--- name: skillzip-expert description: Guided 7-phase interview that turns a expert's raw playbook into a publish-ready Skillzip skill — demo prompt + premium content + pricing + listing copy + JSON ready to paste into /dashboard/expert/skills/new. Use when user says "create skillzip skill", "publish to skillzip", "skillzip-expert", "ship my skill", or pastes a draft prompt and asks "how do I sell this on skillzip". argument-hint: "" level: 3 --- Skillzip-expert is a marketplace-onboarding interview. It takes a expert's fuzzy idea ("I want to sell my cold-email playbook") and walks them through the seven decisions that must be made before a Skillzip skill can ship: identity, demo design, premium content split, pricing, listing copy, demo examples, and validation. The output is a JSON object matching the Skillzip `skills` table schema plus a curl/copy-paste ready payload — the expert can publish in 30–60 minutes instead of staring at a blank dashboard form. Every phase is gated by `AskUserQuestion` calls that force explicit choice. The skill refuses vague answers, surfaces trade-offs, and never advances until each gate has a concrete decision. - User says "skillzip-expert", "create skillzip skill", "publish to skillzip", "ship my skill", "list my prompt on skillzip" - User pastes an existing prompt/playbook and asks how to monetize it on Skillzip - User has the raw expertise but doesn't know how to package it as a Skillzip listing - Expert coming from cold-email outreach (`/for-experts/*` landing) wants to onboard - User wants to *browse* Skillzip as a buyer — point them to https://skillzip.vercel.app/browse - User is asking for general prompt-writing advice with no intent to publish - User wants to build a non-Sales/Marketing skill — politely defer (Skillzip Stage 1 is vertical-locked); revisit when the platform expands - User already has a published skill and wants to *edit* it — use `/skillzip-edit` (separate skill) instead of running the full interview again Most prompt-marketplace listings on Gumroad / similar platforms underperform because the expert skipped 3-5 packaging decisions: their demo gives away too much (no upgrade incentive), their pricing is anchored on what feels "safe" (under-priced by 60%), or their listing copy describes the *prompt* instead of the *outcome*. Each missed decision costs sales for the lifetime of the skill. This skill is the interview the expert would have done with a packaging consultant — captured in code so they can self-serve in an hour. - Phase 0 → 6 each end with at least one `AskUserQuestion` gate. Do not advance until answered. - Show 2-4 concrete options per question. Vague open-text questions are forbidden — always offer Recommended option first with rationale. - After every phase, summarize the locked-in decisions in 2-3 lines so the expert can sanity-check before continuing. - State persists between phases via `state_write(mode="deep-interview", state.source="skillzip-expert")`. Expert can resume with `/skillzip-expert resume`. - Phase 7 (push) requires explicit user confirmation. The skill never auto-publishes — only generates the payload + opens dashboard. - If expert's niche is outside Sales/Marketing, hard-stop with a polite redirect instead of completing the flow. ## Phase 0: Identity snapshot Goal: lock the *who*, *what for*, *what category*. 4 questions, 1 round. ``` AskUserQuestion ([ { question: "Who is the primary buyer for this skill? (drives positioning + pricing anchor)", header: "Buyer ICP", options: [ "SMB founder / solopreneur (Recommended for prompt-heavy skills)", "B2B Sales rep / SDR", "Growth / Performance marketer at startup", "Agency operator selling client services" ] }, { question: "Which Sales/Marketing sub-vertical is the closest fit?", header: "Sub-vertical", options: [ "Cold Outreach (cold email, LinkedIn DM, sequencing)", "Ad Copy (Meta / Google / TikTok creative)", "Growth (audience build, viral hooks, lead magnets)", "Performance Marketing (account audit, optimization, attribution)" ] }, { question: "What's the #1 pain your skill solves? (drives the headline + tagline)", header: "Core pain", options: [ "Buyer can't write at all — needs ground-zero output", "Buyer can write but lacks framework / structure", "Buyer has output but it doesn't convert / land", "Buyer needs scale — wants 10x volume of variants" ] }, { question: "Have you already shipped this prompt to a paying buyer or used it for a paid client deliverable?", header: "Validation", options: [ "Yes — paid use, with measurable result", "Yes — used for client work, no isolated metric", "Beta-tested with friends / network", "Brand new, never shipped" ] } ]) ``` After the round, write state and confirm: > Locked: ICP={buyer}, vertical={vertical}, pain={pain}, validation={state}. > Brand-new prompts can still ship, but we'll position you as "experimental" > and recommend lower price anchor. ## Phase 1: Demo design Goal: design the public demo system prompt so it *hooks* without giving away the master playbook. 