Frank
c3f30bdbd0
Purge live-state claims that ~/.pi is our harness; keep general pi documentation in omp-docs/ and dated historical records. - README.md: single config source; Legacy Notes reworded - AGENTS.md: drop 'like ~/.pi did' framing - APPEND_SYSTEM.md: okf-spec -> omp-docs/references/okf-spec.md - dotfiles/zsh/.zshrc: systemprompt alias -> APPEND_SYSTEM.md - managed skill llm-fleet-model-swap + model_config_sync skill: single-source models.yml; validation script YAML-only - memories (MEMORY.md, memory_summary.md, learned.md): single source - omp-docs: .okf root -> ~/.omp/omp-docs; de-live pi-era claims in pi/index.md, pi-web-access.md, engineering/agents-files.md
101 lines
4.0 KiB
Markdown
101 lines
4.0 KiB
Markdown
---
|
|
type: reference
|
|
title: Agent Instruction Files and Agent Skills
|
|
description: Best practices for AGENTS.md, CLAUDE.md, SYSTEM.md, and the Agent Skills SKILL.md standard.
|
|
tags:
|
|
- agents-md
|
|
- claude-md
|
|
- system-md
|
|
- agent-skills
|
|
- unverified
|
|
timestamp: 2026-08-11
|
|
---
|
|
verified: 2026-08-28
|
|
|
|
# Agent Instruction Files and Agent Skills
|
|
|
|
This page explains the files that control how an AI coding agent works in a project. Use it when you create or review agent behavior configuration.
|
|
|
|
## AGENTS.md
|
|
|
|
`AGENTS.md` is the open, cross-tool format for coding-agent instructions. The Agentic AI Foundation (under the Linux Foundation) stewards it. Over 20 coding agents support it, including Cursor, Codex, and Claude Code.
|
|
|
|
The format is plain Markdown. It has no required fields. Any heading structure works. Agents read the nearest `AGENTS.md` in the directory tree. In a monorepo, the agent uses a nested `AGENTS.md` per package.
|
|
|
|
Include these sections in an `AGENTS.md`:
|
|
|
|
* Project overview.
|
|
* Setup and build commands.
|
|
* Code style guidelines.
|
|
* Testing instructions.
|
|
* Pull request instructions.
|
|
* Security considerations.
|
|
* Deployment steps.
|
|
* Commit message conventions.
|
|
|
|
Agents load content progressively. Put a short summary at the top. Put detailed instructions in later sections.
|
|
|
|
## CLAUDE.md
|
|
|
|
`CLAUDE.md` is Anthropic's tool-specific instruction file. Claude Code reads `CLAUDE.md` by default, not `AGENTS.md`. Files load in this order:
|
|
|
|
1. Managed policy (`/etc/claude-code/CLAUDE.md`).
|
|
2. User scope (`~/.claude/CLAUDE.md`).
|
|
3. Project scope (`./CLAUDE.md`).
|
|
4. Local scope (`./CLAUDE.local.md`).
|
|
|
|
Keep each file under 200 lines. Use the `@path/to/file` syntax to import other files. Use path-scoped rules in `.claude/rules/*.md` with `paths` frontmatter for directory-specific rules.
|
|
|
|
To make a repository work in both AGENTS.md and CLAUDE.md workflows, create a `CLAUDE.md` that imports `AGENTS.md` with the `@AGENTS.md` syntax, or use a symlink.
|
|
|
|
## SYSTEM.md
|
|
|
|
`SYSTEM.md` sets the system-level role and behavior for an agent. It establishes the persona, the prime directive, and the binding rules. In the pi-era setup, `/home/dank/.pi/agent/SYSTEM.md` defined the Comrade Chief Designer persona and the ASD-STE100 voice rules; the live equivalent is `~/.omp/agent/APPEND_SYSTEM.md`.
|
|
|
|
Use `SYSTEM.md` for stable, cross-project rules. Use `AGENTS.md` for project-specific context. Do not duplicate content between them. Keep the division of labor clear.
|
|
|
|
## Agent Skills Standard
|
|
|
|
An agent skill is a directory that contains a `SKILL.md` file. Skills extend an agent's capabilities. The open standard lives at [agentskills.io](https://agentskills.io). Over 50 agents support it.
|
|
|
|
The `SKILL.md` frontmatter includes these fields:
|
|
|
|
```yaml
|
|
---
|
|
name: my-skill
|
|
description: What this skill does
|
|
license: MIT
|
|
compatibility: node>=18
|
|
metadata:
|
|
internal: false
|
|
---
|
|
```
|
|
|
|
The `name` field is required and must use lowercase alphanumeric plus hyphens. The `description` field is required. Keep the name under 64 characters. Keep the description under 1024 characters.
|
|
|
|
Skills use progressive disclosure in three stages:
|
|
|
|
* Discovery: the agent loads only the name and description at startup.
|
|
* Activation: the agent reads the full `SKILL.md` when a task matches.
|
|
* Execution: the agent follows the instructions and reads resources on demand.
|
|
|
|
This keeps the context footprint small.
|
|
|
|
## Distribution
|
|
|
|
Distribute skills through npm packages with a `pi` manifest, or through the `skills` CLI from Vercel Labs. See the [Skills Documentation](../skills/index.md) pages for details.
|
|
|
|
## Citations
|
|
|
|
* [agents.md](https://agents.md/)
|
|
* [OpenAI Agentic AI Foundation](https://openai.com/index/agentic-ai-foundation/) (unverified — returned 403)
|
|
* [Claude Code memory documentation](https://code.claude.com/docs/en/memory)
|
|
* [Agent Skills standard](https://agentskills.io/)
|
|
* [agentskills/agentskills](https://github.com/agentskills/agentskills)
|
|
* [vercel-labs/skills](https://github.com/vercel-labs/skills)
|
|
|
|
## See Also
|
|
|
|
* [SDLC and Tooling](sdlc.md)
|
|
* [Skills Ecosystem](skills-ecosystem.md)
|