The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local CLI tools including git and GitHub CLI (gh) to manage the development lifecycle. Commands are scoped to repository operations such as branch management, staging changes, and pushing commits.- [CREDENTIALS_UNSAFE]: Contains a specific security instruction to avoid staging or committing sensitive files like .env or credential files, reducing the risk of accidental secret exposure.- [DATA_EXFILTRATION]: Transfers repository data (diffs, titles, descriptions) to GitHub via the official gh tool. This is a documented and expected behavior for a pull request utility targeting a well-known service.- [COMMAND_EXECUTION]: Implements a git reset --hard command for cleanup, but wrap it in safety logic that verifies a clean working tree to prevent data loss.- [PROMPT_INJECTION]: Includes strict 'Hard stops' and error-handling instructions that prevent the agent from proceeding with ambiguous or potentially dangerous states, such as pushing directly to a default branch.

submit-task