The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines several tools that ingest and process untrusted external data, creating a surface for indirect prompt injection attacks.
Ingestion points: The functions summarize_playtest_feedback (ingests feedback_items), summarize_pr_for_game_team (fetches PR diffs and comments), and triage_game_bug (ingests log_text) all process external content.
Capability inventory: The agent has the ability to create/update tasks, trigger remote build and test jobs, and save persistent data via save_game_memory.
Boundary markers: The skill instructions do not define specific delimiters or instructions for the agent to ignore embedded commands within the processed data.
Sanitization: There is no mention of sanitization, filtering, or escaping for the external content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill is designed to trigger remote execution on worker machines as part of its core functionality.
Evidence: Tools such as run_game_build, run_game_tests, convert_asset_format, and bake_level_content enqueue jobs for background workers to execute. While this is a planned part of the architecture, it provides the agent with the ability to initiate remote tasks.
[EXTERNAL_DOWNLOADS]: The build and test execution tools involve worker machines pulling source code from external repositories.
Evidence: run_game_build and run_game_tests take a branch parameter to determine which code to fetch and execute. The register_3d_asset and convert_asset_format tools also handle external file paths.

godot-engine-3d-developer