Skip to content

About

Claude skill for distil cli

Resources

Stars

190 stars

Watchers

3 watching

Forks

Latest commit

 

History

84 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 

Repository files navigation

distil labs Agent Skill

An Agent Skill for building task-specific small language models on the distil labs platform: from raw data or production traces, through teacher evaluation and synthetic data generation, to a finetuned, evaluated, deployable student model.

Everything runs on the distil labs platform through the distil CLI, or through the REST API from Python when you prefer it. The only prerequisite is an account. The API backend also needs pip install requests pyyaml, and takes its access tokens from distil access-token.

The skill works with every coding agent that reads the Agent Skills format, including Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot and OpenCode.

Installation

With the distil CLI

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

distil skill install puts the skill in ~/.agents/skills, the shared directory that Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Cline and many other agents read. It also links the skill into Claude Code. Run it again to update the skill. distil skill uninstall removes it. The command needs git. It does not need Node.

With the skills CLI

If you have Node, skills installs the skill without the distil CLI:

npx skills add distil-labs/distil-cli-skill

Use this route for an agent that distil skill install does not cover, with --agent <name>.

Prerequisites

The skill runs every stage through the distil CLI, installed above. Sign in with distil auth, or distil signup without an account; the skill does this itself at the start of a project if you skip it. On Windows, run the CLI inside WSL.

Supported Task Types

Task Type Use Case Example
Question Answering Extract answers from documents, or pick one label from a fixed set Invoice parsing, contract analysis, intent detection, ticket triage
Chat Completion Conversational turns with text, tool calls, or both Support chatbots, API routing, chatbot actions
Agentic Chat Completion Tool-calling loops where tool results feed back into the answer DevOps assistants, file system assistants, database interfaces

Layout

SKILL.md is the entry point. stages/ holds one unit of pipeline work each, workflows/ the two workflows (building a model in a single pass, and autoresearch, where the agent iterates on the model for you), and references/ the shared knowledge, with every distil command in references/execution/cli.md and every REST API request in references/execution/backend-api.md. Swapping the CLI for the REST API changes the commands and nothing else.

Quick Start

Once the skill is installed, ask your agent to build you a model:

"Help me build a model that labels the intent of customer support messages"

The agent starts from the model you already run in production: it puts an inference endpoint in front of it to collect traces, or takes the files you have: traces, a train set, a test set, or any combination. It builds a test set, picks the teacher with a teacher evaluation, builds the train set from relabelled traces and synthetic data, trains the student, and serves it behind a new endpoint that takes the traffic the traces came from. Every stage confirms the setup and the expected cost with you before it submits anything.

Documentation

License

Apache 2.0. See LICENSE for details.

About

Claude skill for distil cli

Resources

Stars

190 stars

Watchers

3 watching

Forks

Releases

Packages

Used by

Contributors