The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes helper scripts located within the project's _bmad/scripts directory (e.g., resolve_customization.py, resolve_config.py) to manage configuration and customization logic.
[DYNAMIC_EXECUTION]: Several steps (e.g., in step-01-validate.md and step-04-validate-and-summary.md) include an On Complete trigger that resolves a command string from the workflow.on_complete configuration and instructs the agent to execute it as a terminal command.
[EXTERNAL_DOWNLOADS]: The generated CI pipeline templates include steps to download the Maestro CLI from its official GitHub repository releases. This targets a well-known service for a specific development tool and is a standard functional requirement.
[INDIRECT_PROMPT_INJECTION]: The skill's evaluation plan rendering logic (step-03b-render-evaluation-plans.md) reads and processes evaluation-ci-plan.json files. These files are used to dynamically modify the CI pipeline structure, creating an ingestion surface where data influences agent-generated code.
Ingestion points: ci/evaluation-ci-plan.json files detected in the project root via git ls-files.
Boundary markers: Processing is constrained by the evaluation plan schema and placement rules, but specific delimiters within the file are absent.
Capability inventory: The skill has capabilities to write CI configuration files, list project files, and execute shell commands via npm exec.
Sanitization: The skill attempts to validate plans by running tea-evaluate check via npm exec before rendering them into the pipeline.

bmad-testarch-ci