Factory Floor

A startup coach that turns Claude into a thinking partner for prioritization and execution. It won't tell you what to do — it'll ask the questions you're avoiding.

Works with Claude Code, Claude Desktop, OpenAI Codex, and any agent that supports the open agent skills standard.

Claude Code

npx @swiftner/factory-floor

Installs to ~/.claude/skills/factory-floor/. Triggers automatically when you talk about priorities, bottlenecks, what to build, or why growth is flat.

OpenAI Codex

npx @swiftner/factory-floor

Installs to ~/.codex/skills/factory-floor/. Codex picks it up automatically — trigger with /skills or $ to mention it directly, or just describe your problem and it will activate implicitly.

Or install via ClawHub:

clawhub install factory-floor

Claude Desktop

1. Open Claude Desktop and create a new Project
2. Set the contents of SKILL.md as the project's Custom Instructions
3. Upload these files as Project Knowledge:
   • stages/pre-revenue.md
   • stages/restart.md
   • stages/growth.md
   • stages/scaling.md
   • references/intake.md
   • references/probes.md
   • references/playbooks.md
   • references/weekly-review.md
   • references/state.md

That's the whole skill — ten files, about 60KB.

Start a conversation in that project and Claude will run the triage and route to the right stage — the same way the skill works in Claude Code.

State layer in Claude Desktop. Claude Code writes .factory/context.md and .factory/journal.md to your working directory so you don't re-explain yourself each session. Claude Desktop has no filesystem, so the state layer works manually: paste your current context.md into the project or the chat at the start of a session, and when Claude suggests a state update (after a diagnosis, experiment, or weekly review), copy the edited version back into your project files. The schema and triggers in references/state.md are the same either way.

What it does

You say "should we build Slack integration?" and instead of a pros-and-cons list, you get:

• "What's your retention like — do people who try it stay?"
• "Where are new trials coming from? Is that number growing?"
• "If retention is 90% but trials are flat... is the real problem that not enough people know you exist?"
• "Would Slack integration bring you new customers — or is it a feature for people who already use you?"

It asks, you decide. Three areas where founders consistently fool themselves:

You're building when you should be selling. The product works — users who find it stay. But nobody's finding it. You don't need features. You need to exist in more people's heads.

You're doing five things and finishing none. Each parallel project costs ~20% in context-switching. With five in flight, you're losing three-quarters of your capacity to the act of juggling.

You're optimizing the wrong thing. Marketing is generating leads but onboarding can't absorb them. You just created inventory, not progress. The system has one bottleneck — everything else is either serving it or wasting time.

It adapts to your stage

A quick triage loads the right playbook:

| Stage | What it covers | |---|---| | Pre-revenue | No customers yet? Don't build. Five tests before you write code. Napkin math. The Mafia Offer. | | Restart | Had customers, lost them. Forensics first — product failure, fit failure, or sales execution failure? Churned customer interviews. Restart sequence. | | Growth | Have customers, small team. Find the constraint, exploit it, run the system. Two-week experiment cycles, WIP limits, brand building vs. activation. | | Scaling | $100K+ MRR or 10+ people. Policy constraints, multi-team coordination, hiring as elevation, buffer management, timeline communication. |

Two modes

Field experiments on AI advice disagree about who benefits — until you sort them by task. On structured, checkable work an assistant lifts the least experienced people most. On ambiguous, judgment-dependent work it makes them worse. Same model, same person; the task decides.

So Factory Floor runs in two modes and sorts every question before answering:

Procedure — funnel arithmetic, the weekly review, napkin math, buffer sizing, WIP limits, interview scripts, CEP mapping, the positioning sprint. The answer is checkable, so it's maximally directive. No hedging, no options.

Judgment — what your constraint is, whether to pivot, whether this is the right segment, why they churned. The answer is a claim about your business that nobody can verify, so it won't prescribe. It takes your theory first, then designs the test that tells the two of you apart — and names a constraint only when a number supports it.

Practically: it will be more opinionated than most tools about how you work, and markedly less confident than most tools about what's wrong with your company.

The prediction ledger

Startups are a terrible place to learn, because feedback is slow, noisy and confounded — so experience alone never corrects your instincts. Well-prepared and poorly-prepared founders turn out to be equally confident.

Factory Floor takes your prediction before it comments, writes it to .factory/journal.md with a confidence number, and scores it when the experiment returns. Each quarter it shows your hit rate. 40–60% right means you're calibrated; above 70% means you're only predicting safe things.

The frameworks underneath

Jobs To Be Done, Theory of Constraints (Goldratt), the Customer Factory (Maurya), How Brands Grow (Sharp), marketing strategy discipline (Ritson), and strategic thinking (Rumelt, Clausewitz, Dixit & Nalebuff).

The skill doesn't explain any of them — the model already knows them. It carries the procedures, the thresholds, and the posture. That's why it fits in ten short files.

Things you can ask

| You say | It does | |---|---| | "What should we work on this week?" | Runs the triage. Finds the bottleneck. Helps you pick three priorities. | | "We have no customers yet" | Problem validation before code. Napkin math, five tests, Mafia Offer. | | "Should we build X or focus on sales?" | Asks where the constraint is and whether X serves it. | | "We're spread too thin" | Figures out what to stop. WIP audit, team state check. | | "Why do deals ghost?" | Walks through the four forces. Where is the deal dying? | | "Nobody knows we exist" | Maps Category Entry Points. Audits physical availability. Builds a reach cadence. | | "How long will this take?" | Helps you build an honest buffer instead of giving you a number. | | "Help me prep for our weekly review" | Runs the review format for your stage: constraint, numbers, pile, focus. |

The weekly review

Same structure, scaled to your stage:

Every review opens by scoring the prediction you made last time.

• Pre-revenue (12 min) — How many conversations? What did we learn? Has the hypothesis survived?
• Growth (12 min) — Name the constraint, check throughput, find where work piles up, set 3 priorities.
• Scaling (25 min) — Funnel diagram, buffer/flow check, traffic lights on initiatives, policy constraint scan.

When an experiment reaches its deadline you get a separate decision point, not another review: persevere, pivot, or pause. Recording an outcome without forcing that fork is how a wrong constraint survives a quarter.

Credits

• Clayton Christensen — The Innovator's Dilemma, Competing Against Luck. Jobs To Be Done.
• Bob Moesta — Demand-Side Sales 101. Forces of progress, switch interviews.
• Tony Ulwick — Jobs to be Done: Theory to Practice. Outcome-Driven Innovation.
• Eli Goldratt — The Goal, Critical Chain. Theory of Constraints.
• Ash Maurya — Running Lean, Scaling Lean. Customer Factory, Lean Canvas, Mafia Offer.
• Byron Sharp — How Brands Grow. Mental and physical availability.
• Mark Ritson — Mini MBA in Marketing. Marketing strategy discipline, STP, positioning.
• Richard Rumelt — Good Strategy Bad Strategy, The Crux. The kernel of strategy, bad strategy signs, proximate objectives.
• Carl von Clausewitz — On War. Fog, friction, center of gravity, culminating point, moral forces.
• Avinash Dixit & Barry Nalebuff — The Art of Strategy. Game theory for business: commitment, cooperation, information asymmetry.
• Les Binet & Peter Field — The Long and the Short of It. Brand building vs. activation budget allocation.
• April Dunford — Obviously Awesome. Positioning from JTBD.
• Douglas Hubbard — How to Measure Anything. Estimation calibration.

Made by Swiftner.