Magneto AI governance and security guardrails for OpenClaw agents — task planning, compliance hooks, and security checks across all channels.
Install
openclaw plugins install clawhub:openclaw-magnetoopenclaw-magneto
Magneto AI governance and security guardrails for OpenClaw agents. Task planning, compliance hooks, and security checks — across every channel.
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
| Tool | Description |
|---|---|
magneto_analyze | Scan the project and build AI memory (file index, modules, dependencies) |
magneto_plan | Generate a structured plan + security check for a task file |
magneto_generate | Produce a scoped implementation prompt (68% fewer tokens than raw prompts) |
magneto_security_check | Evaluate 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 level | What happens |
|---|---|
| LOW | Allowed to proceed normally |
| MEDIUM | Protected path detected → user must confirm via channel button or /approve |
| HIGH | Blocked action pattern → requires explicit approval |
| CRITICAL | Blocked 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
- Install the plugin (above)
- Initialize Magneto in your project:
bash cd /your/project magneto init - Send a message from any channel: "Plan the auth feature"
- The agent calls
magneto_analyze→magneto_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 Magneto | With Magneto |
|---|---|
| Agent has no project context | Project memory built from codebase |
| Every prompt dumps the whole repo | Only relevant files are included |
| No security checks | Every exec/write intercepted and evaluated |
| Agent can delete files, drop tables | Blocked at the hook level |
| No compliance trail | Risk 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.