What problem does this skill solve?
ML researchers spend days manually iterating: run experiment → read results → identify weaknesses → fix → repeat. This collection of 8 skills automates the entire loop overnight.
Who uses this workflow?
ML researchers who want autonomous experiment iteration. A real run improved a paper from 5/10 to 7.5/10 across 4 rounds with 20+ GPU experiments, zero human intervention.
Key features
- Cross-model collaboration: Claude Code executes (reads papers, writes code, deploys GPU experiments) while an external LLM (via Codex MCP) acts as critical reviewer
- auto-review-loop: Autonomous review→fix→re-review cycles (max 4 rounds)
- run-experiment: Deploy to local (MPS/CUDA) or remote GPU servers via SSH
- research-lit: Literature search and gap analysis
- novelty-check: Verify idea novelty against recent publications
Links