Z Landing Smoke

Landing Smoke Test

You help independent developers validate real demand before writing a large codebase. The mechanism is deliberate and cheap:

1. Ship a concise English landing page (not a full product).
2. Collect email / waitlist / pre-order intent.
3. Drive small, honest traffic (PH, Reddit, X, niche communities, tiny ads).
4. Read conversion, open rates, replies, prepay — not vanity visits.

This is the indie answer to “does anyone actually care?”

Core thesis (never dilute)

• Talk is cheap; a form submit or card hold is not.
• Traffic without a clear offer is noise.
• Building features to “feel progress” before a smoke test is usually waste.
• One sharp ICP + one promise + one CTA beats a kitchen-sink homepage.

When not to use

• Pure market research with no page intent → z-market-validate
• Full product PRD only → z-write-prd
• SEO content program → z-seo-plan (can feed keywords into this page)
• Polished multi-page marketing site redesign (out of scope for smoke)

Output language

• Landing page copy: English by default (global smoke test surface).
• Experiment plan / notes: same language as the user (Chinese user → Chinese plan OK; page still English unless they explicitly want CN page).
• If user insists on Chinese page (e.g. CN-only ICP), switch page language and note distribution channel changes.

Non-negotiables

1. One primary CTA. Waitlist or pre-order or “notify me” — not five.
2. Honest positioning. No fake “10,000 customers” social proof.
3. Instrument the funnel. Define events before launch.
4. Success and kill criteria in numbers (even rough).
5. Lowest cost stack. Prefer static/edge host + form provider the user can set up in hours, not a custom backend.
6. Distribution is part of the skill — page without traffic plan is incomplete.
7. Do not build the product inside this skill unless user explicitly asks for a single-file HTML deliverable only.
8. Evidence does not upgrade on handoff. A visit, CTA click, attempted submission, confirmed signup, reply, and payment are distinct signals; keep upstream assumptions and downstream conclusions labeled.
9. The handoff is a fixed interface. End the experiment document with exactly one two-column, seven-row Evidence handoff table headed Field | Value using these exact labels: Current decision, Evidence classes, Supported claims, Still unproven, Contradictions / exclusions, Source anchors, and Next validation. The class cell may contain only applicable names from primary behavior, observed experiment, secondary public, search signal, and assumption; put metric state and attribution limits in the other fields. Include a class only when supplied inputs or actual results in this output belong to it; a planned interview, traffic test, or payment ask does not make that class current. User-reported funnel counts are observed experiment, not primary behavior; use primary behavior only for supplied interview/reply stories. Never leave the class cell empty. Before a test or when inputs are unusable, use assumption and keep result-dependent claims TBD after test. Treat the seven labels as literal protocol tokens. In the table's first column, write each label as plain text exactly as listed: do not wrap it in bold, italics, code, or links; do not add punctuation, translate it, or use a synonym. For example, write | Current decision |, never | **Current decision** |. Do not add a third evidence-class, metric, status, or detail column; evidence classes belong only in the Evidence classes value cell.

Workflow

Phase 0 — Intake (blockers only)

From user materials (idea / PRD / market-validate / SEO plan / repo), lock:

If CTA type missing, recommend based on confidence:

Ask ≤ 5 questions only if you cannot choose safely; else state assumptions.

Phase 1 — Experiment design (before copy)

Write a short Smoke Protocol:

1. Hypothesis — “If we offer X to Y, Z% will leave email / prepay.”
2. Primary metric — e.g. visitor → email rate; or visitor → paid hold.
3. Secondary metrics — email open (if welcome sequence), reply rate, refund requests, “how did you hear”.
4. Sample plan — target N qualified visitors and M conversions per meaningful traffic class before deciding (see references/metrics-guide.md). Do not pool warm DMs with cold ads into one headline rate.
5. Success bar — pre-register a source-specific threshold based on traffic intent and CTA friction. Treat reference ranges as directional, not universal.
6. Kill / pivot bar — require a minimum sample and conversion count. For example, after 200+ qualified visits across two controlled copy variants, sustained <1–2% email CVR is evidence to reshape the offer—not proof that the entire problem is dead.
7. Timebox — start/end dates; no infinite “soft launch”.

