autoresearch

Why we included this project

This is a skill pack that wires Claude Code, OpenCode, and OpenAI Codex into an autonomous loop that keeps working toward a goal without babysitting. You give it a plain-language objective, and it derives a success predicate, confirms it once, then iterates — modify, verify, keep or discard — until it decides the task is done. Teams running long or repetitive engineering chores will appreciate the 14 commands and nine safety hooks, which keep an otherwise tireless agent from drifting past its constraints. It also claims to cut tokens per invocation, so it matters if you're watching API spend on agent-heavy workflows. If you already live inside a terminal-based coding agent and want a structured way to let it self-improve rather than run one-shot prompts, this gives you a tested harness to build on.