The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script using uv run at both the start and end of the workflow. The script path is {project-root}/_bmad/scripts/resolve_customization.py, which is resolved relative to the project working directory.
[COMMAND_EXECUTION]: The activation sequence involves executing instructions defined in activation_steps_prepend and activation_steps_append. These steps are loaded from customize.toml and potentially from team or user override files, allowing for dynamic instruction execution based on local configuration.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from multiple external sources and user inputs to guide its workflow, representing a vulnerability surface for indirect prompt injection. • Ingestion points: Data is retrieved from design-methods.csv, config.yaml, and multiple customization TOML files. The persistent_facts feature also allows loading content via file globs. • Boundary markers: The instructions do not define explicit delimiters or warnings to the agent to ignore potentially malicious instructions embedded within these data sources. • Capability inventory: The skill possesses capabilities for subprocess execution via uv run, arbitrary file reading within the project scope, and writing output files to the filesystem. • Sanitization: There is no evidence of content validation, escaping, or filtering for data retrieved from external files before it is processed or interpolated into prompts.

bmad-cis-design-thinking