Agent Skills: Initialize Agent Memory

Initialize Agent Memory

Create an AGENTS.md file for this repository and symlink CLAUDE.md to it.

Steps

1. Check existing files:

   ls -la AGENTS.md CLAUDE.md 2>/dev/null; file AGENTS.md CLAUDE.md 2>/dev/null | grep -i link
   

2. Explore the codebase to understand:

   • Tech stack (check package.json, Cargo.toml, go.mod, etc.)
   • Project structure (key directories)
   • Build/test commands
   • Any existing documentation

3. Create AGENTS.md with these sections:

# Project Name

Brief description of what this project does.

## Tech Stack

- Language/runtime
- Key frameworks/libraries
- Database (if any)

## Project Structure

src/ # Source code tests/ # Test files ...

## Commands

```bash
# Install dependencies
<package-manager> install

# Run dev server
<command>

# Run tests
<command>

# Build
<command>

Architecture Notes

Key patterns, conventions, or decisions worth knowing.

4. **Create symlink**:
   ```bash
   ln -s AGENTS.md CLAUDE.md

Best Practices (from humanlayer.dev)

• Keep it lean: Under 300 lines. Agents can follow ~150-200 instructions reliably.
• Universal only: Only include what applies to every task. No task-specific guidance.
• Pointers not copies: Reference other docs (e.g., See docs/architecture.md) instead of embedding.
• No style rules: Don't use agents as linters. Use formatters/hooks instead.
• Three pillars: WHAT (tech stack, structure), WHY (purpose), HOW (commands, workflow).

Ask clarifying questions about the project before writing AGENTS.md.