The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local system commands to perform screen captures and launch development tools. Evidence includes:
Calling the grim utility in scripts/capture.py for Wayland sessions.
Using subprocess.Popen in scripts/stage_editor_bridge.py to start the Godot editor with specific project paths.
[DYNAMIC_EXECUTION]: The skill uses dynamic module loading and modifies project configuration files at runtime. Evidence includes:
Using __import__ in scripts/capture.py to load local dependency scripts (asset_sheet, ensure_gitignore, webp_encode) based on the file system layout.
Modifying the project.godot file in scripts/stage_editor_bridge.py to enable a temporary editor plugin.
[INDIRECT_PROMPT_INJECTION]: The skill ingests visual data from the project, creating an attack surface for indirect prompt injection if malicious instructions are present in screenshots or assets.
Ingestion points: Screenshots and assets are loaded via the Pillow library in scripts/capture.py and scripts/webp_encode.py.
Boundary markers: No explicit delimiters or instructions to ignore embedded visual text are present in the review rubrics or instructions.
Capability inventory: The skill possesses file-write permissions and the ability to execute shell commands via subprocess across multiple scripts.
Sanitization: No visual sanitization or instruction filtering is performed on the visual content before it is presented to the agent.

godot-agent-vision