GW
HomeWorkAboutNotesStackLab
Experience atOmnicom Group

Driving financial systems and automation across global teams.

Builder ofEnterprise AI Systems

AI agents, audit workflows, and decision intelligence at scale.

Based inNew York City

Building systems for enterprises around the world.

HomeNotesWorkAboutStackLabPartners
XLinkedInGitHubEmailllms.txt
© 2026 · New York City

Starter Template: agents.md

Feb 3, 2026

This is a starting point for inspiration, not a copy-paste solution. Every project is different.

Better approach: Use the 9 Things for Effective Agent-Driven Development note as context, then ask your agent: "Help me design and implement an AGENTS.md (or CLAUDE.md) for this project based on these strategies." The agent can tailor it to your stack and workflow.

Below is a template that shows the structure. Adapt sections to your repo.


# Agent Configuration

## Project Architecture

### Directory Structure
- `./app`, Next.js application (App Router)
- `./components`, Shared React components
- `./lib`, Utilities, helpers, API clients
- `./types`, TypeScript type definitions

### Key Files
- `./app/layout.tsx`, Root layout, providers, global styles
- `./lib/api.ts`, API client and data fetching
- `./components/ui/`, Base UI components

### Navigation
Before searching the codebase, check for a NAV_MAP.md or README in the target directory. Read it first to understand what each area does.

---

## Behavioral Rules

### Rule of Threes
If the same error appears 3 times in a session:
1. STOP. Do not attempt a 4th fix.
2. Analyze the root cause, not the symptom.
3. Check devlog.md for similar past issues.
4. Propose a structural fix. Log the solution to devlog.md with timestamp.

### Before Starting Any Task
1. Read devlog.md for recent context and successful patterns.
2. If a similar task was done before, ask: "Should I follow the same approach as [previous task]?"
3. Confirm scope before executing. Ask clarifying questions if the task is ambiguous.

### Before Completing Any Task
- Remove unused imports and dead code (no commenting out).
- Remove stray console.logs unless explicitly debugging.
- Keep types correct; avoid `any` unless necessary.
- If a file exceeds ~300 lines, consider splitting.

### When Stuck
1. Don't loop. If 2 attempts fail, pause and explain what's not working.
2. Ask for help or suggest an alternative. "I'm not sure" is better than guessing and breaking things.

---

## Documentation

### devlog.md Format
After significant work, append:
**What / Why / How / Learned** (brief).

### When to Update devlog.md
After fixing a tricky bug, making an architectural decision, discovering a library gotcha, or completing a multi-step feature.

---

## Sub-Agents
Spawn sub-agents for parallel, independent tasks (e.g. UI, API, tests). Give each: the specific task, relevant files to read first, and design constraints. Include "Check devlog.md before starting."

---

## Tech Stack (example)
- Next.js 15, App Router, React 19, TypeScript (strict)
- Tailwind CSS; use CSS variables for theme
- [Your DB and external services, with links to current docs]
- When referencing library APIs, confirm the version in package.json. Don't assume latest.

---

## What NOT to Do
- Don't create new files when you can edit existing ones.
- Don't add dependencies without asking.
- Don't refactor unrelated code while fixing a bug.
- Don't leave TODO comments, do it or note it in devlog.
- Don't guess at file paths, search or ask. Explore the codebase first.

Back to 9 Things for Effective Agent-Driven Development.

Ask a follow-up about this note
$