distil-cli

Distil CLI Skill for Claude

A Claude skill for training task-specific small language models (SLMs) using the Distil Labs CLI and platform.

What This Skill Does

This skill teaches Claude how to help you:

• Train specialized SLMs — Create models up to 70x smaller than large models while maintaining accuracy
• Prepare training data — Generate proper data files for classification, QA, tool calling, and multi-turn tool calling tasks
• Run the Distil CLI — Execute commands for the full training workflow
• Deploy models locally or remotely — Set up your trained models with llama-cpp, vLLM, or Distil-managed infrastructure
• Train from production traces — Convert existing LLM logs into fine-tuned small models

Installation

Claude Code

/plugin marketplace add https://github.com/distil-labs/distil-cli-skill
/plugin install distil-cli@distil-cli-skill

Claude.ai / Claude Desktop

1. Download this repo as ZIP (or click "Code" → "Download ZIP" on GitHub)
2. Go to claude.ai → Settings → Capabilities → Skills
3. Click "Upload skill" and select the downloaded ZIP file
4. Toggle the skill ON

Prerequisites

Install the Distil Labs CLI:

curl -fsSL https://cli-assets.distillabs.ai/install.sh | sh

Supported Task Types

Task Type	Use Case	Example
Question Answering	Extract answers from documents	Invoice parsing, contract analysis, ticket extraction
Classification	Categorize text into fixed classes	Intent detection, sentiment analysis, ticket triage
Tool Calling	Select and invoke functions/APIs	API routing, workflow automation, chatbot actions
Multi-Turn Tool Calling	Multi-step conversations with tool use	DevOps chatbots, file system assistants, database interfaces
Open Book QA (RAG)	Answer questions using provided context	Document QA, support from docs
Closed Book QA	Answer from knowledge learned during training	FAQ bots, domain assistants

Skill Structure

distil-cli-skill/
├── .claude-plugin/
│   ├── marketplace.json
│   └── plugin.json
└── skills/
    └── distil-cli/
        ├── SKILL.md                          # Router + core instructions
        ├── references/
        │   ├── getting-started.md            # Install CLI, auth, quickstart
        │   ├── platform-overview.md          # What Distil Labs is, concepts, value prop
        │   ├── cli-reference.md              # All CLI commands with args and flags
        │   ├── task-selection-guide.md       # Choosing the right task type
        │   ├── model-catalog.md              # Student + teacher models, compatibility, defaults
        │   ├── job-description-guide.md      # Writing job_description.json
        │   ├── configuration.md              # Full config.yaml reference
        │   ├── mutations-guide.md            # Controlling synthetic data diversity
        │   ├── evaluation-metrics.md         # Metrics reference + interpretation
        │   ├── api-reference.md              # REST API setup + endpoints
        │   └── tasks/
        │       ├── prepare-data/
        │       │   ├── overview.md
        │       │   ├── question-answering.md
        │       │   ├── classification.md
        │       │   ├── tool-calling.md
        │       │   ├── multi-turn-tool-calling.md
        │       │   ├── open-book-qa.md
        │       │   └── closed-book-qa.md
        │       ├── upload-dataset.md
        │       ├── upload-and-process-traces.md
        │       ├── teacher-evaluation.md      # Incl. canonical verdict thresholds
        │       ├── training.md
        │       ├── deployment-integration.md
        │       ├── retrieve-predictions.md
        │       ├── analyze-predictions.md
        │       ├── polling-jobs.md            # Canonical polling loop
        │       └── verify-auth.md
        └── workflows/
            ├── dataset-to-model.md            # E2E: dataset → eval → train → deploy
            ├── traces-to-model.md             # E2E: traces → process → eval → train → deploy
            └── improving-a-model.md           # Iteration: ITERATE/RETHINK/RETUNE/ESCALATE

Quick Start

Once the skill is installed, just ask Claude to help you train a model:

"Help me train a classification model for customer support intent detection"

Claude will guide you through:

1. Creating a model with distil model create
2. Preparing your data files
3. Uploading data and running teacher evaluation
4. Training the model
5. Downloading and deploying

Documentation

• Distil Labs Documentation
• CLI Reference

License

Apache 2.0 — see LICENSE for details.