# Juggernaut Harness

> An open-source skill for Claude, Claude Code and Codex that treats your AI agents as a business process: a BPMN map, a nine-tab console for health, cost, security and every report, and ranked fixes. Bring your own logs.

Juggernaut Harness is an open-source skill (Apache-2.0) for Claude.ai, Claude Cowork, Claude Code and OpenAI Codex that treats AI agents as a business process. Describe the agent system you want, or point the skill at one you already have. It draws a BPMN process map of every agent, scores each agent's health and cost from run logs, ranks the fixes worth making with an estimated saving, and reviews whether the agents are safe. Everything appears in one offline HTML viewer with nine tabs: Map, Process, Health, Runs, Optimize, Security, Readiness, Reports and Source.

It is not an agent framework. It sits above whatever already runs the agents (n8n, LangChain, LangGraph, CrewAI, the OpenAI Agents SDK, the Claude Agent SDK, Google ADK, Bedrock, Azure AI Foundry, Lovable and more). It works with your own logs: it imports n8n execution history, OpenTelemetry traces and Langfuse exports, or reads logs written by a small recorder file. Juggernaut Dev never receives any user data: there is no Juggernaut server, account, API key, telemetry or analytics.

Who it is for: developers building multi-agent systems in code; n8n and low-code builders; newcomers who have never seen a BPMN diagram; enterprise architects and security reviewers.

## Install
- Claude.ai and Cowork: download the skill file from the latest release (https://github.com/ORG/juggernaut-harness/releases/latest) and add it in your Skills settings. Then ask in plain language.
- Claude Code: run `/plugin marketplace add ORG/juggernaut-harness` then `/plugin install juggernaut-harness@juggernaut`.
- OpenAI Codex: copy the skill folder into `.agents/skills/` in your repository, then ask Codex in plain language.

## Pages
- [Home](https://harness.juggernautdev.com/): overview, console tour, BPMN explainer, platforms, install, roadmap
- [Example use cases](https://harness.juggernautdev.com/use-cases): what you type, what Claude asks and what you get in Claude Code, Claude.ai and n8n
- [Live demos](https://harness.juggernautdev.com/demo): the real console built from sample data

## Platform guides
- [n8n](https://harness.juggernautdev.com/platforms/n8n): n8n already stores every execution, with the token usage of each model call. Import the workflow file and its history; nothing in your workflows changes.
- [OpenTelemetry](https://harness.juggernautdev.com/platforms/otel): Many agent frameworks emit OpenTelemetry GenAI spans with the model and token usage of every call. Point their exporter at a collector you run, save the traces to a file, and import it.
- [Langfuse](https://harness.juggernautdev.com/platforms/langfuse): Langfuse already stores every trace and generation with its model, token usage and cost. Export the observations with your own keys and import them.
- [OpenAI Agents SDK](https://harness.juggernautdev.com/platforms/openai-agents): The Agents SDK has its own tracing. Bridge it to OpenTelemetry with an instrumentation library such as OpenInference and import the traces, or record each run's usage directly.
- [Google ADK](https://harness.juggernautdev.com/platforms/google-adk): ADK emits OpenTelemetry spans for agent runs and model calls. Send them to a collector you run and import the file.
- [Strands Agents](https://harness.juggernautdev.com/platforms/strands): Strands has built-in OpenTelemetry support. Point its OTLP exporter at your collector and import the traces.
- [Mastra](https://harness.juggernautdev.com/platforms/mastra): Mastra exports traces in OpenTelemetry format. Send them to your collector and import the file.
- [Vercel AI SDK](https://harness.juggernautdev.com/platforms/vercel-ai): The AI SDK emits OpenTelemetry spans for each call when telemetry is enabled. Register an OpenTelemetry exporter in your app, point it at your collector, and import the traces.
- [CrewAI](https://harness.juggernautdev.com/platforms/crewai): Instrument CrewAI with an OpenTelemetry instrumentation library such as OpenInference or OpenLIT and import the traces, or call the recorder from your agents' model wrapper.
- [Pydantic AI](https://harness.juggernautdev.com/platforms/pydantic-ai): Pydantic AI instruments agent runs with OpenTelemetry. Send the spans to your own collector and import the file.
- [LlamaIndex](https://harness.juggernautdev.com/platforms/llamaindex): Instrument LlamaIndex with an OpenTelemetry instrumentation such as OpenInference and import the traces.
- [LangChain and LangGraph](https://harness.juggernautdev.com/platforms/langchain): One callback handler records every model call. LangGraph passes the node name in the callback metadata, so name your graph nodes with the blueprint's step ids.
- [Claude Agent SDK](https://harness.juggernautdev.com/platforms/claude-agent-sdk): Each agent's query ends with a result message that carries its usage and cost. Record it once per agent.
- [Anthropic API](https://harness.juggernautdev.com/platforms/anthropic-api): Wrap each `messages.create` call. Cache reads and writes are counted correctly.
- [OpenAI API](https://harness.juggernautdev.com/platforms/openai-api): Wrap each call. `usage_from` reads both Chat Completions and Responses usage, including reasoning and cached tokens.
- [Amazon Bedrock](https://harness.juggernautdev.com/platforms/bedrock): Wrap each Converse or InvokeModel call. Model names such as `us.anthropic.claude-sonnet-5-v1:0` are matched to the price table automatically.
- [Bedrock AgentCore](https://harness.juggernautdev.com/platforms/agentcore): AgentCore Runtime hosts agents built with any framework. Add the recorder inside the agent code, or import the OpenTelemetry traces AgentCore Observability produces.
- [Azure AI Foundry](https://harness.juggernautdev.com/platforms/azure-foundry): Wrap each Azure OpenAI or Foundry model call, or export Foundry's OpenTelemetry traces. Pass the underlying model name to the recorder, not your deployment name.
- [Lovable](https://harness.juggernautdev.com/platforms/lovable): Lovable apps run their AI work in Supabase Edge Functions. Add the TypeScript recorder, keep the events in a table in your own Supabase project, and download them with one command.
- [xAI Grok](https://harness.juggernautdev.com/platforms/grok): The xAI API is OpenAI-compatible, so the recorder reads its usage, including reasoning tokens.

## Live demo consoles
- [Claims settlement](https://harness.juggernautdev.com/demo/claims.html): A 25-agent claims process with 10 human approvals. The flagship example: the BPMN map, health, cost, ranked fixes, the security review and every report.
- [n8n support triage](https://harness.juggernautdev.com/demo/n8n.html): Built entirely from an n8n workflow export and its execution history, with no hand-written blueprint. It finds a drafter that got worse after its system message was edited.
- [All five pipelines](https://harness.juggernautdev.com/demo/all-pipelines.html): One console across claims, support triage, research, marketing content and code review, to see how the viewer handles several pipelines at once.

## Roadmap
- Available now: the free skill.
- Coming soon: Juggernaut Enterprise, a delivery kit deployed in your own cloud.
- In the future: Juggernaut Cloud, a hosted application. Opt-in, metadata only.

## Optional
- [Full site text](https://harness.juggernautdev.com/llms-full.txt)
- [Source code](https://github.com/ORG/juggernaut-harness)
- [Juggernaut Dev](https://juggernautdev.com): the company behind it