Spielewoy/autoprompt-skill

What it solves

Autoprompt is designed to reduce failures in agentic coding tasks. It prevents the common issue where a single agent might plan, approve, and verify its own work, which often leads to errors. By implementing a structured execution loop, it cuts failures by up to 45% on specific benchmarks (e.g., Terminal-Bench 2.1).

How it works

Instead of a linear prompt-response cycle, Autoprompt uses a hierarchical agent system. It breaks a goal down into several layers:

• Coordination and Management: High-level planning and oversight.
• Execution Lanes: Dedicated agents that perform the actual coding and implementation.
• Independent Checks: Separate agents that review, test, and verify the work to ensure it meets success criteria before sign-off.

Users trigger the skill via a command (like /autoprompt) and provide a goal. The system then handles the scoping, implementation, testing, and repair loop autonomously until the goal is reached or a human decision is required.

Who it’s for

Developers using AI coding agents such as Claude Code, Codex, OpenCode, Kilo Code, VS Code (with Copilot), and Prime Agent who want higher reliability and fewer manual prompts for complex coding tasks.

Highlights

• Multi-Agent Hierarchy: Separates planning, execution, and verification to avoid self-approval bias.
• Broad Compatibility: Supports multiple major coding agent providers.
• Concurrency Controls: Offers modes like tokensaver (caps subagents) and wide (maximizes parallel work) to manage cost and token usage.
• Autonomous Loop: Handles the full cycle of scoping, building, testing, and repairing without requiring step-by-step prompting.