distil-cli-skill is a Claude Code plugin with 1 hand-picked skill, indexed on Flowy in the Machine Learning category.
$ npx -y skills add distil-labs/distil-cli-skill --agent claude-code
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.
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.
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>.
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.
| 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 |
SKILL.md is the entry point. stages/ holds one unit of pipeline work each, workflows/ the
two workflows (the build loop and the improvement loop), 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.
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 credit cost with you before it submits anything.
Apache 2.0. See LICENSE for details.
distil-cli-skill is a Claude Code plugin with 1 hand-picked skill, indexed on Flowy in the Machine Learning category. Install it with the command on its page. It includes distil-cli. Its skills do not fire on their own yet. Request auto-invocation to have Flowy route them as you prompt. Free and open source.
Is this plugin yours?
Claim it with GitHubSubmit a pluginPromote it