supratikpm

supratikpm / gemini-autoresearch

Public

Autonomous goal-directed iteration for Gemini CLI. Inspired by Karpathy's autoresearch. Modify → Verify → Keep/Discard → Repeat forever.

ai-agent autonomous-agent autoresearch gemini gemini-cli
15
3
100% credibility
Found Mar 29, 2026 at 15 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
JavaScript
AI Summary

gemini-autoresearch is a skill for AI assistants that automates iterative experiments to improve measurable aspects of code projects or content overnight.

How It Works

1
💡 Discover the overnight helper

You hear about gemini-autoresearch, an AI tool that improves your projects or content automatically while you sleep.

2
📥 Add it to your workspace

Follow quick steps to place the skill folder into your AI assistant's setup, like Gemini or Antigravity, so it's ready to go.

3
🎯 Set your improvement goal

Describe what you want better, like faster loading, higher scores, or cleaner content, and how to measure success.

4
🚀 Start the magic loop

Tell the AI to begin, and it starts running smart experiments, testing changes, and keeping only the wins.

5
😴 Walk away and rest

Let the AI work through the night, safely trying ideas, checking results, and learning from each step.

🌅 Wake up improved

Open the results log to see a list of gains, lessons learned, and your project or content noticeably better.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 15 to 15 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is gemini-autoresearch?

This JavaScript tool hooks into Gemini CLI to create an autonomous, goal-directed iteration loop—inspired by Karpathy's autoresearch—that runs forever: AI proposes changes, verifies metrics, keeps improvements, discards failures, and learns across runs. Set a goal like boosting test coverage or shrinking bundle size, define a verify command and optional guard check, then let it iterate overnight in headless mode via `gemini --yolo`. You wake up to a log of gains, git-committed changes, and compounding lessons, turning one-shot AI prompts into relentless optimization.

Why is it gaining traction?

Unlike generic autoresearch ports for Claude or Codex, this leverages Gemini's Google Search grounding for real-world verification (e.g., SEO tweaks against top results), 1M-token context for full project history, and true CLI headless mode for unattended runs. Commands like `/autoresearch:plan`, `:ship`, `:debug`, and `:security` auto-configure and fix issues, while the dual verify/guard system plus cross-run lessons prevent regressions better than basic keep/discard loops. Devs dig the measurable, compounding progress on metrics like Lighthouse scores or Docker sizes.

Who should use this?

Frontend devs chasing bundle trims or Lighthouse gains; backend folks optimizing SQL queries or CI pipelines; ops engineers shrinking Docker images. Non-coders in marketing/HR tuning SEO content, email subjects, or job descriptions via simple scripts. Ideal if you're already on Gemini CLI or Antigravity IDE and want autonomous exploration without babysitting.

Verdict

With 15 stars and 1.0% credibility score, it's early-stage and unproven at scale—docs shine but expect tweaks for complex projects. Worth a spin for Gemini CLI users experimenting with autonomous iteration; clone, test on a side project, and contribute if it clicks.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.