Quick Start

Get deterministic knowledge governance running in 5 minutes.

1. Initialize the harness config

kcp-harness init

This creates harness.yaml in your project root:

version: "1.0"

governance:
  domains:
    - manifest: "./knowledge.yaml"
      paths:
        - "docs/"
        - "src/"

  policy:
    fail_closed: true
    audit_all: true
    max_units: 5
    strict: false

audit:
  path: ".kcp-harness/audit.jsonl"

2. Generate agent integration files

Pick your agent:

# Claude Code
kcp-harness integrate claude-code

# Cursor
kcp-harness integrate cursor

# GitHub Copilot
kcp-harness integrate copilot

# All agents at once (dry-run first)
kcp-harness integrate claude-code --dry-run

This generates the MCP config and rules file for your agent. See Agent Integrations for agent-specific setup.

3. Validate the config

kcp-harness check

4. Start coding

Your agent now routes knowledge access through the harness. When it tries to read a governed file:

1. The harness classifies the tool call as knowledge-navigation
2. Runs the 13-gate cascade via kcp-agent
3. Returns content only if all gates pass
4. Logs the decision to the audit log

What happens under the hood

You: "Explain the authentication flow"

Agent: calls kcp_plan(task="explain auth flow", manifest="./knowledge.yaml")
  → Harness: runs 13-gate cascade
  → Returns: plan with 3 eligible units

Agent: calls kcp_load(task="explain auth flow", manifest="./knowledge.yaml")
  → Harness: loads eligible units, tracks budget
  → Returns: unit content + metadata

Agent: uses loaded knowledge to answer your question
  → Harness: logs decision trace to audit.jsonl

Next Steps

• Configuration — customize harness.yaml
• Architecture — understand the proxy pipeline
• MCP Tools — all available governance tools