The teamvince agent harness starter.
Plus a 60-second take from Vince.
Three things, free. Drop the harness into any new project and you have a working spec-first loop in ten minutes. Run your idea through the scope filter to know if it can ship in 14 days. And if you check the optional box, Vince records a 60-second Loom on whether your idea is ready to ship and how to scope it.
Send me the harness
No spam. Unsubscribe any time. Vince personally reviews the first 50 idea submissions per cohort cycle.
Check your email.
The harness, the scope filter, and the next step are on the way to your inbox. If nothing lands in five minutes, check spam, then email [email protected].
The harness, the filter, the take.
Most lead magnets are a teaser. This one is the actual thing Vince uses on his own projects, plus a tool that protects you from picking the wrong project, plus a 60-second outside read.
The agent harness starter
A downloadable repo with AGENTS.md, CLAUDE.md, five battle-tested slash commands (/ship, /diff-review, /verify, /scope, /loom-prep), one starter Skill setup, .env.example, .gitignore. Drop into any project. Edit the placeholders. Done.
The scope filter, interactive
An interactive worksheet that runs your project idea through the seven-question filter. Picks the lane (Internal Tool, Workflow Automation, AI Work Agent, Niche Micro-SaaS), names the graceful fallback if scope blows up, generates a draft AGENTS.md ready to commit. The same tool the cohort uses on Day 1.
A 60-second Vince Loom (limited)
If you check the box and submit your idea, Vince records a 60-second Loom on whether it is scoped right and what to cut. Capped at the first 50 per cohort cycle. Real human review, not a bot. If the queue is full, you still get the harness and the scope filter.
First 50 each cycleWhat you actually unzip.
. ├── AGENTS.md # Open standard, any agent reads it ├── CLAUDE.md # Claude-specific behavior on top of AGENTS.md ├── .env.example # Template for secrets, never commit the real one ├── .gitignore # Keeps .env out of git ├── .claude/ │ ├── commands/ │ │ ├── ship.md # /ship, lint and deploy in sequence │ │ ├── diff-review.md # /diff-review, plain-English diff explanation │ │ ├── verify.md # /verify, three-check verification before merge │ │ ├── scope.md # /scope, run an idea through the scope filter │ │ └── loom-prep.md # /loom-prep, generate the mid-cohort Loom script │ └── skills/ │ └── README.md # Where Skills go, plus the install pattern └── README.md # 10-minute quick start
Why two spec files? AGENTS.md is the open standard stewarded by the Linux Foundation since December 2025. Cursor, Codex, Copilot, Windsurf, Gemini CLI, and Claude Code all read it. Sixty-thousand-plus projects use it. CLAUDE.md adds Claude-specific behavior on top: model picks (Opus plans, Sonnet executes), plan-mode preferences, the verify-before-trust check. When a better agent ships in six months, your spec moves with you. The asset is the harness, not the model.
SMEs who want to ship.
PMs, operators, consultants, and indie builders who have already opened Claude Code at least once. You have the domain expertise. AI is the force-multiplier on top, not a replacement for being a worse developer.
Right fit
You have an idea you'd build if you weren't stuck in "I need a developer" mode. You're willing to own every diff. You want to leave with something real, not a toy.
Wrong fit
You want a tour of every AI tool. You want passive video. You've never opened Claude Code. Start there first, then come back. You want to vibe-code without reviewing the code.
What's next
Use the harness on something real. Want a 4-week structured push to ship one tool live with a small cohort? The May 12 founding cohort opens at $497. Future cohorts $697+.
FAQ
What's in the harness, exactly?
A downloadable repo with two spec templates (AGENTS.md, CLAUDE.md), five slash commands as markdown files, an .env.example, a .gitignore, and a README with a 10-minute quick start. About 14 files, 19 KB unzipped. You unzip, fill in the placeholders, you have a working spec-first loop.
Do I need Claude Pro to use the harness?
To run the slash commands and use plan mode, yes. Claude Pro is $20 a month and is the minimum for serious agentic work. You can read AGENTS.md and CLAUDE.md as plain documents without Pro, but the loop only fires inside Claude Code.
What's the Vince Loom take, exactly?
If you submit a project idea and check the consent box, Vince records a 60-second Loom video addressing one specific question: is this scoped right to ship in 14 days, and if not, what to cut. Capped at the first 50 submissions per cohort cycle (monthly). Real human review. If the queue is full when you submit, you still get the harness and scope filter, plus a note when the next cycle opens.
Does this replace the course?
No. The harness is the starter scaffold. The course is the structured 4-week push to actually ship one tool live with a small cohort, async Loom mid-cohort demos, project approval, build buddies, and demo day. The harness is the lead magnet. The cohort is the product.
Why is this free?
Two reasons. One, this is the harness Vince uses on his own projects, so giving it away costs nothing extra. Two, it qualifies the lead. People who use the harness and want to push further into a structured cohort are the right buyers. People who don't get value from the harness are not the right fit.
How current does the harness stay?
The harness is versioned. When Anthropic, OpenAI, or Cursor ships something material (new model, new feature like Skills, new MCP standard), Vince updates the harness and email subscribers get the diff. Same living-syllabus mechanic the cohort uses.
From Chatbot to Builder.
The 4-week founding cohort starts Tuesday May 12, 6pm ET. One shipped tool, one workflow you trust, one next-project plan. Founding price $497. Future cohorts $697+.
See the course