The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a shell script scripts/probe_asset_credentials.sh and provides instructions in references/asset-recovery.md to execute commands within a login shell context using zsh -lc or bash -lc. This pattern is employed to source user profile configuration files (e.g., .bashrc, .zshrc) to ensure the agent has access to environment variables, specifically API keys required for external asset generation services like Tripo, Gemini, and ElevenLabs.- [INDIRECT_PROMPT_INJECTION]: The skill orchestrates a complex workflow that involves reading and writing to several artifact files, such as artifacts/game-progress.md, artifacts/evidence.json, and artifacts/final-evidence.md, as well as interacting with "sibling skills." This creates a potential surface where external data or previous outputs could influence subsequent agent instructions.
Ingestion points: artifacts/game-progress.md, artifacts/evidence.json, artifacts/final-evidence.md, and sibling skill instructions.
Boundary markers: Not explicitly defined in the manifest or progress files, though the agent is instructed to follow specific schemas.
Capability inventory: Subprocess execution for Python and Node.js scripts, shell command execution via bash/zsh, and file system write access.
Sanitization: The artifact verification script scripts/check_evidence.py implements basic sanitization by filtering resolved paths against a whitelist of media/data file extensions.

threejs-game-director