The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the user and the agent to run local shell commands for environment preparation (pip install), model management (spacy download), and executing the included training and evaluation scripts.
[EXTERNAL_DOWNLOADS]: Documentation and scripts reference the installation of official software packages from PyPI and the download of pre-trained NLP models from the spaCy official registry.
[PROMPT_INJECTION]: The provided FastAPI serving script (scripts/serve_model.py) processes untrusted text input, which creates an indirect prompt injection surface.
Ingestion points: Data enters the agent context via the POST /classify and POST /classify/batch endpoints in scripts/serve_model.py.
Boundary markers: None; the script passes the raw input string directly to the model pipeline.
Capability inventory: The primary capability is model inference (nlp(text)), which typically produces classification labels but can involve custom code execution if a loaded model contains malicious components.
Sanitization: Input is validated for length (1 to 100,000 characters) using Pydantic models, but no content-based sanitization is performed.
[SAFE]: No evidence of credential theft, obfuscated code, or persistence mechanisms was found. The operations described in the skill are standard practices for the Python NLP ecosystem and align with the skill's stated educational and technical purpose.

using-spacy-nlp