# Autoprompt

A coding-agent workflow skill that reviews, fixes, and rechecks an agent's own work, reported to cut task failures by 45% in the project's own benchmark against unmodified OpenCode.

Autoprompt is a coding-agent skill that wraps a task in a review, fix, and recheck workflow to reduce failures, installed as a CLI that detects and configures a chosen coding agent from a list including Claude Code, Codex, OpenCode, Kilo Code, VS Code, Prime Agent, Oh My Pi, DeepSeek Harness, Reasonix, Hermes Agent, and Grok Build.

Version 1 benchmarks reported on a fixed OpenCode comparison track show OpenCode alone solving 60 of 89 tasks (67.42%) versus 73 of 89 (82.02%) with Autoprompt added, with 16 failures instead of 29; the README notes that version 2 benchmarks will follow. Installation supports both a released package through npm and building from a source checkout, and an installed setup can be checked, updated, repaired, or removed through `autoprompt doctor` and `autoprompt uninstall` commands.

It requires Node.js 20 or later, Python 3.11 or later with PyYAML, and Bash 4.3 or later on macOS or Linux.

Website: https://github.com/Spielewoy/autoprompt-skill