Drop one folder into your project. Claude Code now must check your positioning says something specific, verify your copy answers 'who is this for?' and 'why does it matter?', plan how a test ends before it starts, and confirm any AI performance claims have a real source — or it cannot mark the work done.
Not edge cases. The default outcome when no growth quality enforcement exists.
The positioning passed every internal review. It failed the "so what?" test with every prospect. Competitors said exactly the same thing. Three steps into the chain of consequences, nobody could explain why it mattered for their specific problem.
The hero copy tested well with the team. "Who cares?" — there was no audience identifiable from the copy alone. "Why now?" — there was no urgency. It could have been written for any product, any year, any market.
The experiment ran for two weeks. It looked good at week two, so they stopped. It looked good because of random variation — the sample wasn't there yet. The feature shipped. The lift didn't hold in production.
The claim shipped without a source. It wasn't wrong — but it wasn't sourced, the population wasn't scoped, and the baseline wasn't stated. A journalist asked for the methodology. There wasn't one.
One rule blocks messaging from shipping without sourced claims. Five rules enforce how every piece of growth work is done.
Each rule requires a specific output with real values. Claude cannot say copy passed without showing the actual 'so what?', 'who cares?', and 'why now?' test results.
An agent that skips the positioning audit cannot identify the audience from copy alone or verify the "so what?" chain terminates in ≤ 2 steps.
That is the enforcement.
Each specialist has a narrow focus. All produce output at specific file paths — no chat-only results.
Reviews positioning, copy, and experiment designs against all 6 rules. Verdict: pass, fix these things, or blocked. Writes the report to a real file.
Designs experiments before they start — calculated sample size, exactly when to stop, and what decision you'll make in each of the four possible outcomes. Writes it to a real file.
Reviews any copy that makes claims about AI. Checks every number has a source, every claim has a defined scope, and every 'AI' label says what the AI actually does. Writes the report to a real file.
The difference is questions answered before work ships, not problems discovered after it does.
Installs only skills/ and .claude/agents/. Non-destructive — won't overwrite existing files.
positioning-audit before writing any copy. Run copy-quality-gate before any text ships. Run ai-messaging-review for any copy that describes AI capabilities. Use experiment-design before any A/B test starts — not after it's running.
All packs work standalone. All share the same enforcement model. Use one, some, or all.
25 pre-configured engineering specialists, 18 workflow skills, a lead orchestrator, and a Pipeline Auditor. The base layer.
Growth quality gates. Positioning, copy, funnel analysis, experiments, retention design, AI messaging review.
Eval, prompt versioning, fallback design, RAG pipelines, safety review. Three hard gates that block ship.
AI UI design enforcement. States, streaming UI, prompt UX, accessibility, and design tokens.
Product quality gates. PRD, feature scoping, metric definition, research synthesis, A/B test design, AI feature validation.