The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script resolve_customization.py using the uv run command during the initialization phase to resolve workflow settings. This occurs automatically upon skill activation.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by design, as it is intended to process external, potentially untrusted data to create design documents.
Ingestion points: Processes user brain dumps, pitch decks, design notes, prototypes, and digests from web-research subagents (e.g., genre landscape and market trends).
Boundary markers: Instructions do not explicitly require the use of delimiters or specific safety warnings when interpolating extracted content into the main design brief prompt.
Capability inventory: The skill has permissions to write to the local filesystem (brief.md, addendum.md, .decision-log.md), execute shell commands via uv run, and interact with external services via MCP tools defined in the external_handoffs configuration.
Sanitization: The skill employs a mitigation strategy where subagents perform relevance-filtered extraction of source artifacts rather than loading full documents, which reduces the attack surface.
[DYNAMIC_EXECUTION]: The skill framework supports executing arbitrary instructions or sequences of commands defined in the on_complete, activation_steps_prepend, and activation_steps_append fields of the customize.toml configuration file. While intended for post-processing and setup, this allows for runtime execution of logic defined in local project files.

gds-create-game-brief