The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill scaffolds workflows that download standard, well-known GitHub Actions. It mitigates supply-chain risks by explicitly instructing and templating the pinning of these actions to specific commit SHAs rather than mutable tags.\n- [CREDENTIALS_UNSAFE]: The skill manages secrets by delegating them to the native encrypted storage of the CI platform (e.g., GitHub Secrets). The generated templates follow security best practices by avoiding the logging of secrets and ensuring temporary credential files on runners are deleted immediately after use.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted repository data (e.g., package.json, requirements.txt) to detect the project stack. It mitigates this attack surface by requiring the agent to present an 'Assumptions block' for user review before making any file changes, and by using the detected data to select static templates rather than for raw command interpolation.\n- [COMMAND_EXECUTION]: The skill utilizes a set of restricted CLI tools (gh, vercel, fly, etc.) to configure deployment environments. These operations are scoped to the intended deployment targets and are subject to platform-level tool restrictions.

setup-ci