The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external content such as transcripts, email threads, and URLs to generate notes that downstream agents or users are instructed to act upon.
Ingestion points: Raw input text from users, local files (transcripts, voice memos), and content fetched from external URLs.
Boundary markers: The instructions mandate that every item must trace back to the raw source and include an 'Unclear in the source' section to separate ambiguous content.
Capability inventory: The agent can execute local Python scripts via uv run, read/write to the file system, and use gh to read remote content.
Sanitization: A mandatory 'Review' step is implemented where a second pass (or a subagent) compares the generated note against the raw source to identify hallucinations or omissions.
[COMMAND_EXECUTION]: The skill executes local utility scripts located within the project's directory to handle its internal logic.
Evidence: Execution of resolve_config.py, resolve_customization.py, and list_templates.py via the uv tool from the {project-root}/_bmad/scripts/ folder.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of related vendor tools if they are not already present in the environment.
Evidence: Commands to add the bmad or bmad-customize skills using the npx skills add bmad-code-org/BMAD-METHOD pattern.

bmad-ingest