2 rounds. ### Round 1.1 — anatomy ``` AskUserQuestion ([ { question: "What's the smallest unit of useful output a buyer should see in the demo?", header: "Demo output", options: [ "1 polished example (Recommended — highest hook-rate)", "3 short variants (good for ad copy / hook generation)", "A skeleton / outline (good for frameworks / multi-step)", "A diagnosis / first-pass critique (good for audit-style skills)" ] }, { question: "What MUST stay hidden? (drives the demo's natural ceiling)", header: "Locked behind paywall", options: [ "The full multi-step framework / sequence", "The voice / tone calibration rules", "Edge cases + 'if this then that' branches", "Curated examples library + templates" ], multiSelect: true }, { question: "How long should each demo response be?", header: "Demo length", options: [ "Under 80 words (Recommended — Anthropic cost guard, demo cap)", "80–150 words", "150–250 words (only if buyer needs to *see* a flow to understand value)" ] } ]) ``` ### Round 1.2 — draft & pick Generate 2 demo system prompt drafts. Each must: - Be ≤300 tokens (Skillzip rate-limit cap) - End with a one-line teaser for premium ("Premium unlocks…") - Stay strictly within the chosen "must stay hidden" boundary ``` AskUserQuestion ([ { question: "Pick a demo system prompt direction. (You can iterate after — this is just the v1.)", header: "Demo prompt", options: [ "Draft A — direct, output-first (Recommended for ad copy / hooks / one-shot skills)", "Draft B — diagnostic, asks 1 clarifying Q first (better for consulting-style skills)" ], preview: "" } ]) ``` After: write the chosen prompt to state. Show character count + token estimate. ## Phase 2: Premium content extraction Goal: pull the master playbook out of the expert's head into a structured artifact that buyers will pay for. 2 rounds. ### Round 2.1 — what exists ``` AskUserQuestion ([ { question: "How developed is your master playbook today?", header: "Playbook state", options: [ "Polished — written, tested, ready to paste", "Sketched — bullet list, needs assembly", "Mental model only — never written down", "I want help generating it from scratch (we'll co-author it)" ] }, { question: "What format will buyers find most valuable?", header: "Premium format", options: [ "Master system prompt + expert commentary (Recommended for prompt buyers)", "Step-by-step SOP + decision tree", "Template library (e.g., 12 ready-to-fill examples)", "All of the above (priced higher)" ] } ]) ``` ### Round 2.2 — extract or co-author If "Polished": ask expert to paste it; validate length ≤8K chars. If "Sketched"/"Mental"/"Help": run a co-author sub-loop with up to 5 turns: - Ask the expert 3-5 targeted Socratic questions to extract the framework - Generate a draft master prompt - Ask expert for 1 round of edits before locking ``` AskUserQuestion ([ { question: "Lock the premium content?", header: "Premium ready", options: [ "Lock it — looks good (Recommended)", "One more revision pass", "Replace entirely with new draft" ] } ]) ``` ## Phase 3: Pricing Goal: set price anchors that match buyer ICP + content depth. 1 round. The skill computes a recommended price band based on Phase 0 (ICP) + Phase 2 (content depth): | ICP | Depth = prompt only | Depth = prompt + SOP + templates | |-----|---------------------|-----------------------------------| | SMB founder | $39–$79 | $99–$199 | | B2B Sales rep | $49–$99 | $129–$249 | | Growth/Performance | $59–$129 | $149–$299 | | Agency operator | $79–$149 | $199–$399 | ``` AskUserQuestion ([ { question: "Pricing model — how do you want buyers to pay?", header: "Pricing model", options: [ "One-time only (simple, lower friction)", "Subscription only (recurring, lower churn for evolving playbooks)", "Both — buyer picks (Recommended — Skillzip data shows ~30% pick subscription)" ] }, { question: "Within the recommended band ({computed_band}), where do you want to anchor?", header: "Price anchor", options: [ "Floor — fastest to first sale, leaves room to raise", "Mid (Recommended)", "Ceiling — premium positioning, slower start" ] }, { question: "What's a 1:1 consult or done-for-you version of this skill worth?", header: "Anchor reference", options: [ "$200–$500 / engagement", "$500–$1,500 / engagement", "$1,500–$5,000 / engagement", "$5,000+ / engagement" ] } ]) ``` After: compute final price. Show the breakdown (gross → 12% Skillzip → 88% to expert) so expert sees their per-sale take in cents. ## Phase 4: Listing copy Goal: convert "what the skill does" → "what the buyer wins". 