The Harness station · Setup Pack · $49 · included with any plan
Stop explaining your project to every agent.
The files your agents read before they write. Nine questions in, you get one project brief in CLAUDE.md, AGENTS.md and your Cursor rules, an .mcp.json that references only servers which exist, and a QA & E2E playbook (TESTING.md) that makes agents show evidence. (On the ladder — Levels 0–5, after Dan Shapiro — that is the move from Level 2 to Level 3.)
One brief
The project section is byte-identical across all three tools. That is the point — one description of your project, not three that drift.
Checked, not generated
Only MCP servers that are published and still maintained upstream. Nothing from the archived repo, and none of the packages models keep inventing.
Twenty minutes
Every step in SETUP.md ends with how to confirm it loaded. A config that is silently ignored looks exactly like one that works.
Verification, not vibes
TESTING.md makes your agents show evidence: fail-first tests, mutation spot-checks, E2E rules — and a hook that refuses to let an agent stop on red.
Couldn’t an AI generate this? It will, and it will sound right. Ask one about Codex approval modes and there is a good chance you get --approval-mode, which is not a flag — the real one is --ask-for-approval. Copy the invented one and Codex rejects the command outright — not a wrong permission posture, just an agent that will not start. What you are paying for is that someone checked.
Setup Pack · The Harness
Nine questions. Then a working setup.
Everything below goes into the files your agents actually read. Nothing is sent anywhere — the pack is generated from these answers and downloaded straight to you.
The preview is the complete, working brief for one tool — not a teaser. The full pack adds the other tools, the MCP config, and TESTING.md, the QA & E2E playbook.
Or start the 7-day trial: the pack is included with every plan.