The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local assets like assets/methods.csv and configuration fields in customize.toml to define method intents and output patterns. These strings are interpolated into the agent's prompts to guide the refinement process. While this creates a surface for indirect prompt injection, the sources are local files intended for customization and are managed by the user or vendor.
Ingestion points: assets/methods.csv, customize.toml (additional_methods field).
Boundary markers: None. Method descriptions are used directly as instructions.
Capability inventory: Local script execution via uv run, file reading.
Sanitization: None.
[COMMAND_EXECUTION]: The skill uses uv run to execute internal Python scripts for managing its methods catalog and resolving project-level configurations. These scripts are part of the skill's distribution or the established vendor infrastructure.
Evidence: Executions of scripts/pick_methods.py, _bmad/scripts/resolve_customization.py, and _bmad/scripts/resolve_config.py.

bmad-advanced-elicitation