1 round (with auto-generated drafts, multi-select). The skill auto-generates: - 3 candidate titles (outcome-led, framework-led, ICP-led) - 3 candidate taglines (≤120 chars each) - 1 description draft (Problem → Solution → Outcome → Proof) ``` AskUserQuestion ([ { question: "Pick a title direction.", header: "Title", options: ["Outcome-led: ''", "Framework-led: ''", "ICP-led: ''"] }, { question: "Pick a tagline. (Skillzip homepage cards show ~80 chars on mobile.)", header: "Tagline", options: ["", "", ""] }, { question: "Description tone — match how your buyers actually talk?", header: "Tone", options: [ "Operator (direct, no fluff, numbers-first) — Recommended for B2B/Sales", "Coach (warm, second-person, journey-driven) — better for Growth/audience", "Analyst (precise, framework-first, evidence-led) — best for Performance / Audit" ] } ]) ``` ## Phase 5: Demo examples Goal: 3 starter prompts buyers can click in the playground. 1 round. ``` AskUserQuestion ([ { question: "What 3 starter prompts give the strongest first-impression?", header: "Starters", options: [ "Auto-generate from my premium content (Recommended)", "I'll paste the 3 I already use", "Mix — auto-gen 2, I'll write 1 hero starter" ] } ]) ``` If auto-gen: derive 3 starters from the master prompt's most common buyer input shapes. If user paste: validate ≤200 chars each. ## Phase 6: Preview + validation gate Goal: assemble the full skill payload, render a visual preview, gate publish. Render a side-by-side preview: - Left: the listing card as it'll appear on `/browse` - Right: the full skill detail page mock (header + demo + premium-locked + pricing panel) ``` AskUserQuestion ([ { question: "Skill payload is ready. What next?", header: "Ship it", options: [ "Publish as draft to Skillzip (Recommended — review on dashboard before going live)", "Copy JSON payload to clipboard (manual paste into /dashboard/expert/skills/new)", "Edit a section first (back to Phase 1/2/3/4/5)", "Save state and come back later" ] } ]) ``` ## Phase 7: Push ### Option A — publish as draft If user is signed in to Skillzip in this session AND `SKILLZIP_API_TOKEN` env is set: - POST the payload to Skillzip's `/api/skills` endpoint with `status: "draft"` - Return the draft URL: `https://skillzip.vercel.app/dashboard/expert/skills/{id}` If not signed in: - Generate a signed magic link to `/dashboard/expert/skills/new` with the payload pre-filled in URL state (base64-encoded) - Open the link in the user's browser ### Option B — copy JSON Output the payload as a fenced JSON block + a one-line instruction: > Open https://skillzip.vercel.app/dashboard/expert/skills/new, paste each > field. Or pipe to `xclip` / `pbcopy` and refer back here. ### Final report Always end with: ``` ✅ Skill scaffold complete Title: {title} Tagline: {tagline} Category: {sub-vertical} Pricing: {one_time_or_sub} / ${price} Demo cap: {demo_chars} chars / ~{demo_tokens} tokens Premium: {premium_chars} chars Starters: 3 prompts ready Per-sale math: ${gross} gross → 12% Skillzip → ${gross_x_88} to you At {target_buyers}/mo → ${monthly_take} MRR Next steps: 1. Review draft at {dashboard_url} 2. Run 3-5 demo runs yourself in playground to QA 3. Flip status to "published" when ready 4. Share your skill page URL in cold outreach reply ``` - `AskUserQuestion` — every phase gate; never advance without explicit user choice - `state_write(mode="deep-interview", state.source="skillzip-expert")` — persist between phases for resume - `state_read` at start to detect resume scenario - `WebFetch` to https://skillzip.vercel.app/api/skills/categories if expert picks "Other" sub-vertical (returns current allowed list) - `Read` if the expert paste-references a local file containing their existing playbook - `Write` to scratch a working draft in `.skillzip-expert/draft-{slug}.json` so expert can resume - Never invoke `autopilot`, `ralph`, or other execution skills — this is a content-generation skill, not a code skill User: "skillzip-expert I want to sell my cold email framework that books me $5k MRR clients" Phase 0: ICP=B2B Sales rep, Vertical=Cold Outreach, Pain=#3 (output exists but doesn't convert), Validation=#1 (paid + measurable) Phase 1: Demo = 1 polished email opener (≤80 words), hides full 5-step sequence + voice rules Phase 2: Premium = master prompt + 12-template library Phase 3: Both pricing, mid-band → $99 one-time / $29 mo Phase 4: ICP-led title "Cold Email That Books $5K MRR Demos" Phase 5: 3 auto-gen starters using common SaaS / agency / B2B prospect shapes