The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/check_action_pins.py uses the subprocess module to execute the gh (GitHub CLI) tool. This is used to resolve version tags to commit SHAs via the gh api command. The execution is performed by passing a list of arguments and does not involve a shell, which prevents command injection from the file content being scanned.
[EXTERNAL_DOWNLOADS]: The skill's workflow and the assets/examples/verify.sh script recommend using uvx to download and execute specialized security tools, including zizmor and check-jsonschema. These tools are fetched from the Python Package Index (PyPI), which is a well-known service, and are used for pipeline security validation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process CI/CD pipeline configuration files from the repository, which constitutes an external data source.
Ingestion points: The agent reads pipeline YAML files from directories like .github/workflows/ or specific files like .gitlab-ci.yml.
Boundary markers: The skill instructions emphasize using env: blocks for interpolating external values, which provides a structural boundary in the generated configurations.
Capability inventory: The agent can modify local repository files and execute validation tools (gh, uvx, actionlint, shellcheck).
Sanitization: The instructions provide specific rules to avoid direct interpolation of untrusted event data into shell scripts, requiring the use of environment variables and quoting to prevent injection vulnerabilities.

configure-ci-cd-pipelines