Skill Doctor: score and improve your agent skills
Score past Claude Code, Codex, and Warp agent conversations against tested rubrics, then get concrete edits to your SKILL.md files — free and open sou...
AI Agent Framework
Summary
An open-source framework that transforms AI coding assistants into autonomous agents via an iterative Modify-Verify-Repeat loop. Inspired by Karpathy, it leverages mechanical metrics and Git-based rollbacks to achieve continuous, compounding improvements in software development.
Highlights
- Implements an autonomous loop with automatic Git-based reversion for failed attempts.
- Features a plain-language orchestrator that derives success predicates from goals.
- Includes specialized subcommands for debugging, security, and deployment tasks.
- Optimized architecture significantly reduces token consumption per invocation.
auto-generated
Context
Audience
Software Engineers and AI Developers
DomainAI Agent Orchestration
Formatopen source software
Accessfree online
Source Overview
Converts Claude Code, OpenCode, or OpenAI Codex into an autonomous loop that iterates atomic changes against a single mechanical metric, auto-reverting failures via git. v2.2.0 adds a plain-language orchestrator that classifies a typed goal, derives a success predicate, then chains subcommands (debug, fix, security, ship, evals) until done.
Architecture was rebuilt in v2.1.0 to cut token cost from ~100K to 5–8K per invocation across 12 self-contained command files.
Topics
Related
Claude CodeOpenAI CodexKarpathy autoresearch
