The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes local Python scripts using uv run to handle task-specific logic. These include memlog.py for state management, resolve_customization.py for configuration, and lint_spine.py for deterministic document validation.
[EXTERNAL_DOWNLOADS]: The skill references the bmad-code-org GitHub organization for method definitions and skill updates. Additionally, the test suite for the linter script declares a dependency on the well-known pytest package.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted external data, such as project specifications, raw ideas, and existing source code, to generate architectural artifacts.
Ingestion points: Reads from SPEC.md, user-provided ideas, and specified codebase paths in the project directory.
Boundary markers: Employs a multi-stage review process using specialized subagents and a mechanical linter script to ensure output integrity.
Capability inventory: Capable of executing shell commands through the uv tool and writing files to the project's documentation workspace.
Sanitization: Instructs the agent to verify all referenced technology versions against official web sources and project constraints rather than relying solely on training data.

bmad-architecture