---
license: apache-2.0
library_name: sklearn
pipeline_tag: text-classification
tags:
- ai-systems
- agents
- agent-tasks
- routing
- automation
- tool-use
- multi-agent
---

# Agent Task Classifier

**Agent Task Classifier** is a lightweight reference model that maps short agent-task descriptions to practical operational categories.

It is published under the **ai-systems** handle as a transparent example for agent routing, task decomposition, and AI-system prototyping.

## Task Labels

- `browser`
- `coding`
- `communication`
- `control`
- `data-analysis`
- `memory`
- `multi-agent`
- `planning`
- `research`
- `retrieval`
- `tool-use`
- `verification`
- `workflow-automation`

## Examples

Input:

```text
Search the web for recent papers and summarize the findings.
```

Expected category:

```text
research
```

Input:

```text
Fix a bug in a Python project and run the tests.
```

Expected category:

```text
coding
```

Input:

```text
Require human approval before sending the final message.
```

Expected category:

```text
control
```

## Usage

```python
from joblib import load

classifier = load("agent-task-classifier.joblib")

text = "Delegate subtasks to specialized agents and merge their results"
prediction = classifier.predict([text])[0]

print(prediction)
```

## Model Architecture

The model uses:

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

It is intentionally small and easy to inspect.

## Training Data

The reference model was trained on a small curated set of agent-task descriptions covering:

- research
- coding
- browser interaction
- data analysis
- retrieval
- communication
- workflow automation
- planning
- verification
- memory
- tool use
- multi-agent coordination
- control and approval

## Intended Use

Suitable for:

- agent-routing demos
- task taxonomies
- educational examples
- workflow prototypes
- lightweight agent orchestration experiments
- AI-system documentation

## Limitations

This is a **reference model**, not a production-grade task router.

The training set is intentionally small, so ambiguous or out-of-domain requests may be misclassified.

It should not be used to make medical, legal, financial, safety-critical, or other high-impact decisions.

## Related Dataset

`ai-systems/ai-system-patterns`

## Related Models

- `ai-systems/system-router`
- `ai-systems/capability-classifier`

## License

Apache-2.0