← ClaudeAtlas

kleinlisted

Run Klein Auto Research studies — process-verifiable research for AI for Science on the user's data, public data, a simulator, or a verifier alone, closed with a claims lock, an independent referee verdict, and a teaching artifact. Use for any research study (predict, estimate, test, simulate, replicate, discover, optimize), an experiment loop / autotuning, comparing or exploring a method, auditing data before modeling, studying an unfamiliar method, synthesizing findings, refereeing a study, or building a tutorial — whenever the user wants to research, experiment with, compare, or learn about methods on a question, EVEN IF they don't say "klein".
Xiang-Shan/klein-auto-research · ★ 2 · AI & Automation · score 68
Install: claude install-skill Xiang-Shan/klein-auto-research
# /klein — study lifecycle router > `/klein` is the Claude Code packaging of Klein's lifecycle. Everything it routes — > the reference protocols, profiles, templates and helper scripts below — is plain > markdown/Python usable with ANY coding agent or by hand; the tool-neutral entry > point is the repo-root `AGENTS.md`. Klein runs research **studies** under `studies/NN-slug/` through a fixed lifecycle of seven stages and four gates. The last three stages are what make it *research*, not just experiment-running. ``` new ─▶ CONSULT ─▶ DATA ─▶ METHOD ═══▶ EXPERIMENT/SWEEP ─▶ SYNTHESIZE ─▶ REFEREE ─▶ TUTORIAL Gate 0 Gate 1 Gate 2 └ the honest loop ┘ findings.md Gate 3 report/ ``` **Hard-block rule:** no modeling until CONSULT, DATA, and METHOD are recorded and the DATA artifact says GO; no `finalize` until REFEREE is recorded. Acknowledgements and explicit overrides are recorded with `klein gate`; prose in `program.md` alone cannot unlock a run. ## Stages `/klein <stage>` routes the lifecycle stages below. The STAGES are protocol routes; the `klein` CLI verbs (`new gate preflight run-one recover status finalize noise-floor verify headroom stop predict claims replicate sweep doctor generation`) are the machine actions each stage uses — the two are mapped here, not identical. Run stages in lifecycle order for a new study; `status` any time. The reference PROTOCOLS are the source of truth — the worker agents in `.claude/agents/` are optional accelera