Part Four: Live Demo

Live Demo — Governance You Can Use Monday

The Setup

The tools your developers are running right now — Claude Code, OpenClaw, Claude Flow — have varying levels of access control, permissions, and sandboxing. Some are quite good. None of them govern the agent.

The orchestrator runs locally. The cognition — the actual decisions about what actions to take — lives in a remote LLM. Current governance constrains the local tool. Nothing governs the remote decision-maker. That's the gap this demo fills.

What We Build Live

We take Claude Code — the tool with the strongest existing governance in this space — and add a Web4 governance plugin. Not a different tool. Not a replacement. An upgrade to the tool you already use.

The plugin adds:

Witnessed session provenance. Every action the agent takes is signed, hash-chained, and attested by independent observers. Not a log file — a tamper-evident record of what happened, who authorized it, and what trust level the agent had at the time.
Trust evolution from behavior. T3 scores (Talent, Training, Temperament) per project, per role. Trust accumulates from consistent behavior and decays without reinforcement. An agent that has been reliable for 1,000 actions earns more latitude than one that started today.
Role-contextual permissions. An agent trusted as a code reviewer is not automatically trusted as a deployment approver. Trust is scoped to what the agent has demonstrated competence in.
Governance config integrity. The governance configuration is not a JSON file the agent can overwrite. Policy is witnessed and hash-linked. Modifying your own governance is itself a witnessed action that triggers trust consequences.
Session accountability report. At session end: what the agent did, which actions were consequential (R7), how trust changed, what was witnessed, any anomalies flagged. One page your security team can read.

What You Leave With

A link to the plugin. Open source. Installable on your existing Claude Code setup. No migration, no rewrite, no disruption to your current workflow. The plugin hooks into Claude Code's existing lifecycle events — it adds governance alongside what you already have.

Monday morning, your agents have witnessed provenance and trust evolution. Same Claude Code. Same workflow. Different accountability.

You will understand:

What the plugin does and how it works
What the roadmap looks like — the future steps from basic witnessing to full Web4 entity participation
How to install it, configure it, and start getting accountability data from your agents immediately
What it does NOT do yet — honest about the current scope and the path forward

The Bigger Picture

The plugin is step one of the Web4 migration path. Context files → memory → trust scoring → attestation → federation. Each step stands alone. Each adds clear value. The plugin gives you the first three: context (what the agent knows), memory (what it observed), and trust scoring (how reliable it's been).

Attestation and federation come later — and when you're ready for them, the foundation is already in place. The governance grows with your needs. It doesn't require you to adopt the entire ontology on day one.

Background: The Orchestrator Landscape

For context on where the current tools stand and what governance gaps exist across the industry, see the orchestrator comparison. The short version: every tool constrains the tool. None governs the agent. The row of ❌ in the comparison table is what this plugin begins to fill.