my-auto-goal-plugin

my-auto-goal-plugin

A Claude Code plugin for autonomous goal-achievement loops.

Inspiration

This plugin is inspired by Andrej Karpathy's autoresearch - an autonomous research loop that iteratively improves on a research goal. I've generalized the concept beyond research to support six different domains: research, code optimization, debugging, app development, content writing, and web/SEO optimization.

The core idea: give it a goal, let it iterate until it gets there or cleanly pauses.

How It Works

┌─────────────────────────────────────────────────────────────┐
│  1. CLARIFY GOAL                                            │
│     Define objective, success criteria, constraints         │
│                           ↓                                 │
│  2. APPROVE PLAN                                            │
│     Review configuration, approve once                      │
│                           ↓                                 │
│  3. AUTONOMOUS LOOP                                         │
│     ┌──────────────────────────────────────┐                │
│     │  Evaluate current state              │                │
│     │           ↓                          │                │
│     │  Generate hypothesis                 │                │
│     │           ↓                          │                │
│     │  Implement change                    │                │
│     │           ↓                          │                │
│     │  Measure improvement                 │                │
│     │           ↓                          │                │
│     │  KEEP (improved) or DISCARD (worse)  │                │
│     │           ↓                          │                │
│     │  Log results, update state           │                │
│     └──────────────────────────────────────┘                │
│                    ↺ repeat until goal met                  │
└─────────────────────────────────────────────────────────────┘

Supported Domains

Domain	Metric Type	Example Goal
Research	Rubric score (0-100)	"Research API rate limiting best practices"
Code	Objective metric	"Reduce test runtime by 50%"
Debug	Bug status (0/50/100)	"Fix silent login failure"
App	Build/test pass	"Implement user authentication"
Content	Rubric score (0-100)	"Write API documentation"
Web-SEO	Lighthouse scores	"Achieve SEO score >= 95"

Installation

From GitHub

Add to your ~/.claude/settings.json:

{
  "extraKnownMarketplaces": {
    "my-auto-goal": {
      "source": {
        "source": "github",
        "repo": "akhil41/my-auto-goal-plugin"
      }
    }
  },
  "enabledPlugins": {
    "my-auto-goal-plugin@my-auto-goal": true
  }
}

Then restart Claude Code or run /reload-plugins.

Local Installation

git clone https://github.com/akhilvemuri/my-auto-goal-plugin ~/.claude/plugins/my-auto-goal-plugin

Add to settings.json:

{
  "extraKnownMarketplaces": {
    "local": {
      "source": {
        "source": "directory",
        "path": "~/.claude/plugins/my-auto-goal-plugin"
      }
    }
  },
  "enabledPlugins": {
    "my-auto-goal-plugin@local": true
  }
}

Quick Start

/my-auto-goal

The plugin guides you through goal clarification, then runs autonomously:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Goal:        Reduce test suite runtime by 50%
Domain:      code (LOCKED)
Metric:      runtime (lower-is-better)
Eval:        pytest --durations=0
Will stop:   goal met | stagnation | user stop
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Proceed? (yes / modify first)

Example: Research Loop

Goal: Research best practices for API rate limiting
Success: rubric_score >= 85

Iteration 0: baseline = 30
Iteration 1: ✅ KEEP add token bucket section → 48 (+18)
Iteration 2: ✅ KEEP add sliding window section → 62 (+14)
Iteration 3: ✅ KEEP add comparison table → 75 (+13)
Iteration 4: ✅ KEEP add implementation examples → 86 (+11)

Goal met: 86 >= 85

Example: Debug Loop

Goal: Fix login failing silently
Success: bug_status = 100

Iteration 0: bug reproduces, bug_status = 0
Iteration 1: ❌ HYPOTHESIS WRONG - API returns 401 correctly
Iteration 2: ❌ HYPOTHESIS WRONG - handler fires correctly
Iteration 3: ✅ FIX VERIFIED - stale closure in useEffect
             Added regression test, bug_status = 100

Goal met!

Key Features