The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md explicitly direct the agent to execute shell commands defined in configuration fields (activation_steps_prepend, activation_steps_append, and on_complete). These commands are retrieved from customize.toml or project-level override files, allowing for arbitrary command execution controlled by configuration.
[DYNAMIC_EXECUTION]: The workflow utilizes uv run to execute a local Python script (resolve_customization.py) which dynamically determines the configuration and steps to be executed at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and process external Game Design Documents (GDD) and Game Briefs from the project's output folder to generate its deliverables. This ingestion of untrusted external content represents a vulnerability surface where malicious instructions could be embedded in the game documentation to influence the agent's behavior.
Ingestion points: Reads files matching *gdd*.md and *brief*.md in the {output_folder}.
Boundary markers: None explicitly defined for the ingested file contents.
Capability inventory: Ability to execute shell commands via uv run and the activation/completion step mechanisms.
Sanitization: No sanitization or validation of the content of the ingested Markdown files is described.

gds-playtest-plan