The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses uv run to execute local Python scripts located within the project's _bmad/scripts/ directory, such as memlog.py, resolve_customization.py, and lint_spine.py. These scripts are used for initializing the decision log, resolving project-specific configurations, and performing mechanical validation of the architecture document. This execution is scoped to the project environment and supports the skill's core functionality.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data, such as project specifications, raw ideas, and existing codebases, to distill them into an architecture spine. This content is eventually passed to subagents for review. While this creates a theoretical surface for indirect injection, the risk is managed through a deterministic linter and a structured review rubric that focuses on architectural consistency.
Ingestion points: Project source code, SPEC.md files, and user-provided raw ideas (as described in SKILL.md).
Boundary markers: No specific delimiters or isolation markers for untrusted content are defined in the prompts.
Capability inventory: File system access for reading/writing documentation and execution of local Python scripts via uv run.
Sanitization: The skill includes a dedicated linter (lint_spine.py) that checks for placeholders and structural errors before the documentation is finalized.

bmad-architecture