The Agent Skills Directory

[COMMAND_EXECUTION]: The skill functions as an orchestrator for developer tools, including git, gh (GitHub CLI), and pull request stack managers like Graphite (gt), Git Town, and gh stack. These tools are executed locally via subprocess calls to manage repository state and publication workflows.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data derived from the repository environment, such as Git diffs, log messages, and existing pull request metadata.
Ingestion points: External data enters the agent context through the outputs of git diff, git log, and gh pr view.
Boundary markers: The instructions explicitly mandate the use of single-quoted heredoc delimiters for multi-line strings (like commit messages) to protect against shell interpolation of special characters such as backticks and dollar signs.
Capability inventory: The skill has the capability to write files, create commits, push to remotes, and open pull requests via the gh CLI.
Sanitization: The skill utilizes internal scripts (check-pr-body and validate-publication) to validate the integrity and safety of generated pull request content before any publication commands are executed.
[CREDENTIALS_UNSAFE]: The skill implements a proactive security measure by requiring the agent to inspect git status and explicitly reject the staging of credentials, .env files, or unexplained large binaries. This mitigates the risk of accidental secret exposure in the repository history.

plate