The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's bootstrap instruction in SKILL.md requires the execution of a project-local script (_bmad/scripts/render_skill.py) using the uv tool. This is a controlled execution of a framework-specific component within the project's own directory.
[DYNAMIC_EXECUTION]: The workflow orchestrates the creation and execution of specialized, context-free subagents for coding and review tasks. These agents are directed by dynamically generated prompts based on the current project state and developer specifications.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an inherent attack surface by ingesting external user intent (e.g., story descriptions or issue links) to generate specifications. It mitigates this risk through several design patterns:
Ingestion Points: User-provided intent prompts and external files processed in step-01-clarify-and-route.md.
Boundary Markers: Uses <frozen-after-approval> tags to delimit human-verified intent from agent-generated content.
Capability Inventory: Capabilities include file system writing, local script execution, and subagent orchestration.
Sanitization: The planning process in step-02-plan.md acts as a distillation phase, and the skill explicitly instructs the agent to ignore any directives within the intent that attempt to bypass workflow steps.

bmad-build