The Agent Skills Directory

[SAFE]: The skill consists of documentation and reference material. No malicious code or suspicious behaviors were detected.
[EXTERNAL_DOWNLOADS]: The skill references official SDKs (such as typesafe-sdk and @typesafe-ai/sdk) and model repositories (e.g., ggml-org on Hugging Face). These are legitimate resources from the technology's vendor and well-known community organizations.
[COMMAND_EXECUTION]: Provides standard command-line examples for curl, ollama, and llama.cpp to demonstrate how to interact with classification APIs and serve local models.
[INDIRECT_PROMPT_INJECTION]: The skill addresses the processing of untrusted data by providing specific safety patterns, such as 'confidence-gated routing' and 'deterministic rails,' to ensure that agent actions based on model outputs are validated and controlled.

decision-models