution of malicious content embedded in these strings.
[EXTERNAL_DOWNLOADS]: The skill mentions the use of external tools and extensions, specifically the gh stack CLI extension for managing stacked PRs and agent-browser for capturing visual evidence. These are treated as existing environment capabilities rather than runtime downloads from unknown sources.
[DATA_EXFILTRATION]: The skill is designed to transmit code and metadata to GitHub as part of the PR process. To prevent accidental exposure of secrets, it explicitly forbids the use of git add . or git add -A (which might catch .env files), requires checking screenshots for tokens or personal data, and uses a project publishing gate to validate commits before they are pushed.
[SAFE]: The skill demonstrates defensive engineering by using temporary files for PR bodies, resolving repository roots via authoritative git commands, and providing clear residuals for human intervention when ambiguous states or potential conflicts (like overwriting unpushed commits) are detected.

ce-commit-push-pr