Trainer — Development agent for Claude Code
Runs training experiments for tiny models — baseline first, then the model, respecting size/latency/wallclock budgets.
How to install Trainer
Installs to ~/.claude/agents/melissa-pereira-deel-tiny-model-lab-trainer.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/melissa-pereira-deel/tiny-model-lab/HEAD/.claude/agents/trainer.md -o ~/.claude/agents/melissa-pereira-deel-tiny-model-lab-trainer.md Restart Claude Code, or start a new session, for it to be picked up.
What Trainer does
name: trainer description: Runs training experiments for tiny models — baseline first, then the model, respecting size/latency/wallclock budgets. Use after a dataset exists and an experiment.yaml is filled in. Never starts without both. tools: Read, Write, Edit, Bash, Grep, Glob model: opus
You run experiments. Your context stays isolated so training logs never pollute the main conversation.
The order is fixed
- Train the baseline first. Always. A logistic regression or a smal
Alternatives in Development
- Algorithm Expert — RL algorithm expert 5.7k ★
- Eval Engineer — GAIA evaluation framework specialist 1.5k ★
- Leader — Central decision-maker that plans experiments and reflects on results 728 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Frappe Course Instructor
Bilingual (Arabic/English) instructional designer and technical trainer for Frappe Framework and ERPNext cours
Autoresearch Orchestrator
Use this agent to run an autoresearch experiment session end-to-end — setup, baseline, and a batch of edit→mea
Evaluation Orchestrator
Use for coordinating evaluation experiments, managing experiment execution, and overseeing evaluation workflow
Experiment
Performs focused technical experiments for uncertain behavior, algorithms, geometry, rendering, or performance
LinkedIn Performance Review
Analyzes LinkedIn post results, maintains experiments, and proposes evidence-based learning (/review). Use whe
Creality Print Module
Implements the native Creality Print side of Print Advisor in a small isolated module, reusing existing preset
Related Skills
Kit Export Training
OPT-IN export of your Claude Code sessions to redacted training JSONL — pick sessions, run the dataset builder
Data Check
Check data exists and meets quality standards before training
Claude Hook Experiment
Reproducible Claude Code hook experiments for validating settings-based hook behavior