Execute wiki-planned experiments by generating code, deploying, and collecting metrics.
npx skills add https://github.com/skyllwt/AutoSci --skill exp-run-skyllwt
Please help me install this Agent Skill. Skill: /exp-run Source: https://github.com/skyllwt/AutoSci/tree/main/.claude/skills/exp-run Command: npx skills add https://github.com/skyllwt/AutoSci --skill exp-run-skyllwt
It eliminates the repetitive, error-prone workflow of turning a planned experiment spec in the wiki into runnable code, deploying it (locally or remotely), monitoring it, and collecting standardized results back into the wiki.
Run the deployment for the planned experiment by telling the AI: "Deploy and monitor the planned experiment with slug 'my-exp' locally, and after a user confirmation and sanity checks, start the run."
High-intent search queries and answers about installing and using this skill.
Automating experiment deployment involves generating code from a wiki spec, running it locally or remotely via SSH, and persisting JSON results. This workflow updates the wiki with status, metrics, and mean±std outcomes automatically.
Yes, you can execute deploy-only or collect-only modes independently. This allows splitting the experiment lifecycle, so you can deploy code first and collect standardized metrics later without a full end-to-end run.
Before running experiment code remotely, the system requires explicit user confirmation, validates datasets and configs, and executes a small-scale sanity check. These guardrails prevent erroneous deployments and ensure config correctness.
Wiki experiment tracking handles multiple seed runs by saving individual run outcomes as JSON files under results/{slug}/seed_{N}.json. It then aggregates these results to report mean±std metrics in the wiki experiment markdown file.
An LLM code review is optional before local deployment. You can proceed with the experiment execution after passing the mandatory user confirmation gate and dataset validation checks without the LLM review.
The best way to manage an experiment lifecycle is using a full end-to-end run mode. It generates code from wiki/experiments/{slug}.md, applies sanity checks, deploys via remote SSH, and appends outcomes and logs back to the wiki.
LambenthanDeploy and collect machine learning experiments from wiki plans across local and SSH environments.
kamensRefreshes a wiki by collecting trial results, approvals, and guidelines with provenance.
Leo1349Deploy machine learning experiments to local or remote GPU servers via SSH.
Shallow-WConvert experiment plans into executable training code and deploy GPU runs.
