The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes run_shell_command to perform operations such as resource management via gcloud and environment discovery via git. This functionality is essential to its role as a CI/CD assistant and is controlled by a structured process requiring architectural approval from the user before execution.
[INDIRECT_PROMPT_INJECTION]: The skill scans local repository files to identify application archetypes and gather infrastructure context.
Ingestion points: Scanning of project manifest files (e.g., pom.xml, package.json, go.mod) as described in SKILL.md and references/how_to_build_cloudbuild_yaml.md to determine the build environment.
Boundary markers: None explicitly defined for the file scanning process, which relies on the agent's interpretation of existing project structure.
Capability inventory: run_shell_command, file write operations, and specialized GCP management tools (e.g., create_build_trigger).
Sanitization: None present for the ingested file metadata, although the skill logic is mapping file existence to fixed patterns rather than executing embedded instructions.
[EXTERNAL_DOWNLOADS]: The skill references official documentation and community best practices from sources including Google Cloud, Docker, and GitHub. These references are provided for user guidance and informational support rather than runtime execution of untrusted remote code.

google-cicd-pipeline-design