Phase 6: Publish as draft Phase 7: Returns dashboard URL + per-sale math End-to-end: ~25 minutes with expert answering ~12 multi-choice questions. User: "skillzip-expert help me sell something on skillzip" Phase 0 question 1 fires. User answers "I dunno" → skill SHOULD pause and ask for the one-sentence skill description before continuing. Don't auto-fill placeholders. Skill detects user's niche is "personal productivity" (outside Sales/Marketing). Skill SHOULD politely stop with: "Skillzip is currently vertical-locked to Sales/Marketing. Want me to flag your niche to the founders for the next expansion phase?" — not silently complete the flow with a non-fitting category. - Stop and ask for clarification if user provides no concrete inputs to Phase 0 (skill description blank or 3 words long) - Hard-stop with niche-redirect message if vertical is outside Sales/Marketing - Pause + state_write if expert says "stop", "later", "save" - If `/api/skills` push returns 4xx, fall back to Option B (JSON copy) and surface the error - If 3 phases pass with expert selecting "I'll paste it" but never pasting, prompt explicitly: "Paste the content here, or pick auto-gen" - [ ] Phase 0 locked ICP + vertical + pain + validation - [ ] Phase 1 demo prompt ≤300 tokens with paywall teaser - [ ] Phase 2 premium content ≥500 chars (substantive) - [ ] Phase 3 pricing within recommended band - [ ] Phase 4 title + tagline + description all chosen, not auto-defaulted - [ ] Phase 5 has 3 starter prompts - [ ] Phase 6 user explicitly approved publish/copy/edit/save - [ ] Phase 7 produced either a draft URL or a JSON payload — never silent failure - [ ] State written so expert can resume - [ ] Final report shows per-sale math (88% take after 12%) ## Configuration Optional settings in `.claude/settings.json`: ```json { "skillzip": { "expert": { "apiBase": "https://skillzip.vercel.app", "apiToken": "${SKILLZIP_API_TOKEN}", "defaultPricingModel": "both", "demoTokenCap": 300, "premiumMinChars": 500, "autoOpenDashboard": true } } } ``` ## Resume If interrupted, run `/skillzip-expert resume`. The skill reads `state_read(mode="deep-interview")`, checks `state.source === "skillzip-expert"`, and resumes from `current_phase`. The draft JSON at `.skillzip-expert/draft-{slug}.json` is the canonical resume artifact. ## Co-authoring sub-loop (Phase 2.2) When expert picks "I want help generating it from scratch", run this inline mini-interview: 1. "What does the buyer give you as input? (1 sentence)" 2. "What 3-7 transformations do you apply, in order?" 3. "What's the expected output shape?" 4. "What 2 things would make a buyer say 'this is wrong / not for me'?" 5. "What 2 examples (input → output) would make them say 'oh, this is exactly what I need'?" Feed answers into a draft master prompt template, generate, present to user for one revision pass. Cap at 5 questions to avoid death-by-interview. ## Pricing band derivation The recommended band is computed from Phase 0 ICP × Phase 2 depth, not hand-picked. If Skillzip later publishes per-vertical conversion data, this table updates accordingly. Treat the band as evidence-based default — expert can override but must select "ceiling" with explicit acknowledgment. ## Why not auto-publish? Skillzip-expert never auto-publishes (`status: "published"`) because: 1. The expert hasn't QA'd the demo against their own input yet 2. The dashboard preview uncovers issues a JSON dump won't 3. Drafts are reversible; published listings build search history The skill always ends at `status: "draft"` + dashboard URL. ## Integration with `/for-experts/*` cold outreach When a expert arrives via `/for-experts/founding` (Founding 50 program), the skill should detect this in `state.source.referrer` and: - Auto-tag the application with `founding_50: true` - Skip Phase 3 ceiling caps (founding experts can price freely) - Add the gold founder badge to the listing payload When arriving via `/for-experts/earn`, Phase 3 should weight pricing toward mid–ceiling band (expert already self-selected for revenue motivation). When arriving via `/for-experts/grow`, Phase 5 should auto-suggest 1 starter prompt that mirrors a real published skill in the same sub-vertical (showcases the demo pattern that already works). ## What this skill does NOT do - Does NOT write the master prompt for experts who don't have one — it *helps extract* what the expert already knows. Net-new prompt authoring belongs in a separate skill. - Does NOT optimize SEO / Skillzip ranking — that's `/skillzip-rank-up`. - Does NOT handle published-skill edits — that's `/skillzip-edit`. - Does NOT process Stripe payouts setup — expert does that once via `/dashboard/settings`.