The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Downloads AI models for text and image processing from Hugging Face, a well-known model repository.
[COMMAND_EXECUTION]: Manages a local llama-server process using subprocess and utilizes system utilities like lsof and ps to verify local port binding and process health.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data (text and images) for anonymization. While this presents an attack surface, the skill uses structured semantic tags and provides specific instructions for agents to define boundaries when sending anonymized data to external models, mitigating the risk of instructions being executed as commands.
[DATA_EXFILTRATION]: The skill incorporates a security check that restricts the inference server to the local loopback interface (127.0.0.1), preventing sensitive data from being sent to external endpoints during the model inference process.

has-anonymizer