Ship Press — launch kit from a release

Turn a shipped release into everything you need to launch it publicly: a press release, the copy for a Product Hunt (or HN / Reddit / IH) post, and a launch-day checklist. The input is the work you already shipped — this skill reframes it for the public, on-brand, with zero fabrication.

The shipping log says what changed. The user guide says how to use it. The launch kit says "here's why the world should care, today." It is attributable and quotable, so the honesty bar is absolute: no invented quotes, customers, traction, or metrics — ever.

Part of the "ship" pack

One repo, one docs/, one brand file. This skill consumes the pack's outputs:

docs/VERSION              ← shared semver anchor (ship-release owns the bump)
docs/brand.json           ← brand-as-data (colors + fonts) — themes every HTML output
docs/SHIPPING-LOG.md      ← the release this kit announces   (← shipping-log / ship-changelog)
   └─ ship-press          → docs/press/release.md + .html, docs/press/launch-kit.md   (this skill)
docs/brand/logos/*        ← thumbnail / gallery art          (← logo-pack)
docs/screenshots/*        ← gallery images                   (← screenshot-capture)
docs/user-guide/*         ← "how to use it" companion link   (← user-guide-builder)

Cross-links: get the release from shipping-log / cut it with ship-release; get the thumbnail from logo-pack; get gallery images from screenshot-capture.

Outputs

launch-kit.md is intentionally markdown-only — it's paste-ready copy (PH/HN/Reddit fields + an operational checklist) meant to be copied straight into a submission form or a task list, so it gets no HTML render. Only release.md is rendered (it's the one public, link-shareable artifact). That asymmetry is by design, not a missing output.

Workflow

1. Discover what exists (always first — never act blind)

Run the bundled discovery script from the repo root. It reads the shipping log, brand.json, and docs/VERSION, extracts the latest release (version, date, themed line, bullets by section) and the brand voice (brand.voice / tone / personality, plus do/don't/words lists), lists existing logos/screenshots to attach, and flags any prior launch kit so you don't clobber it.

node "<skill-dir>/scripts/collect-launch.mjs" --json

<skill-dir> is this skill's folder (e.g. ${CLAUDE_PLUGIN_ROOT}/skills/ship-press). Read the JSON; do not guess the release from memory. Honor every warnings[] entry:

• No shipping log → that's a blocker. A launch kit needs a real release. Offer to run shipping-log first, or ask the user to point --log at a changelog.
• No brand.json → HTML falls back to a neutral slate theme (never purple). Proceed, note it.
• voice present (report.voice) → write the press release and the maker's comment in that voice — honor its tone/personality and its do/don't/words lists. Don't improvise a tone when the brand has declared one. voice absent → use the skill's default plain-English press voice.
• No themed line → write the lede from the bullets, grounded only in them.
• Prior kit exists → confirm before overwriting; offer to update/diff instead.

2. Confirm the launch framing (one or two questions, only if needed)

Resolve only what the source can't tell you: the maker/spokesperson name (for attribution), the launch channel (Product Hunt? Show HN? just a press release?), the canonical URL and media contact, and the price/availability if the kit should state it. If unknown, you'll use honest [ADD …] placeholders — do not invent any of these.

3. Write the press release → docs/press/release.md

Follow references/press-release-template.md exactly: the headline → dated lede → problem → what-shipped paragraphs → "What's new" list → optional quote → availability → about → contact. Map the "What's new" list 1:1 to the latest release's bullets.

The hard rules (full list in the template):

• No fabricated quotes. A maker statement may be first-party and attributable ("— The {Product} team" or a real name). No real source → omit the Quote section or leave [ADD CUSTOMER QUOTE].
• No fabricated metrics. No "10x", "saves N hours", "trusted by thousands", "#1" unless the exact number is in the source. Otherwise leave [ADD METRIC] and report it.
• Honest ship status. "Available now" only if it truly is; else "rolling out" / "in {plan} tier".
• Plain English, present tense, third person, one screen.

4. Write the Product Hunt assets + launch-day checklist → docs/press/launch-kit.md

Follow references/product-hunt-assets.md:

• Tagline — 3 ranked options, each ≤60 chars, benefit-first, from brand.json.tagline.
• Description — 2–3 sentences (~260 chars): problem → what it does → who it's for.
• Maker's first comment — ~120–180 words, first person, the arc: who → the itch → what it does today (honestly scoped) → what's next → a real ask. Never a fake persona. Match the brand voice from discovery if one was declared.
• Gallery notes — list the assets discovery found; flag what's missing and which pack skill makes it.
• Launch-day checklist — Before / Launch hour / During / After, terse and checkable, adapted to the channels the user actually has (don't assume a Twitter/X account exists).

