@rijuvashisht

Magneto AI

Magneto AI governance and security guardrails for OpenClaw agents — task planning, compliance hooks, and security checks across all channels.

Current version
v1.0.1
code-pluginCommunitysource-linked

openclaw-magneto

Magneto AI governance and security guardrails for OpenClaw agents. Task planning, compliance hooks, and security checks — across every channel.

npm License: MIT


What it does

This plugin wires Magneto AI into your OpenClaw gateway as the governance and reasoning layer for all AI engineering agents.

Tools added to every agent session

ToolDescription
magneto_analyzeScan the project and build AI memory (file index, modules, dependencies)
magneto_planGenerate a structured plan + security check for a task file
magneto_generateProduce a scoped implementation prompt (68% fewer tokens than raw prompts)
magneto_security_checkEvaluate risk of any command or action before running it

Compliance hook — before_tool_call

Every call to exec, bash, write, edit, or apply_patch passes through Magneto's security engine:

Risk levelWhat happens
LOWAllowed to proceed normally
MEDIUMProtected path detected → user must confirm via channel button or /approve
HIGHBlocked action pattern → requires explicit approval
CRITICALBlocked action + protected path → hard blocked, not permitted

Blocked patterns include: rm -rf, DROP TABLE, DELETE DATABASE, curl | bash, credential file access, and more.

Bundled skill

A magneto.SKILL.md is registered with the agent, teaching it when and how to invoke all four tools and follow Magneto's governance workflow.


Install

Requires magneto-ai globally on your gateway host:

npm install -g magneto-ai

Install the plugin:

# Users install it from their own gateway
openclaw plugins install clawhub:magneto-ai/openclaw-magneto
openclaw gateway restart

Quick start

  1. Install the plugin (above)
  2. Initialize Magneto in your project:
    cd /your/project
    magneto init
    
  3. Send a message from any channel: "Plan the auth feature"
  4. The agent calls magneto_analyzemagneto_plan → returns a structured plan with risk level

How Magneto works

User → Telegram / Slack / WhatsApp / Discord
          ↓
     OpenClaw Gateway
          ↓
     Agent (reads magneto.SKILL.md)
          ↓
     magneto_analyze   → understands the codebase
     magneto_plan      → structured plan + security check
     magneto_generate  → scoped implementation prompt (68% fewer tokens)
          ↓
     before_tool_call hook intercepts any destructive exec/write calls
          ↓
     Governed, safe response back to user

Why this matters

Without MagnetoWith Magneto
Agent has no project contextProject memory built from codebase
Every prompt dumps the whole repoOnly relevant files are included
No security checksEvery exec/write intercepted and evaluated
Agent can delete files, drop tablesBlocked at the hook level
No compliance trailRisk levels logged per action

Configuration

No configuration required after install. Optional: set the projectRoot parameter on any tool call to point to a specific project directory.


Links

Source and release

Source repository

rijuvashisht/Magneto

Open repo

Source commit

4286e79d922e4a59552f82516adfb5cc9f911397

View commit

Install command

openclaw plugins install clawhub:openclaw-magneto

Metadata

  • Package: openclaw-magneto
  • Created: 2026/04/19
  • Updated: 2026/04/19
  • Executes code: Yes
  • Source tag: develop

Compatibility

  • Built with OpenClaw: 2026.3.24-beta.2
  • Plugin API range: >=2026.3.24-beta.2
  • Tags: latest
  • Files: 8