Autoresearch Builder
- 32 repo stars
- Updated March 20, 2026
- jung-wan-kim/autoresearch-builder
autoresearch-builder is a plugin marketplace that adds one Claude Code plugin running an autonomous modify-test-keep/discard experiment loop.
About
autoresearch-developer is a one-plugin Claude Code marketplace from Jung-Wan Kim that implements an autonomous experiment loop: propose a change, run tests, decide keep or discard, and repeat. developers use it when validating technical bets—model tweaks, refactor paths, or feature prototypes—without babysitting each iteration. The design echoes karpathy/autoresearch, making it familiar to developers who want empirical iteration inside their agent workflow rather than ad-hoc prompts. It spans validate and build because successful keeps become product code, but the marketplace’s primary home is early scoping and proof. Expect intermediate complexity: you must supply meaningful tests and safe boundaries so the loop does not churn destructive changes. Ideal for developers running tight feedback cycles on APIs, agents, or research codebases.
- Single autoresearch-builder plugin (v1.1.0) for closed-loop experiments
- Workflow: modify, test, keep or discard, repeat without manual orchestration
- Designed for any project type, not only ML notebooks
- Inspired by karpathy/autoresearch pattern for agent-driven iteration
- Productivity category plugin aimed at autonomous research sessions
Autoresearch Builder by the numbers
- Data as of Jul 7, 2026 (Skillselion catalog sync)
/plugin marketplace add jung-wan-kim/autoresearch-builderAdd your badge
Show developers this marketplace is listed on Skillselion. Paste this into your README.
| repo stars | ★ 32 |
|---|---|
| Last updated | March 20, 2026 |
| Repository | jung-wan-kim/autoresearch-builder ↗ |
What it does
Run a karpathy-style autonomous modify-test-keep/discard loop from Claude Code to explore ideas without manual experiment bookkeeping.
Who is it for?
Best when you're validating hypotheses with automated tests and want agent-driven iteration inspired by autoresearch.
Skip if: Production-only hotfix workflows with no test harness or teams that forbid autonomous file changes.
What you get
After install, the agent can loop experiments automatically so only changes that pass your tests survive in the repo.
- One autoresearch-builder plugin wired to your repo
- Iterated patches that pass your test gate before keep
- Documented experiment outcomes from repeated agent cycles
By the numbers
- Marketplace contains 1 plugin named autoresearch-builder at version 1.1.0
- Core loop steps: modify, test, keep/discard, repeat
Plugins in this marketplace
1 plugin - install individually after you add the marketplace.
Recommended Marketplaces
How it compares
Autonomous experiment-loop plugin, not a static planning skill or passive code reviewer.
FAQ
Who is Autoresearch Builder for?
Developers and researchers using Claude Code who want hands-off experiment loops with explicit keep/discard decisions.
When should I use Autoresearch Builder?
Use it while scoping or prototyping when you have tests or checks and need many quick iterations without manual glue code.
How do I add Autoresearch Builder to my agent?
Register the autoresearch-developer marketplace in Claude Code and enable the autoresearch-developer plugin from the repo root source path.