trae-evolve 0.2.0

Universal self-evolution layer for AI agents — brings Hermes-style skills, memory, and self-improvement to any agent

trae-evolve 🧬

Universal self-evolution layer for AI agents — brings Hermes-style skills, memory, and self-improvement to Trae IDE and beyond.

Your AI agent learns from experience. After completing complex tasks, it automatically saves reusable workflows as skills and remembers facts about your project. Next time, it loads that knowledge and gets better.

---

Quick Start

1. Install

pip install trae-evolve

2. Initialize your project

cd your-project
trae-evolve init
trae-evolve trae-sync

3. Restart Trae IDE

That's it. Your AI agent now automatically evolves.

---

How It Works

Agent completes a complex task
         ↓
Automatically calls MCP tools → skill_manage(action='create', ...)
         ↓
Knowledge saved to .evolve/skills/ → synced to .trae/skills/
         ↓
Next session → Agent reads skills → reuses proven workflows
         ↓
Curator tracks usage → stale after 30 days → archived after 90 days

MCP Tools (Agent calls these automatically)

Tool	What it does	When Agent calls it
skill_manage	Create/patch/delete skills	After complex tasks
memory	Add/update/remove memory	After learning new facts
skill_view	Read a skill's full content	Before starting tasks
skills_list	List all available skills	Before starting tasks

---

Commands

Skills (procedural knowledge — "how to do X")

trae-evolve skills list                    # List all skills
trae-evolve skills view <name>             # View a skill
trae-evolve skills create <name> -d "..." -c "..."  # Create
trae-evolve skills patch <name> -o "old" -n "new"    # Patch
trae-evolve skills delete <name> -y       # Delete

Memory (declarative knowledge — facts)

trae-evolve memory read                    # Read all
trae-evolve memory read memory             # Agent notes only
trae-evolve memory read user               # User profile only
trae-evolve memory add memory "..."        # Add agent note
trae-evolve memory add user "..."          # Add user preference
trae-evolve memory replace memory "old" "new"  # Replace
trae-evolve memory remove memory "..."     # Remove

Trae Integration

trae-evolve trae-sync          # Generate .trae/ files (rules + skills + mcp.json)
trae-evolve inject             # Generate AGENTS.md / CLAUDE.md / .cursorrules

Lifecycle Management

trae-evolve curator status     # Show skill lifecycle status
trae-evolve curator run        # Run maintenance cycle
trae-evolve curator pin <name> # Prevent skill from being archived
trae-evolve status             # Overall status

---

Directory Structure

Global (shared across projects)

~/.evolve/
├── skills/          ← Global skills
├── memories/
│   ├── MEMORY.md    ← Agent notes (2200 char limit)
│   └── USER.md      ← User profile (1375 char limit)
└── .archive/        ← Archived skills

Per-project (created by trae-evolve init)

your-project/
├── .evolve/
│   ├── skills/      ← Project-specific skills
│   └── memories/    ← Project-specific memory
└── .trae/           ← Created by `trae-evolve trae-sync`
    ├── mcp.json     ← MCP Server config
    ├── rules/
    │   └── evolve-project-rules.md  ← Always-apply rules
    └── skills/      ← Synced Trae-native skills

---

SKILL.md Format

---
name: deploy-to-prod
description: How to safely deploy to production
version: 1.0.0
category: devops
tags: [deployment, production]
---

# Deploy to Production

## Steps
1. Run tests: `pytest -x`
2. Build: `make build`
3. Deploy: `./deploy.sh`

## Pitfalls
- Never deploy on Fridays
- Check DB migrations first

Memory Format

Entries separated by §:

Uses Edge browser
§
NVIDIA GPU + CUDA
§
Project uses FastAPI + PostgreSQL

---

Trae IDE Integration

trae-evolve trae-sync generates:

1. .trae/mcp.json — Connects Trae to evolve MCP Server
2. .trae/rules/evolve-project-rules.md — Always-apply rules with skill index + memory
3. .trae/skills/ — Skills in Trae-native format

After restart, Trae's AI agent gets 4 new MCP tools and reads the rules. It will:

• Automatically save skills after complex tasks
• Automatically remember user preferences
• Load existing skills before starting work

---

Agent Compatibility

Agent	Integration	Command
Trae IDE	MCP + Rules + Skills (native)	trae-evolve trae-sync
Claude Code	CLAUDE.md	trae-evolve inject
Codex	AGENTS.md	trae-evolve inject
OpenCode	AGENTS.md	trae-evolve inject
Cursor	.cursorrules	trae-evolve inject

---

Requirements

• Python 3.9+
• pyyaml (installed automatically)

---

License

MIT