mpommrehn

Agent Meta Prompt — AI skill for Claude Code

AI community

Sample meta-prompt to direct LLMs and Agent Harnesses like Claude Code, Codex, OpenCode, Grok Build, Hermes, etc., to operate they way YOU want them to.

How to install Agent Meta Prompt

This entry records only its repository, not the path inside it, so there is no exact command to give. Open mpommrehn/agent-meta-prompt and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agent Meta Prompt does

Sample meta-prompt to direct LLMs and Agent Harnesses like Claude Code, Codex, OpenCode, Grok Build, Hermes, etc., to operate they way YOU want them to. Includes best practice metaprompts from 3 AI companies plus my preferences as an example.

Alternatives in AI

Repomix — 📦 Repomix is a powerful tool that packs your entire repository into a single, AI-friendly file 22.7k ★
Manifest — Real-time cost observability for OpenClaw agents — track tokens, costs, messages, and model usage 4.1k ★
Agentrq — AgentRQ: Human-in-loop realtime conversational task manager for AI Agents 1.1k ★

README

agent-meta-prompt

A collaboration meta-prompt for coding agents, and the record of how it was built from the published prompting guides of Anthropic, Google, and OpenAI. With it: the working habits that grew up around it over a summer of daily use, written so you can copy the ones you want.

It is for two readers. Someone about to work with the author and an agent on a shared codebase, who needs to know how the agent has been told to behave and why. And anyone setting up Claude Code, Codex, or a similar harness for themselves, who would rather start from a tested prompt than from a blank `AGENTS.md`.

What is here

File	What it is
`META-PROMPT.md`	The prompt itself: role, communication style, skepticism, asking for input, task framing, agentic execution, durable-config discipline, unattended runs, testing and security policy. A placeholder stands where the principal's name goes.
`HOW-IT-WAS-BUILT.md`	The four sources, what was taken from each and what was skipped and why, the calibration decisions, and the version history.
`SESSION-HYGIENE.md`	Context-rot thresholds, the checkpoint prompt, and the rules for gates and delegated work.
`EVOLUTION-LOG.md`	The rule that every project keeps a construction and evolution log, and what an entry holds.
`MACHINE-CONTEXT.md`	One `AGENTS.md` across several machines: imports instead of symlinks, per-machine notes files, presence checks.
`VOICE-PATTERN.md`	How to give an agent a personal writing voice without publishing the voice: the pattern, not the author's files.
`agent-meta-prompt-EVOLUTION.md`	This repository's own evolution log, from the first commit.

How to use it

Copy the XML block from META-PROMPT.md into your AGENTS.md. Replace {{PRINCIPAL}} with your name and the role line with one sentence about who you are to the agent.
If your harness reads CLAUDE.md, make that file the single line @AGENTS.md. Claude Code expands the import