Matt Pocock's Prompt for Cleaning Up AGENTS.md
Matt Pocock's Prompt for Cleaning Up AGENTS.md
A three-subagent prompt helps simplify AGENTS.md, remove ineffective instructions, and move coding standards into a separate file.
Matt Pocock shared a prompt for working with AGENTS.md — the instruction file for coding agents that often becomes a source of wasteful token consumption and chaotic rules.
/writing-for-agents my AGENTS.md is a complete mess. Propose a series of structural changes that:
- Remove instructions that do not affect anything.
- Reveal details only when necessary.
- Move instructions into CODING_STANDARDS.md.
Do this work using three subagents, each acting more radically than the previous one. Create one PR.
AGENTS.md is one of the most common sources of inefficient token use. The prompt tackles the problem at the root: instead of accumulating random rules, it suggests progressively simplifying the structure, removing useless instructions, and moving coding standards into a separate document.
/writing-for-agents is a skill from Matt Pocock's set for working with agent instructions, including AGENTS.md and CLAUDE.md. Before using the prompt, the skill needs to be installed: skills.
Why it matters
AnalysisA clean AGENTS.md reduces token waste, prevents contradictory agent instructions, and makes AI-assisted coding more predictable.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
OpenMotion: An Open-Source CapCut Alternative for AI-Agent Video Editing
OpenMotion has appeared on GitHub as a full-featured CapCut alternative that creates and edits videos using AI agents, templates, and brand kits.
Loop Engineering: Architecture of Autonomous Agentic Loops and the Post-Prompting Era
A deep dive into Loop Engineering: self-sustaining execution loops for Claude Code and autonomous agents, the DOER/CHECKER topology, deterministic verification, the /goal primitive, and reward hacking mitigation.
Short-Term and Long-Term Memory of Agents
The architecture of multi-tier memory for autonomous agents. It distinguishes between Short-Term Memory (active context window of the current session) and Long-Term Memory (vector databases, user profile files, and episodic diaries).
Few-Shot Prompting (In-Context Learning)
A prompt engineering methodology where the model adapts to a specific format, style, or logic during inference (In-Context Learning) by showing 2–5 reference examples.