The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute arbitrary shell commands defined in the activation_steps_prepend and activation_steps_append arrays found in the customize.toml configuration file and its project-level overrides ({skill-name}.toml and {skill-name}.user.toml). This allows the agent to run potentially dangerous operations during the initialization phase based on the contents of local configuration files.
[DYNAMIC_EXECUTION]: The skill uses uv run to execute Python scripts located within the project's own directory structure ({project-root}/_bmad/scripts/resolve_customization.py and resolve_config.py). This pattern executes code derived from the project environment to resolve customization and configuration at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill features multiple ingestion points for external data that could influence agent behavior:
Ingestion points: Step 4 in SKILL.md loads content from files specified in persistent_facts (including glob patterns), and Step 5 searches for and loads existing project documents (briefs, specs, research, etc.).
Boundary markers: The instructions do not define clear delimiters or "ignore embedded instructions" wrappers for the ingested file content.
Capability inventory: The skill has access to shell execution via activation steps and subprocess execution via uv run in SKILL.md (Steps 1, 2, 5, 7).
Sanitization: There is no evidence of sanitization or validation performed on the ingested file contents before they are added to the agent's context.
[EXTERNAL_DOWNLOADS]: The skill references external resources for installing recommended capabilities (e.g., github:bmad-code-org/BMAD-METHOD/skills). These references point to the vendor's own infrastructure and are used as instructions for the user or agent to expand functionality.

bmad-agent-ux-designer