Skip to content

quickstart-en.md

Latest commit

 

History

History
138 lines (104 loc) · 3.99 KB

quickstart-en.md

File metadata and controls

138 lines (104 loc) · 3.99 KB

MOLT Quickstart Guide

What is MOLT?

MOLT (蜕界) is a reflexive co-evolution engine for OpenClaw. It observes your agent system, identifies structural friction (recurring problems that stem from how your system is organized), and proposes evidence-based improvements at three levels: task, team, and host environment.

Prerequisites

  • OpenClaw installed at ~/.openclaw/
  • Python 3.6+ available in PATH

Install

git clone <molt-repo-url>
cd molt
bash product/install.sh

This installs:

  • Data directory at ~/.openclaw/molt/ (CLI, schemas, evidence store)
  • OpenClaw Skill at ~/.openclaw/skills/molt/ (use /molt in any session)
  • OpenClaw Agent at ~/.openclaw/workspace-molt/ (background analysis with cron)

To install only specific components: bash product/install.sh data|skill|agent

First Run: Discover Your System

bash ~/.openclaw/molt/molt-cli.sh discover

This scans your OpenClaw installation and builds a System Model (model.json) — a structured description of your system: how many agents you have, their roles, collaboration mechanisms, and host capabilities.

Example output:

Model created: ~/.openclaw/molt/model.json
  Agents: 3 (main, researcher, writer)
  Topology: hub_spoke
  Multi-agent: True
  Dreaming: False
  Sessions: 142
  Shared artifacts: 5

View the model summary anytime:

bash ~/.openclaw/molt/molt-cli.sh model

First Scan

Via Skill (interactive)

In any OpenClaw session, type:

/molt scan

Via Agent (automatic)

If you installed the agent form and configured cron, MOLT runs daily at 9am and produces a weekly report on Mondays at 10am.

Via Claude Code

/molt scan

A scan reads your recent sessions, analyzes them against the System Model, and produces structured findings:

  • Frictions — observed problems with source references
  • Attributions — root cause hypotheses (agent / team / environment level)
  • Evidence — tracking whether problems are being addressed
  • Evolution proposals — suggested improvements

All findings are appended to ~/.openclaw/molt/molt.jsonl.

Review Findings

# See current status
bash ~/.openclaw/molt/molt-cli.sh status

# List active frictions
bash ~/.openclaw/molt/molt-cli.sh frictions

# Approve a friction (marks it as reviewed)
bash ~/.openclaw/molt/molt-cli.sh approve FRI-20260410-001

# Trace the full evidence chain for any object
bash ~/.openclaw/molt/molt-cli.sh chain FRI-20260410-001

Or use the interactive review:

/molt review

Weekly Report

bash ~/.openclaw/molt/molt-cli.sh report

Generates a markdown report at ~/.openclaw/molt/reviews/week-YYYY-WNN.md summarizing friction counts, attribution distribution, evidence status, and active proposals.

Where Data Lives

~/.openclaw/molt/
├── model.json          # System Model (your system's structure)
├── molt.jsonl          # Evidence Store (all MOLT objects, append-only)
├── molt-cli.sh         # CLI tool
├── schema/             # JSON schemas for validation
├── reviews/            # Weekly reports
└── backups/            # Automatic backups

Key Commands

Command Description
scan Run analysis (requires LLM — use via /molt skill or agent)
discover Build/update System Model
model Display model summary
model-diff Check for system changes since last scan
status Current evidence store summary
frictions List active frictions
chain <ID> Trace evidence chain
report Generate weekly report
approve <ID> Approve a friction
promote <CAN-ID> Promote candidate to pattern
propose <level> <title> Create evolution proposal
backup Backup evidence store

Next Steps

  • Run your first scan and review the findings
  • Approve or reject frictions based on your judgment
  • Watch for patterns emerging across multiple scans
  • Consider evolution proposals when evidence is strong enough