Why Zigrix

Most agent-heavy workflows break down in the same places:

• work gets delegated, but not consistently tracked
• specialist routing depends on memory, not policy
• outputs arrive without structured evidence
• final reporting is manual and hard to audit

Zigrix gives that flow a control surface: dispatch, worker lifecycle, evidence, and finalization in one runtime.

Is Zigrix for you?

Zigrix is a strong fit if you want:

• one-time operator setup, then agent-driven execution
• repeatable orchestration rules instead of ad-hoc delegation
• recoverable local runtime state for task progress and reports
• OpenClaw compatibility without requiring a plugin-based architecture

Zigrix may be a poor fit if your main goal is a hosted, multi-tenant control plane.

Supported environments

• Operating systems: macOS, Linux
• Runtime: Node.js (current LTS recommended)
• Package manager: npm
• OpenClaw: optional, but recommended for full orchestration flow

Zigrix still works as a standalone CLI when OpenClaw is not installed.

Quick Start

1) Install

npm install -g zigrix

From source:

./install.sh

2) Onboard

zigrix onboard

3) Verify

zigrix doctor
zigrix dashboard

That is the intended entry flow: install → onboard → done.

Usage model: human vs agent

| Role | Primary responsibility | |---|---| | Human operator | Install Zigrix, run zigrix onboard, verify readiness (zigrix doctor), then step out of day-to-day orchestration | | OpenClaw / automation agents | Run task/worker/evidence/report commands to execute and complete orchestration work | | Human (maintenance mode) | Use zigrix configure or zigrix reset only when reconfiguration or recovery is needed |

This split keeps normal operation agent-driven while keeping setup and governance human-controlled.

Core Workflow

Typical operational flow:

Human setup
  install -> zigrix onboard -> doctor check

Agent execution
  zigrix task dispatch
    -> worker prepare/register/complete
    -> evidence collect/merge
    -> task finalize + report

Common commands:

# Dispatch orchestration work
zigrix task dispatch --title "Implement auth module" --description "..." --scale normal --json

# Check runtime health
zigrix doctor

# Launch dashboard
zigrix dashboard --port 3838

# Maintenance surfaces
zigrix configure --section agents
zigrix configure --section skills
zigrix reset state --yes

What onboard does

zigrix onboard prepares runtime and integration in one pass:

1. creates runtime directories from zigrix.config.json (paths.*) and default config/state structure
2. seeds rule files from bundled templates
3. ensures zigrix is reachable from the runtime-visible PATH
4. detects OpenClaw and imports agents from openclaw.json
5. registers bundled OpenClaw skills into ~/.openclaw/skills/ (including oz and the zigrix-* packs)
6. leaves the environment ready for agent-led orchestration

OpenClaw Integration

When OpenClaw is present, Zigrix is optimized for this model:

• import agent definitions and normalize roles
• establish orchestrator ownership for task execution
• register bundled OpenClaw skills for agent readiness checks
• expose /oz as the public Zigrix chat entrypoint after onboarding
• keep CLI reachability stable for OpenClaw runtime

After zigrix onboard, the expected chat-side entrypoints are:

• /oz fix the auth flow
• 이거 맡겨
• route this through Zigrix

/oz is force-delegate. Natural-language delegation is judged semantically by the installed skill and should enter the standard Zigrix chain (zigrix task dispatch → orchestrator spawn) instead of falling back to direct execution.

Read the full integration contract in docs/openclaw-integration.md.

Dashboard

Zigrix ships with a built-in dashboard for real-time visibility into task orchestration.

zigrix dashboard --port 3838

Task Detail

Track execution units, evidence submissions, and spec previews — all in one view.

Event Log

Every state change is recorded as an immutable event — from dispatch through finalization.

Live Progress

Monitor in-progress tasks with real-time execution unit status and one-click cancellation.

Task Spec

Full task specification with metadata, scope definition, and routing rules — visible before and during execution.

Docs

• Quickstart
• Install
• OpenClaw Integration
• CLI Spec
• Architecture
• Troubleshooting

Contributing

See CONTRIBUTING.md.

Support

See SUPPORT.md.

Security

See SECURITY.md.

License

Apache-2.0