The Agent Skills Directory

Indirect Prompt Injection (LOW): The scripts validate_tres.py and validate_tscn.py read local Godot files and output extracted data to the agent context.\n
Ingestion points: scripts/validate_tres.py (line 33) and scripts/validate_tscn.py (line 33) ingest untrusted file content via the read_text() method.\n
Boundary markers: No delimiters or explicit instructions are used to separate untrusted file content from the script's output results.\n
Capability inventory: The scripts are limited to standard library regex processing and stdout display; they do not have network access or file-writing capabilities.\n
Sanitization: Content extracted from files, such as node names or resource IDs, is printed directly to the console without validation or escaping.\n
Risk: Maliciously crafted file content (e.g., node names containing instructions) could attempt to influence the agent's behavior through the validation output.

godot