The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install a companion asset skill using the command npx skills add ericzakariasson/skills --skill game-builder-blender-assets. This download targets the author's own repository and uses standard package management tools.
[COMMAND_EXECUTION]: The Orchestrator agent manages the execution of local build commands, loop verification scripts, and specialized capture tools (e.g., tools/capture and tools/make-sxs) developed during the build process to facilitate visual review.
[INDIRECT_PROMPT_INJECTION]: The skill has a defined attack surface for indirect injection via external data processing:
- Ingestion points: External visual references (images) and their metadata (URLs) are stored in refs-locked/ and SOURCES.md for the Critic agent to evaluate.
- Boundary markers: The Critic role is mandated to run in a fresh session with no access to the Builder's chat history or the Orchestrator's internal commentary; verdicts must follow a strict, validated schema.
- Capability inventory: Includes spawning subagents, reading/writing project files, and executing local tools.
- Sanitization: The skill includes a pre-handoff verification phase that scans all artifacts for credentials, secrets, and model identifiers. It explicitly forbids using external references as direct inputs for asset generation to prevent prompt leakage.
[DYNAMIC_EXECUTION]: To support visual parity testing, the Builder agent generates and commits capture scripts to the tools/ directory. These are then executed by the Orchestrator to produce the artifacts required for the Critic's evaluation loop.

game-builder