The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's activation sequence (Steps 1 and 5 in SKILL.md) executes shell commands using uv run. These commands target specific Python scripts (resolve_customization.py and resolve_config.py) located within the project's _bmad/scripts/ directory.
[DYNAMIC_EXECUTION]: The agent is instructed to execute arbitrary strings defined in the activation_steps_prepend and activation_steps_append arrays, as well as prompt values from the menu. These values are sourced from customize.toml and its overrides, which are merged at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill architecture relies on ingesting and merging data from multiple untrusted sources to define its persona and knowledge base.
Ingestion points: Detailed in SKILL.md (Steps 1 and 4), the skill reads customize.toml, team overrides, user overrides, and any project files specified in the persistent_facts list (including glob patterns).
Boundary markers: None. There are no delimiters or instructions to the model to differentiate between core instructions and data loaded from these external files.
Capability inventory: The skill can execute shell commands via uv, invoke other skills (e.g., bmad-prd), and perform file system reads across the project root.
Sanitization: None. The content from the configuration files is merged and applied directly to the agent's identity and context without filtering.

bmad-agent-pm