The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses dynamic context injection (the ! syntax) in the YAML frontmatter to execute git commands (git branch, git status) at load time. These commands are benign and used strictly for project grounding.
[SAFE]: The instructions include specific mitigations against command injection. It explicitly warns against interpolating user-provided text into shell command strings and instead mandates writing text to a file before processing it as a CLI argument for the gh tool.
[SAFE]: The skill enforces a 'read-only' contract, explicitly stating it must not modify source code or open pull requests. External actions like filing issues are described as optional hand-off steps requiring user confirmation.
[SAFE]: Data ingestion of untrusted content (via Glob, Grep, and user input) is handled safely. The skill uses placeholders like (unknown — needs reporter confirmation) for missing data rather than fabricating information or executing untrusted logic.

write