The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several Git and GitHub CLI commands (git status, git commit, gh pr create, gh pr merge) to manage the repository lifecycle. These are the primary intended functions of the skill.
[COMMAND_EXECUTION]: The skill runs local scripts (python3 scripts/quality_score.py and ./scripts/backtest.sh) to enforce quality and consistency gates. These scripts are assumed to be part of the local repository environment.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests data from multiple sources:
Ingestion points: Processes user-provided $ARGUMENTS for commit messages, reads 'passport' files in quality_reports/passports/, and analyzes repository state via git status and backtest.sh.
Boundary markers: Uses EOF (heredoc) syntax for shell commands involving user-supplied text, which mitigates simple shell command injection.
Capability inventory: Has the ability to read files, execute shell commands, create branches, push code, and merge Pull Requests.
Sanitization: The instructions explicitly warn against staging sensitive files like .claude/settings.local.json or files containing secrets, and provide logic to halt if quality scores are low.

commit