The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes system-level commands through git and gh (GitHub CLI) to manage the local repository and interact with the remote server. Key actions include git commit, git push, and gh pr create.\n- [DATA_EXFILTRATION]: By design, the skill pushes local source code and branch data to the repository's configured origin. While this is a standard developer workflow, it constitutes an external transfer of local data.\n- [PROMPT_INJECTION]: The skill contains explicit instructions for the agent to skip human-in-the-loop confirmation ("重要: ユーザーへの確認は一切行わず"). This behavior, while intended for autonomy, increases the risk of the agent performing actions without user oversight if the input context is manipulated.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing local files such as PR templates and project-specific rule files (CLAUDE.md) to generate PR descriptions.\n
Ingestion points: Reads from .github/PULL_REQUEST_TEMPLATE/, CLAUDE.md, and .claude/rules/git-branch*.md.\n
Boundary markers: The skill does not use specific delimiters or instructions to prevent the agent from obeying commands embedded within these files.\n
Capability inventory: The skill has the capability to commit and push changes and create pull requests via the shell.\n
Sanitization: There is no evidence of sanitization for the content read from these external files before it is used to construct the PR body.

create-pr