Then verify the counts programmatically — the 60-char tagline limit is hard, and an eyeball count is exactly where it breaks. Don't trust the mental tally:

node "<skill-dir>/scripts/check-counts.mjs" --in docs/press/launch-kit.md

It reads every Tagline option and the Description back out of the file and prints OK/OVER with the real code-point count; it exits non-zero if any tagline exceeds 60 (or the description blows the max). Tighten any OVER line and re-check before moving on. (Ad-hoc while drafting: check-counts.mjs --tagline "…" --description "…".)

5. Render the press release to HTML

node "<skill-dir>/scripts/render-press.mjs" --in docs/press/release.md --out docs/press/release.html \
  --brand docs/brand.json --version docs/VERSION

Self-contained single file, themed from brand.json via :root CSS variables, neutral slate default (never purple), print-friendly, no backdrop-filter, no SVG feTurbulence. If brand.json is absent it themes from the neutral defaults — still on-brand-safe.

Render verification (don't skip the exit code). The renderer scans the generated HTML for any remaining [ADD …] / [CONFIRM …] placeholders — which the release carries honestly by design — and prints a HTML still contains N placeholder(s) line listing each one. If any remain it exits non-zero (3) and refuses to certify the page as final, embeds a machine-readable comment, and draws a visible "Draft — N placeholders" banner on the page so a kit is never published with raw [ADD CONTACT] showing. The fix is to fill the placeholders and re-render. If you deliberately want a labelled draft for review, pass --allow-placeholders (exit 0, banner stays). Capture that N for the report.

6. Report

Summarize: the files written, the release described (version + date), the full list of [ADD …] / [CONFIRM …] placeholders the user must fill — quote the renderer's HTML still contains N placeholder(s) line so the count is explicit — the check-counts result (any tagline that was OVER), any missing gallery assets (and which skill produces them), and the ship-status note. Then offer the next pack step (refresh screenshots, or ship-release to cut the next version).

Worked example

See references/worked-example.md for a complete input→output pass: a real brand.json + VERSION + shipping-log block in, and the resulting release.md, Product Hunt assets, maker comment, checklist, and themed HTML out — with notes on why each passes the honesty bar (capabilities traced to bullets, quote omitted, placeholders flagged).

Quality bar

A launch kit is good when:

• Every claim traces to a real input. Each "What's new" line maps to a shipping-log bullet; each number appears in the source. If you can't trace it, it's not in the kit — it's an [ADD …] flag.
• No fabricated social proof. Zero invented quotes, customers, logos-of-companies, user counts, or "industry-leading" language. A maker statement is first-party and attributable, or absent.
• Benefit-first, plain English. Outcomes for the reader, not implementation. No marketing throat-clearing ("We're thrilled to announce…"). Cut every non-load-bearing adjective.
• Specific to this product. If the press release would fit any company by swapping the name, it's too generic — anchor it in this release's actual capabilities and this product's audience.
• Counts respected — and verified. Tagline ≤60 chars, description ~260, press release one screen, maker comment 120–180 words. The tagline/description counts are checked programmatically (check-counts.mjs), not by eye — check-counts must come back with no OVER.
• On-brand voice. If brand.json declares a voice/tone, the release and maker's comment read in it (honoring its do/don't/words) — not a generic press tone.
• No raw placeholders shipped. The rendered HTML is scanned; render-press reports the remaining [ADD …] count and refuses to certify a page that still has any (a labelled draft is explicit).
• Honest about gaps. Missing assets, unknown facts, and staged-not-live work are surfaced as placeholders and report notes — never papered over.
• On-brand, portable HTML. Themed from brand.json, never purple by default, self-contained, no backdrop-filter / feTurbulence. Works in any repo — nothing hardcoded to one app.

Reusable contents

• scripts/collect-launch.mjs — discovery + raw material (run, don't reinvent): latest release, brand, brand voice/tone, version, existing assets, prior outputs, warnings. --json for machine-readable.
• scripts/render-press.mjs — markdown press release → self-contained, themed HTML page; scans for unfilled [ADD …]/[CONFIRM …] placeholders and refuses to certify a page that still has any.
• scripts/check-counts.mjs — programmatic char-count check for the PH tagline (hard 60) + description; reads a finished launch-kit.md (--in) or ad-hoc strings (--tagline/--description).
• references/press-release-template.md — exact press-release structure + the honesty rules + voice.
• references/product-hunt-assets.md — PH tagline/description/maker-comment specs + launch-day checklist.
• references/worked-example.md — one full input→output pass with rationale.