open-thoughtsOfficial·27 Agent Skills Included

OpenThoughts-Agent

Distributed model training, evaluation, and cluster job management

Launches and monitors distributed training, data generation, and model evaluation jobs across HPC and cloud GPU/TPU clusters. Eliminates manual cluster babysitting with automated job sweeps, failure diagnosis, checkpoint publishing, and registry cleanup. Helps researchers run reinforcement learning, fine-tuning, and benchmark workflows faster with consistent, repeatable procedures.

276|37

↗

npx skills add open-thoughts/OpenThoughts-Agent --all -g -y

Available:

Tells the agent how to route requests to the correct launch, cleanup, monitoring, and analysis skills, and enforces standing rules like environment paths, cluster sync discipline, and safety guardrails.

All Skills in This Repository (27)

Pure Emerald Level Indicators

📦 In Repo

open-thoughts

datagen-launch-iris

Orchestrate datagen jobs on Marin's Iris TPU cluster from launch to cleanup.

rl-agentic-launch-jupiter

Launch agentic RL experiments on Jupiter with SkyRL, Harbor, and Daytona.

monitor-job-tables

Format HPC job-status data into bucketed box-drawing tables.

monitor-restore

Restore the 3-hour Jupiter+Leonardo cluster sweep loop via /loop 3h or CronCreate.

monitor-restore-iris-cron

Re-register the every-3-hours Iris job-monitor cron via CronCreate.

datagen-job-cleanup

Upload verified datagen traces to HF and free disk space.

sft-job-cleanup

Publish finished LLaMA-Factory SFT jobs to Hugging Face and clean up HPC cluster state.

eval-agentic-launch-iris

Manages the full lifecycle of Iris evaluation jobs via the command line.

crud-otagent-supabase

Query, aggregate, and safely write OT-Agent eval data in Supabase.

sft-cleanup-hf-only

Publish HF-only SFT cells to Hugging Face without database registration.

eval-standard-cleanup

Consolidate finished standard eval jobs into a SCORES.md tracker.

sft-launch-jupiter

Launch SFT experiments on Jupiter via hpc.launch with LLaMA-Factory.

Official

Advanced

Frequently Asked Questions

FAQPage Schema

How to install OpenThoughts-Agent?▼

Run `npx skills add open-thoughts/OpenThoughts-Agent --all -g -y` in your terminal to install all skills in this collection globally.

What does OpenThoughts-Agent do?▼

It manages the full lifecycle of training small agentic models: generating trace data, running SFT and RL jobs on HPC clusters, evaluating models on benchmarks, and publishing results to Hugging Face and a registry.

How do I launch a training job on a cluster?▼

The skills route you through a unified launcher (`python -m hpc.launch --job_type <type>`) with per-cluster configs for SFT, RL, datagen, and eval jobs, including node counts, reservations, and restart chains.

Can it monitor and clean up finished jobs automatically?▼

Yes. Dedicated skills run recurring cluster sweeps that detect completions and failures, then trigger cleanup workflows that upload checkpoints, register models, and free disk space.

Does it work with Claude Code and other coding agents?▼

Yes. All skills follow the standard SKILL.md format and run in Claude Code, Cursor, Codex, and other compatible agent environments.