The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands using the Git CLI and GitHub CLI (gh). This is the primary function of the skill and is handled with significant safety constraints. Commands include branch management, commits, and pushing to remote repositories.
[PROMPT_INJECTION]: The skill includes explicit instructions for the agent to reject user commands that would violate established safety rules (such as directly committing to protected branches). This provides a layer of defense against malicious or accidental instructions that could disrupt the repository structure.
[DYNAMIC_EXECUTION]: For commit messages and Pull Request bodies, the skill mandates the use of the shell Heredoc format (cat <<'EOF'). This is a security best practice that prevents the shell from interpreting the content of the message as code, effectively mitigating command injection risks from user-provided or AI-generated text.
[DATA_EXPOSURE]: The skill accesses project-specific configuration files (CLAUDE.md, AGENTS.md) to verify project-specific rules. This is a standard procedure for context-aware agents and does not involve sensitive credential access or unauthorized data exfiltration.

pr