Harness Engineering

The harness is what separates "generic AI output" from "code that fits your project."

💡

Without harness: "Here's a React component" (generic)

With harness: "Here's a functional component using your design tokens, following your naming convention, tested against your business rules"

Setup

Harness is generated automatically during npx buildcrew:

npx buildcrew
# Step 3: Generate project harness? (Y/n) → scans your codebase
# Step 4: Pick additional templates

Or generate separately:

npx buildcrew init

Open Directory

.claude/harness/ is an open directory. Not limited to templates.

.claude/harness/
├── project.md          ← core (auto by init)
├── rules.md            ← core (auto by init)
├── erd.md              ← template
├── architecture.md     ← template
├── design-system.md    ← template
├── my-custom-notes.md  ← your own file
└── anything.md         ← agents read it all

Agent Routing

The orchestrator routes the right harness files to the right agents:

File	Routed to
`project.md`, `rules.md`	All agents
`erd.md`, `architecture.md`, `api-spec.md`	developer, reviewer, security-auditor, investigator
`design-system.md`	designer, design-reviewer
`glossary.md`, `user-flow.md`	planner, designer, browser-qa, canary-monitor
`env-vars.md`	developer, security-auditor
ALL files	reviewer, security-auditor, thinker, architect, qa-auditor

Managing

npx buildcrew harness          # Show status
npx buildcrew add              # List templates
npx buildcrew add erd          # Add template
npx buildcrew init --force     # Regenerate core files (backs up existing)

When to Update

Stack changes → architecture.md
New DB table → erd.md
New business rule → project.md or glossary.md
Bad pattern found → rules.md "What to avoid"
New API → api-spec.md

Design Review Templates