Glossary
The vocabulary this site uses, defined in plain language. Each entry is written for AI Agent Hub and links back to the guide or directory page where the term matters.
Agent skill
A packaged prompt and a short description that narrows what an AI coding agent does for one job.
Benchmark score
A number from a standard test set. It ranks models on that test, not on your job.
Context window
The maximum number of tokens a model can consider at once, counting both your input and its output.
Directory score
The ranking number used to sort this site's listings. It is a sorting hint, not a benchmark we ran.
DSH plugin
An installable extension for DeepSeek Harness, added by command or prompt rather than by editing its source.
Install prompt
The block of text you paste into an AI chat so the agent writes a skill or plugin into place for you.
Model card
The documentation page a model publisher writes, describing what the model is, what data trained it, and what it should not be used for.
Open weights
A model whose parameters can be downloaded, which is not the same thing as a permissive license.
Quantization
Storing model weights at lower precision to make them fit in less memory, at some cost to quality.
VRAM
The memory on a graphics card that has to hold model weights, the context, and the runtime when you serve a model yourself.