fainir/most-capable-agent-system-prompt
A copy-paste system prompt that has your coding agent (Claude Code, Codex, Cursor) build a self-improving agent system for any computer work. Try the lead follow-up demo at the website link.
What it solves
Most AI coding agents are "stateless"—they forget their progress between sessions, often claim tasks are finished without verifying them, and stall once a specific prompt is completed. This project provides a comprehensive system prompt that transforms these agents into durable, self-improving operating systems capable of managing long-term, multi-month projects.
How it works
When you paste this prompt into an agent (like Claude Code, Cursor, or Codex), it instructs the agent to stop simply chatting and start building a persistent project infrastructure. The agent creates a set of "operating files" (such as tasks.md, plan.md, and knowledge.md) that act as a memory bank. It then establishes a "closed-loop" system where it defines goals, executes tasks, verifies the results, updates its own memory, and learns from failures. You can choose between "harness-wrapper mode," which builds this system around your existing agent, or "native runtime mode" for deeper control.
Who it’s for
This is for developers and power users who want to move beyond simple code generation and build autonomous agents that can handle complex, multi-step workflows like software engineering, data analysis, or desktop automation. It is best suited for users who are comfortable with agents that write files, run shell commands, and require a dedicated workspace.
Highlights
- Persistent Memory: Moves beyond session-based chat by using file-based logs to maintain state across runs.
- Verification-First: Enforces a "closed-loop" design where tasks are not considered done until they are verified with evidence.
- Self-Improving: Includes instructions for the agent to build its own evaluation harnesses and refine its workflows over time.
- Universal Compatibility: Works with any agent that can read/write files and execute shell commands, including Claude Code, Codex, and OpenHands.
- Structured Architecture: Guides the agent to build a control plane, task graphs, and specialized worker sub-agents for complex projects.
Related
- Project
- Project
- Project
- Project