AgentsMD Kit

A compact AGENTS.md starter kit for coding-agent projects.

This repository is for project-level agent instructions and lightweight task memory. It deliberately does not include the Multi-Agent skill; that lives in a separate repository so the roles/brief workflow can evolve independently.

What Is Included

• AGENTS.md: default project-level rules for agent behavior.
• templates/agent_memory/context.md: stable project context.
• templates/agent_memory/progress.md: current task state.
• templates/agent_memory/bugs.md: active issues, failed attempts, and repeated pitfalls.
• docs/Agent-Memory.md: explanation of the memory files and when to update them.

What Belongs Here

Use this kit when you want an agent to:

• read relevant files before non-trivial changes,
• keep changes scoped and traceable,
• record durable context outside the chat,
• preserve current task state across long sessions,
• keep AGENTS.md short instead of turning it into a giant playbook.

Quick Start

Copy these into a project root:

AGENTS.md
agent_memory/
  context.md
  progress.md
  bugs.md
  archive/

Create agent_memory/archive/ as an empty directory. Copy the three templates unchanged, then let the agent fill them during work.

Runtime Fit

AgentsMD Kit is plain Markdown and works on macOS, Linux, WSL2, and Windows. It has no package install step.

The template is model-agnostic, but stronger reasoning improves outcomes on large repositories, long debugging sessions, and tasks that require evidence tracking. For workflows like the author's, a Codex-style local agent with file access, command execution, and a GPT-5.5-class reasoning model at high or xhigh reasoning is the known-good profile.

Verification Checklist

After copying the kit into a project, confirm these files exist:

AGENTS.md
agent_memory/context.md
agent_memory/progress.md
agent_memory/bugs.md
agent_memory/archive/

Keep archive/ as a directory. The three active memory files should start from the templates in this repository, then be updated by the agent during real work.

Relationship To Multi-Agent Skill

AGENTS.md is project orientation. It answers: how should an agent behave in this repository?

The Multi-Agent skill is task orchestration. It answers: how should a complex task be scoped, delegated, reviewed, or handed off?

Keep them separate so global project rules stay small and the role brief workflow can be installed only when needed.