The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a Python script using uv run during activation (resolve_customization.py). It also executes arbitrary "activation steps" defined in the agent configuration block (activation_steps_prepend, activation_steps_append).
[DYNAMIC_EXECUTION]: The skill's activation logic dynamically merges and executes instructions from external .toml configuration files located in the project root (_bmad/custom/). This allows for runtime modification of the agent's execution flow based on files that can be modified by other processes or users in the environment.
[INDIRECT_PROMPT_INJECTION]: The skill ingests potentially untrusted data from multiple external sources to define its persona and knowledge base.
Ingestion points: Project configuration files (config.yaml), team/user overrides (.toml), and local files referenced via file: globs in persistent_facts.
Boundary markers: No specific delimiters or "ignore instructions" warnings are applied to the content loaded from these files.
Capability inventory: The agent has capabilities for shell command execution, file system access, and skill dispatching.
Sanitization: There is no evidence of sanitization or validation for the instructions loaded from the activation_steps or persistent_facts arrays.

gds-agent-game-architect