evoloop

README

Evoloop

A CLI tool that runs self-improvement loops on local Git projects.

Evoloop inspects your project, detects quality issues, generates patch proposals via LLM, evaluates them against configurable policies, and tracks the results — all locally.

How It Works

evoloop init       # Set up .evoloop/ config
evoloop run        # Run the full improvement loop automatically
evoloop inspect    # Detect project structure and available commands
evoloop analyze    # Run quality checks and generate improvement issues
evoloop propose    # Generate a patch proposal for an issue using LLM
evoloop evaluate   # Evaluate a proposal against tests, lint, and policy
evoloop history    # View execution history

Automated Loop

evoloop run executes the full loop in a single command:

analyze → propose → evaluate → (apply)

evoloop run                              # single iteration, review only
evoloop run --auto-apply                 # apply accepted patches automatically
evoloop run --auto-apply --max-iterations 3  # up to 3 iterations
evoloop run --max-failures 2             # stop after 2 consecutive failures

Step-by-Step

You can also run each step manually:

1. inspect detects the Git project, branch, dirty state, and available test/lint/typecheck commands.
2. analyze runs those commands and generates ImplementationIssue entries for any failures.
3. propose takes an issue, builds a prompt, calls the LLM, and saves the resulting patch.
4. evaluate applies the patch in a temporary directory, runs quality checks, and decides Accept or Reject.
5. history shows all issues, executions, and evaluations stored in SQLite.

Installation

go install github.com/kiwamoto1987/evoloop@latest

Or build from source:

git clone https://github.com/kiwamoto1987/evoloop.git
cd evoloop
go build -o evoloop .

Quick Start

# Initialize in your project
cd your-project
evoloop init

# Edit config if needed
vim .evoloop/config.yaml

# Run the automated improvement loop
evoloop run --auto-apply

# Or run step by step
evoloop inspect
evoloop analyze
evoloop propose --issue <ISSUE_ID>
evoloop evaluate --execution <EXECUTION_ID>
evoloop history

Configuration

evoloop init generates .evoloop/config.yaml:

project_name: my-project

llm:
  provider: claude
  model: sonnet
  command: "claude"

evaluation:
  test_command: "go test ./..."
  lint_command: "golangci-lint run"
  typecheck_command: "go build ./..."

policies:
  max_changed_files: 5
  max_changed_lines: 200
  deny_paths:
    - ".github/**"
    - ".evoloop/**"

Evaluation Policy

Every proposal is evaluated against:

• TestPassed — all tests pass
• LintPassed — no lint violations
• TypeCheckPassed — type checks pass
• ChangedFileCount — within max_changed_files limit
• ChangedLineCount — within max_changed_lines limit

All criteria must pass for a proposal to be accepted.

Requirements

• Go 1.22+
• Git
• Claude CLI (for propose command)

License

MIT