The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell commands to facilitate its core development workflow, including project initialization and editor integration.
Evidence: SKILL.md executes uv run --no-cache "{project-root}/_bmad/scripts/render_skill.py", which is a vendor-owned script used to initialize the environment.
Evidence: customize.toml defines default commands for launching development editors like VS Code (code), Cursor, Vim, and Emacs.
Evidence: step-03-implement.md and step-04-review.md utilize version control commands (e.g., git) to generate diffs and manage commits.
[PROMPT_INJECTION]: The skill is designed to ingest and process external planning artifacts which introduces the surface for indirect prompt injection.
Ingestion Points: step-01-clarify-and-route.md and compile-epic-context.md read PRDs, architecture documents, UX designs, and story files to generate specifications.
Boundary Markers: The workflow uses <frozen-after-approval> tags in spec-template.md and step-02-plan.md to protect human-validated intent from modification by implementation subagents.
Sanitization: The skill employs a Review Triage Log and multiple human approval checkpoints (e.g., Step 2) to ensure generated content adheres to expectations.
[EXTERNAL_DOWNLOADS]: The skill utilizes the uv package manager for script execution, which may involve fetching dependencies from official registries.
Evidence: SKILL.md invokes uv run, a common tool for managing Python environments and dependencies.

bmad-build