The workflow

From zero to production-ready agent

Four steps to give your AI agent persistent memory, real skills, and deep project context.

Step 1

Initialize

aes init scaffolds a .agent/ directory tailored to your project type — ML, Web, DevOps, Research, or Assistant.

$ aes init
# Choose your domain:
#   ML, Web, DevOps, Research, Assistant
# Scaffolds .agent/ with:
#   agent.yaml, permissions.yaml,
#   skills/, memory/, commands/
Step 2

Define

Configure your agent’s identity in agent.yaml, define skills with runbooks, and set fine-grained permissions.

# agent.yaml
name: my-project-agent
version: "1.0"
aes_version: "1.3"
description: Full-stack web agent
skills:
  - scaffold
  - test
  - deploy
Step 3

Sync

aes sync translates your .agent/ directory into the native format of each AI tool. One source, six outputs.

$ aes sync -t claude
✓ Generated CLAUDE.md
✓ Generated .claude/settings.local.json
✓ Generated .claude/commands/skills/

$ aes sync -t cursor
✓ Generated .cursorrules

$ aes sync -t openclaw
✓ Generated .openclaw/
Step 4

Share

Publish skills and templates to the registry. Install proven patterns from the community. Reuse across every project.

$ aes publish --skill deploy
✓ Published [email protected]

$ aes install aes-hub/deploy@^1.0
✓ Installed to .agent/registry/

What each tool gets

Claude

CLAUDE.md + .claude/settings.local.json + .claude/commands/skills/*.md

Cursor

.cursorrules

Copilot

.github/copilot-instructions.md

Windsurf

.windsurfrules

OpenClaw

.openclaw/openclaw.json + workspace Markdown + SKILL.md files

Codex

AGENTS.md + .agents/skills/<id>/SKILL.md

Ready to start?

pipx install aes-cli && aes init