The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates agent templates that contain shell script steps (run: shell: bash) for automation tasks such as SCM provider detection, repository management, and environment variable configuration. These scripts use standard Linux utilities (e.g., git, sed, echo) to interact with the CI/CD environment.
[INDIRECT_PROMPT_INJECTION]: The generated templates are designed to process untrusted data from repositories (code, pull request metadata) which is then interpolated into LLM prompts for analysis or code generation.
Ingestion points: Inputs defined in generated pipeline.yaml files, such as repo, branch, and pullReq as seen in references/agent-examples.md.
Boundary markers: Not present in the generic templates; the skill relies on the user to implement specific delimiters or safety instructions in the resulting agents.
Capability inventory: The templates include capabilities for shell execution, file system writes to build outputs, and network operations via CI/CD plugins (e.g., PR creation, posting comments).
Sanitization: No specific sanitization or validation logic is included in the base templates for external inputs.

create-agent-template