The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script using uv run to resolve its activation configuration. This occurs in SKILL.md at Step 1: uv run {project-root}/_bmad/scripts/resolve_customization.py.
[DYNAMIC_EXECUTION]: The skill is instructed to iterate through and execute arbitrary steps defined in the activation_steps_prepend and activation_steps_append arrays within customize.toml and its overrides. While the defaults are empty, this mechanism allows for the execution of arbitrary instructions or actions provided by external configuration files at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a significant attack surface for indirect prompt injection as it derives its persona, role, and core instructions from external project files.
Ingestion points: Configuration is read from SKILL.md (Step 1 and 5) via customize.toml, {project-root}/_bmad/custom/{skill-name}.toml, and {project-root}/_bmad/gds/config.yaml. Additionally, Step 4 loads arbitrary files from the project root specified in persistent_facts.
Boundary markers: None. The skill instructs the agent to treat configuration entries as foundational facts and to execute steps without explicit delimiters or warnings to ignore embedded instructions.
Capability inventory: The skill can execute shell commands via uv run, trigger other skills through the menu system, and read arbitrary project files.
Sanitization: None. The instructions perform no validation or sanitization of the content loaded from the configuration files before adopting the persona or executing steps.

gds-agent-game-designer