The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/stage-hunks.py uses subprocess.check_output and subprocess.run to interact with Git. These calls are implemented securely using the list-based argument format (preventing shell injection) and include the -- separator to ensure that user-provided file paths are not interpreted as command-line options.
[DATA_EXPOSURE]: The skill processes local file differences to facilitate Git staging. No sensitive data exfiltration or unauthorized file access was detected.
[PROMPT_INJECTION]: The instructions in SKILL.md define a clear task-oriented workflow. There are no patterns suggesting attempts to bypass safety guidelines or override agent constraints.
[INDIRECT_PROMPT_INJECTION]: As the skill involves reading file diffs which could contain untrusted content, there is a theoretical surface for indirect injection. However, the workflow includes verification steps (git diff --cached) and instructions for the agent to report groupings to the user, providing a natural human-in-the-loop checkpoint.

stage-by-concern