autoprompt-skill
Autoprompt is a coding-agent skill that cuts failures by 45% on agentic coding tasks.
What is it?
What it is
A coding-agent skill CLI that reduces failures by 45% on agentic coding tasks by orchestrating multi-agent workflows with coordinated execution lanes, independent reviews, and test-driven verification loops.
Why it exists
To handle the full execution loop for autonomous coding agents—so users provide goals and constraints once while Autoprompt manages scoping, implementation, testing, review, repair, and verification across a layered multi-agent hierarchy.
Who should use it
Developers using coding agents (e.g., Claude Code, Codex, OpenCode) who want to reduce failures and automate the execution loop of agentic coding tasks.
Who should avoid it
Beginners unfamiliar with coding agents or agentic workflows, users seeking full manual control, and those working on trivial tasks where the overhead is not justified.
How it works
A quick walkthrough in plain English
How autoprompt-skill works
Step 1 of 3
You interact with it
Open autoprompt-skill, send a request, or connect it to your stack.
Features
Advantages
- 45% reduction in failures on agentic coding tasks compared to baseline
- Higher solve rate (73/89 = 82.02%) versus OpenCode alone (60/89 = 67.42%)
- Works with nine different coding agents and providers
- Reduced failure count from 29 to 16 in benchmarked tasks
- Extended functionality across multi-agent systems, code review, automated testing, and workflow automation
- Active development with regular releases (version 1.0.4)
- Strong community support with translations in multiple languages
- Flexible concurrency control through mode and max_subs parameters
Disadvantages
- Requires Node.js 20+ and Python 3.11+ with PyYAML dependency
- Increased computational cost (~3x time, ~2x tokens per task)
- Some providers don't support custom model routing (inherits active model)
- Potential complexity in setting up custom agents beyond standard configurations
- Performance gains may vary significantly for small tasks vs large ones
Installation
native
npm install -g autoprompt-skill autoprompt Choose your coding agent, confirm its path, and install. N means enter another path. For another CLI or IDE, choose Custom coding agent and use the compatibility guide. Install from source: git clone https://github.com/Spielewoy/autoprompt-skill cd autoprompt-skill npm install -g . autoprompt
FAQ
What does Autoprompt do, and how much does it improve coding-agent performance?
Autoprompt is a coding-agent skill that reduces failures on agentic coding tasks. On Terminal-Bench 2.1 with OpenCode 1.18.7, it solved 73 of 89 tasks (82.02%) versus 60 of 89 (67.42%) without Autoprompt — a +14.61 point gain and 45% fewer failures. The trade-off is roughly 3x the time and 2x the tokens.
How do I install Autoprompt?
Run `npm install -g autoprompt-skill`, then launch the installer with `autoprompt`. Choose your coding agent, confirm its path, and install. To check an existing installation use `autoprompt doctor --strict` (or `autoprompt doctor PROVIDER --strict` for one provider). To uninstall use `autoprompt uninstall` or `autoprompt uninstall PROVIDER`. You can also install from source by cloning the repo and running `npm install -g .`.
Which coding agents does Autoprompt support?
Nine providers are listed as Working: Claude Code (`claude`, 2.1.219+), Codex (`codex`, subagent-capable), OpenCode (`opencode`, 1.18.7+), Kilo Code (`kilo`, 7.4.22+), VS Code (`vscode`, 1.133+), Prime Agent (`prime`, 0.7.2), Oh My Pi (`omp`, 17.4.0+), DeepSeek Harness (`deepseek`, 0.1.0-rc.7+), and Reasonix (`reasonix`, 1.30.0+). For other CLIs or IDEs, choose `Custom coding agent` and follow the compatibility guide.
What are the system requirements?
You need Node.js 20+, Python 3.11+ exposed as `python` with PyYAML installed, Bash 4.3+ on macOS or Linux, and Git only if you are using the GitHub checkout install method.
How do I invoke Autoprompt and control its runs?
Use `/autoprompt <goal>` in most agents, `$autoprompt <goal>` in Codex, and `/skill:autoprompt` in Oh My Pi. Concurrency is controlled with `mode=`: `mode=tokensaver` caps active subagents at six, `mode=wide` opens every ready lane, and `mode=custom max_subs=N` sets your own ceiling. Custom model routing via `agents=` is available in Claude Code and Codex; other providers inherit the active or parent model.
How autonomous is Autoprompt, and what is the multi-agent structure for?
Autoprompt can scope, implement, test, review, repair, and verify a goal on its own, but it stops for choices that change the result, actions requiring your authority, or blockers it cannot safely resolve. The layered agent hierarchy separates coordination, management, execution, and independent judgment so that the same agent does not plan, approve, and verify its own work.
Featured in Videos
YouTube tutorials and walkthroughs for autoprompt-skill
Alternatives
Similar projects ranked by category, topics, and text overlap.
