Skill clusters
When to use which skills
One app, from the first paragraph to the week it has users. Pick a phase. The bar says when that work sits. The chips are the toolkit, not the whole catalog.
Start here with a paragraph, not a repo. mega-eval is step one of the idea path.
Name who you think has the problem, and paste notes from real conversations if you have them.
product-pitch is step two of the idea path. Open general-counsel only when there is a claim, a clause, or a letter.
System design and architecture run from a description. blurb-to-app returns a spec and code inline. Deploy stays with you.
idea-to-launch is step three of the idea path. Research can run here. Build and deploy come back as plans.
- idea-to-launchLocalResearch stages run for real. Build and deploy stages come back as plans and code.
- testing-strategyRuns end to endA test plan for the feature you are about to build, from the code or the description.
- documentationRuns end to endDocs for the code, API, or process, aimed at a named reader.
Paste the diff, the tree, or the release the catalog asks for. Add payment-audit only if money moves.
- code-reviewRuns end to endA review of the diff or PR, with what the change is for.
- repo-doctorAdaptedA scan of the tree and the files that matter, aimed at what to clean up.
- expert-dev-reviewAdaptedA harder pass on the same kind of tree, when the risk is higher than a checklist.
- deploy-checklistRuns end to endThe release itself: changes, migrations, flags, environments.
- payment-auditAdaptedCheckout, webhooks, and pricing copy. Skip it when the product does not charge.
Launch wants a description, a live URL, an audience, and a date. Social metadata wants the layout and theme.
These need something live. supporter-loop does not send the mail. pulled-feedback needs a URL.
- supporter-loopAdaptedThe first-ten-users loop: state in, updated files out. You still send the messages.
- pulled-feedbackRuns end to endOne locked question for a tester, against a live URL.
- growth-analyticsRuns end to endWhat to grow, from the live product, current numbers, and where traffic comes from.
Incident notes, a weekly digest, debt, and standup run from what you paste. Audits of a live app come back as a plan: this runner cannot click production.
- incident-responseRuns end to endA live incident or a postmortem, from alerts, a timeline, and impact.
- app-auditLocalA review of a live URL and the routes you name. Nothing can be clicked from here.
- fix-bugsLocalA plan for a broken production URL. Local: the fix is not applied here.
- weekly-qa-digestRuns end to endSeveral weekly QA reports, read together.
- tech-debtRuns end to endWhat to pay down, from the tree and the files that matter.
- standupRuns end to endA short read of commits, PRs, or rough notes, once the app is in motion.
Notes
- Slugs, categories, fit, and the paste hint in each detail come from src/config/skills.ts. Engineering entries set fit to full. Any other entry uses the fit on that slug, or the catalog default adapted when fit is omitted.
- Runs end to end, Adapted, and Local mean what README.md describes under “What runs end to end means.” A Local skill here returns a plan or dry run. Building, deploying, clicking a live app, and sending mail still happen outside Skill Runner.
- mega-eval, product-pitch, and idea-to-launch are the three-step path on /idea, written up in docs/from-idea-to-app.md. This schedule puts them in Frame, Say it, and Make. It does not replace that page.
- /map and src/config/journey.ts group the whole catalog by job verb: Research, Write, Build, Fix, Test / Ship. This page is a smaller toolkit grouped by where you are in the life of one app. A skill can belong in both pictures.
- Bars overlap on purpose. They are not dates, durations, or a critical path. Framing does not end when listening starts, and review comes back after the first users arrive.
- supporter-loop’s catalog hint is a first-ten-users loop. The skill returns updated state files. Sending stays with you.
- pulled-feedback asks for a live product URL, so it sits with first users, not with a blank idea.
- payment-audit belongs in Prove when the product takes money. Skip it when nothing is charged.