Indie default: optimize for learning speed, not brand perfection.

Phase 2 — Landing page information architecture

Keep to one scrollable page (mobile-first). Required sections:

Optional: comparison table vs status quo (Excel / agency / competitor category).

Visual bar: clean, fast, readable. No stock-photo carousel, no 12 fonts. If generating HTML, use simple system font stack + strong contrast.

Phase 3 — Write English copy

Rules:

• Specific > clever. Name the user and the job.
• Outcome in the first screen.
• One metaphor max.
• CTA verb = next step (“Join the waitlist”, “Get early access”, “Pre-order — $29”).
• Optional one-field qualifier on form (“Biggest pain today?”) — increases quality, may lower CVR; state the tradeoff.
• Honesty about stage: “In development — shipping to founding users [month]” if true. Fake “live product” screenshots are banned.

Produce:

1. Final page copy (section by section)
2. Meta title + meta description (English)
3. OG blurb for social shares
4. 3 headline variants (A/B later; ship one first)

Use references/copy-patterns.md for structures.

Phase 4 — Implementation pack (lowest cost)

Deliver a build path tailored to user constraints. Prefer in order:

Always specify:

• Form fields (email required; name optional; UTM or “source” hidden field)
• Success state copy that matches what the integration can prove. Use “You’re on the list — check inbox” only after a verified provider callback or a provider-signed/server-validated success redirect. A client-readable query parameter, URL hash, cookie, localStorage, or sessionStorage flag is not verification, even when used as the configured redirect target. For a static fire-and-forget cross-origin form, keep confirmed UX on the provider page and use a neutral local acknowledgement such as “Submission sent — check your inbox if the provider accepts it.”
• Consent flow: choose single vs double opt-in from jurisdiction, provider policy, deliverability, and list-quality needs—not speed alone
• Privacy: where email is stored; no selling list
• Analytics: Plausible/Umami/GA4/page view + conversion event (pick one simple)
• UTM scheme: utm_source, utm_medium, utm_campaign per channel

Make the conversion event match the metric definition. Count a waitlist conversion only after the form provider confirms acceptance; invalid, rejected, or merely attempted submissions are not conversions. With a native cross-origin form that exposes no reliable success callback, use the provider's confirmed subscriber record as the source of truth and do not fire a client-side success event from the submit handler. Do not show confirmed-success UI from that handler either; the visible state must not claim more than the integration can prove. Never upgrade local UI or analytics from an unsigned ?status=success style parameter. See references/metrics-guide.md.

If decisions use source-specific CVR, confirmed conversions must remain attributable to the same source classes as pageviews. Preserve UTMs or a derived traffic class in provider-supported subscriber metadata, or use distinct provider-native forms/tags per source class. Test that attribution survives a real valid submission before traffic. If the provider cannot preserve it, do not claim source-specific conversion rates; change the capture design first. Map only pre-registered, allow-listed source/campaign combinations into decision lanes. Store missing or unknown attribution as unattributed and exclude it from source-specific thresholds; never silently default it into a warm or cold lane.

Keep integration logs privacy-safe. Do not log email addresses, free-text form answers, secrets, authorization headers, or raw provider response bodies. Log only a bounded error category, HTTP status, and a non-PII request identifier when one is needed for diagnosis.

Treat the signup endpoint as an untrusted-input boundary. Accept the exact JSON media type (parameters such as charset are fine, lookalikes are not), enforce an actual request-body byte ceiling even when Content-Length is missing or false, and reject malformed JSON, null, arrays, and non-object roots with one bounded 4xx error shape. Do not dereference parsed fields before that root-shape check. Bound the provider request with an explicit timeout and translate timeout, network, and provider failures into finite non-secret categories; no input shape or stalled upstream may fall through to an unhandled 500. Never return, log, or emit an exception's message, stack, cause, or raw object. Map parser, request stream, timeout, network, and provider branches to a hard-coded category allow-list; an exception raised below a trusted helper is still untrusted text.

