The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation includes code snippets demonstrating the use of subprocess.run to execute external command-line tools like ffmpeg within background tasks.
[INDIRECT_PROMPT_INJECTION]: The skill describes a framework that ingests and processes untrusted data from multiple external sources, creating an attack surface where malicious input could influence agent behavior.
Ingestion points: Untrusted data enters the context through HTTP request bodies (request.json()), query parameters (Query), multipart file uploads (UploadFile), and WebSocket message streams (websocket.iter_text()).
Boundary markers: The framework utilizes msgspec.Struct and Python type hints to define data boundaries and enforce schemas, though it does not explicitly provide instructions to the agent to ignore embedded natural language commands within these fields.
Capability inventory: The documented capabilities include database write operations (acreate, asave), serving local system files (FileResponse), and shell command execution (subprocess.run).
Sanitization: Input validation is handled via msgspec serialization and validation plans, which perform type and structure checking on incoming data.

django-bolt