Agent Skill (for AI Coding Agents)
Transformer Lab ships an agent skill that teaches AI coding agents β Claude Code, and other compatible assistants β how to drive Transformer Lab through the lab CLI.
Once installed, your agent knows how to:
- check job status and stream logs
- queue, upload, and edit training tasks
- list, add, and configure compute providers
- create models and upload/download datasets
- publish job outputs
β¦all by running lab commands on your behalf, from inside your normal coding session.
Installβ
The easiest way is the bundled CLI command:
lab install-agent-skill
Behind the scenes this runs the following on your behalf:
npx skills add transformerlab/transformerlab-app --skill transformerlab-cli
You'll need Node.js installed (npx ships with it). If npx is missing, the command will tell you.
If you prefer to run the underlying npx invocation directly, that one-liner above works on its own. (You'll still need the lab CLI installed for the skill itself to be useful β the skill drives Transformer Lab by calling lab commands.)
What it teaches the agentβ
The skill is the same definition that lives in this repo at .agents/skills/transformerlab-cli/. It's a small bundle of instructions and command references that tells the agent which lab subcommands to use for common workflows, so you can ask things like:
- "What jobs are running on the lab cluster?"
- "Queue a training task using the dataset I just uploaded."
- "Stream logs for the last training job and tell me when it finishes."
β¦and the agent will translate those into the right lab calls.
Updatingβ
Re-run lab install-agent-skill (or the npx command) any time you upgrade Transformer Lab to pick up the latest version of the skill.