The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The documentation describes a system designed to ingest external data, which inherently possesses an indirect prompt injection surface.
Ingestion points: The bmad-ingest and bmad-deep-recon components process raw transcripts, chat threads, and web research results as described in help/ingest.md and help/research.md.
Boundary markers: The framework utilizes structured Markdown templates (located in ingest-templates/) to organize processed data.
Capability inventory: The documented framework has the ability to write files, perform network research, and execute internal Python scripts via uv run.
Sanitization: The bmad-ingest process includes a documented 'review pass' where the system checks the generated output against the raw source to verify accuracy.
[COMMAND_EXECUTION]: The help files reference the use of uv run to execute internal framework scripts (e.g., resolve_customization.py, list_templates.py) located in the project's _bmad/scripts/ directory. This is standard operational behavior for the toolset's management and configuration resolution.
[EXTERNAL_DOWNLOADS]: The bmod.toml manifest specifies a trusted GitHub repository as the source for updates. Additionally, the documentation notes that the tool can fetch templates from user-provided GitHub URLs, which is a core feature for team-wide template sharing.

bmod-core-tools