The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions specify the execution of Python scripts using uv run. These include a discovery script bundled with the skill (scripts/list_customizable_skills.py) and a resolver script expected to be located within the project's directory structure (_bmad/scripts/resolve_customization.py).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by scanning and reading SKILL.md descriptions and customize.toml configuration files from other installed skills in the environment. This data is displayed to the user and used by the agent to generate new configuration files.
Ingestion points: The scripts/list_customizable_skills.py script reads frontmatter and TOML content from sibling directories in the skills root.
Capability inventory: The agent has the ability to read and write files within the project root (specifically under _bmad/custom/) and execute local shell commands via uv run.
Boundary markers: The instructions do not implement specific boundary markers or "ignore" instructions for the metadata ingested from other skills.
Sanitization: Content is extracted via regex and parsed as TOML without explicit sanitization to filter out potential prompt injection attacks embedded in skill descriptions or configuration fields.

bmad-customize