---
license: apache-2.0
library_name: sklearn
pipeline_tag: text-classification
tags:
- ai-systems
- capabilities
- reasoning
- planning
- memory
- agents
- world-models
- verification
- reliability
---

# Capability Classifier

**Capability Classifier** is a lightweight reference model that maps short AI task descriptions to practical capability categories.

It is published under the **ai-systems** handle as a transparent demonstration model for AI-system analysis.

## Capability Labels

- `adaptation`
- `agents`
- `coding`
- `memory`
- `multimodal`
- `planning`
- `reasoning`
- `reliability`
- `science`
- `tool-use`
- `verification`
- `world-modeling`

## Examples

Input:

```text
Break a complex objective into subtasks and replan after failure.
```

Expected category:

```text
planning
```

Input:

```text
Predict how the environment will change before acting.
```

Expected category:

```text
world-modeling
```

Input:

```text
Run tests before accepting generated code.
```

Expected category:

```text
verification
```

## Usage

```python
from joblib import load

classifier = load("capability-classifier.joblib")

text = "Use a browser and API to complete the task"
prediction = classifier.predict([text])[0]

print(prediction)
```

## Model Architecture

The reference model uses:

- TF-IDF text features
- unigram and bigram features
- logistic regression classification

The model is intentionally small so the classification approach remains easy to inspect and reproduce.

## Training Data

The model was trained on a small curated set of short AI-task descriptions covering capability areas such as:

- reasoning
- coding
- planning
- memory
- tool use
- agents
- multimodal understanding
- world modeling
- verification
- reliability
- adaptation
- science

## Intended Use

Suitable for:

- capability explorers
- educational tools
- lightweight taxonomy experiments
- AI-system documentation
- prototyping
- simple routing demos

## Limitations

This is a **reference model**, not a benchmark and not a production-grade classifier.

It was trained on a small curated dataset. Predictions outside the covered task descriptions may be unreliable.

The model should not be used for medical, legal, financial, safety-critical, or other high-impact decisions.

## Related Dataset

`ai-systems/ai-system-patterns`

## Related Model

`ai-systems/system-router`

## License

Apache-2.0