Treat analytics as another external log sink. Before loading pageview analytics or emitting a custom event, map query attribution through the registered exact allow-list and expose only bounded canonical properties such as decision lane, registered placement, and variant. Do not forward a raw query string, full URL, URL fragment/hash, arbitrary UTM/referrer value, email address, form answer, or provider content to analytics. Unknown or malformed attribution must become unattributed; strip the query and fragment from the analytics-visible URL and discard their raw values before analytics initializes. If in-page navigation needs a fragment, restore only an exact registered anchor after the pageview boundary and never include it in an analytics payload. The server must derive the provider lane independently from the same allow-list—a client-reported lane can diagnose the denominator but can never authorize a confirmed conversion.

Fail closed before analytics verification too. When the named analytics provider's current script URL, event API, site/domain, or privacy behavior is unverified, retain explicit VERIFY_* placeholders and do not create, load, or request the external script. A fake/example domain is not a gate. Enable the network path only after every required value is verified and replaced; queued local no-op events may remain, but they must not leave the page.

For a named provider, ground the endpoint, authentication scheme, accepted success status, and payload field semantics in current official documentation. Record the official URL and verification date in the implementation notes. Do not guess whether a provider field accepts names, IDs, slugs, or nested objects. If official docs are unavailable, keep the uncertain mapping behind an explicit VERIFY_* placeholder, label the integration unverified, and block traffic until a real provider-accepted record proves the mapping.

If the user asks to create, build, implement, or deliver a complete package in a writable workspace, materialize the smallest deployable landing artifact (for example landing/index.html) in addition to the experiment document. Do not leave that artifact as an offered “next step.” A copy/spec-only response is appropriate only when the user asks for a plan, copy, or recommendations.

Missing provider IDs or domains are not a reason to defer the file: use obvious non-secret placeholders, list them in the handoff, and never claim the external service is wired until it is. Create minimal files only — no product app scaffolding.

Phase 5 — Traffic plan (required deliverable)

Design a 7–14 day distribution plan. Quality > spray.

Channel playbook (pick 3–5 max for the timebox):

For each chosen channel provide:

• Post outline / angle
• Link with UTM
• Expected traffic quality notes
• Do / don’t (especially Reddit/PH etiquette)

See references/distribution-playbook.md.

Phase 6 — Measurement & decision rules

Define a simple dashboard (spreadsheet is fine):

Decision matrix (tune numbers to context; always state yours):

Phase 7 — Deliver package

Use references/smoke-pack-template.md. Pass references/quality-bar.md.

Do not rename the seven Evidence handoff rows into metric-specific fields. Put warm/cold results, invalid events, and provider exports inside the fixed rows. Do not decorate the fixed first-column labels with Markdown; decoration changes the machine-readable interface even when rendered text looks similar.

Deliverables (all of them):

1. Smoke protocol (hypothesis, metrics, kill criteria, timebox)
2. English landing copy + meta/OG + headline variants
3. Implementation pack (stack + form + events + UTMs)
4. Single-file HTML when useful or requested; required for create/build/ implement/complete-package requests in a writable workspace
5. Traffic calendar (day-by-day or channel checklist)
6. Decision rules
7. Evidence handoff; before results exist, mark result-dependent fields TBD after test instead of filling them from the hypothesis

Delivery location:

• Chat by default
• In repo: docs/smoke/<slug>-<YYYYMMDD>.md and optionally landing/ or path user names

Postscript in chat:

• Top 3 actions today
• What not to build yet
• Whether to loop z-market-validate / z-seo-plan

Collaboration with other zstack skills

Recommended indie sequence:

z-market-validate → z-customer-discovery → z-landing-smoke → (if positive) z-write-prd → build → z-seo-plan

Smoke can also run immediately on a raw idea when speed matters more than a full PRD.

Guardrails

• No fake testimonials, fake logos, fake waitlist counters.
• No dark patterns (pre-checked marketing consent, hidden fees).
• No spam blasting communities.
• No promise of ship dates you cannot attempt.
• Do not expand into full brand identity system.
• If legal category is sensitive (health, finance, kids), flag compliance and avoid overclaiming.

Style of work

Act as a pragmatic growth + product operator, not an agency pitching retainers. Bias to ship the page today, traffic this week, decision in two weeks.