@damiangilgonzalez/harness 1.3.7

package/skills/linkedin-carousels/SKILL.md

@@ -0,0 +1,228 @@
+---
+name: linkedin-carousels
+description: "Use when building a LinkedIn carousel / document post — the swipeable multi-page PDF — slide by slide: cover hook, narrative arc, one idea per slide, CTA closer, PDF export spec. NOT the account/format plan (that is `linkedin-strategy`), NOT a text or single-image feed post (that is `linkedin-content`), NOT a 16:9 talk deck (that is `presentations`)."
+tags: [linkedin, carousel, document-post, slide-design, visual-hierarchy, hook, b2b, social-content]
+recommends: [linkedin-content, linkedin-strategy, brand-identity, brand-voice, presentations, social-publisher]
+origin: risco
+---
+
+# LinkedIn carousels
+
+Design **one swipeable LinkedIn document post**. Your artifact is the *slides of a phone-read
+PDF* — sequence and visual hierarchy are the product — handed back as a slide-by-slide build
+ready to assemble in Canva/Figma/Slides plus a one-page design+export checklist. Everything
+adjacent (the account plan, the feed copy, the upload, a talk deck) routes out; see Handoffs.
+
+## What you produce
+
+- **A cover slide spec** — the one slide that wins or loses the swipe to slide 2.
+- **Interior slides** — each carrying a single focused idea, large type, a clear hierarchy.
+- **A CTA closer** — the payoff slide plus one explicit ask.
+- **A design + export spec** — canvas, safe area, type floor, PDF checklist (one page).
+- **A one-line caption starter** — the first feed line; full caption tuning routes out.
+
+Precondition: a topic, an angle, or a source (post/article). If the ask is the account plan or
+a block of feed copy, you are in the wrong skill — see the handoffs table.
+
+## Ground first
+
+A carousel inherits an identity; it does not invent one. Read the brand kit (colors, type,
+logo — owned by [`../brand-identity/SKILL.md`](../brand-identity/SKILL.md)) and the voice
+([`../brand-voice/SKILL.md`](../brand-voice/SKILL.md)) from `docs/` if they exist; don't pick
+arbitrary fonts. Pull the angle from any plan or post already produced by
+[`../linkedin-strategy/SKILL.md`](../linkedin-strategy/SKILL.md) or
+[`../linkedin-content/SKILL.md`](../linkedin-content/SKILL.md) — reuse the hook, don't reopen
+it. If none exist, name your assumptions on slide 0 of the spec and proceed; don't stall.
+
+## The swipe economics (why every rule below exists)
+
+Document/carousel posts are the **highest-engagement native format on LinkedIn right now** —
+roughly 6.6% average engagement, ahead of every other format, and top-performing pages post far
+more of them than median pages. The reason is **dwell time**: a swipeable deck holds the thumb
+on screen, and dwell — not likes — is the algorithm's main distribution signal.
+
+So the whole bet is the **cover slide and the swipe-to-slide-2 bounce**. If the cover doesn't
+earn the first swipe, your slide count, your arc, and your CTA never get seen; every swipe after
+that extends dwell and signals quality, which is why each slide must *pull to the next*. Design
+backwards from that: stop the thumb, then keep it moving.
+
+## The cover slide
+
+This slide is the entire campaign. Build it to one job: earn the swipe.
+
+- **Lead with a curiosity or benefit headline** — promise a payoff, not a topic label. Why: a
+  label ("Onboarding tips") gives no reason to swipe; a stake ("Why 60% of users quit in week
+  one") does.
+- **One dominant element** — the headline reads first, before anything else. Why: a cover with
+  three competing focal points reads as zero.
+- **Intentional white space** — let the headline breathe. Why: a dense cover looks like work,
+  and the thumb keeps scrolling.
+- **An explicit swipe cue** — an arrow, "Swipe →", or a 1/8 counter. Why: tell people the
+  format is a deck so they know there's more.
+
+The test: **would this stop *my own* thumb mid-scroll on a crowded feed?** If you hesitate,
+the cover is too quiet.
+
+```text
+Bad  (topic label, no stake, no swipe reason)
+  ┌──────────────────────────┐
+  │  Tips for better          │
+  │  user onboarding          │
+  │                           │
+  └──────────────────────────┘
+
+Good (stake + payoff + swipe cue, one dominant line)
+  ┌──────────────────────────┐
+  │  60% of new users quit    │   ← largest element, reads first
+  │  in week one.             │
+  │                           │
+  │  Here's where they fall   │   ← supporting line, smaller
+  │  off — and the fix.       │
+  │                  Swipe →  │   ← explicit cue
+  └──────────────────────────┘
+```
+
+More cover-hook shapes (number / contrarian / mistake / outcome / question) with Bad→Good
+rewrites are in [`references/carousel-patterns.md`](references/carousel-patterns.md).
+
+## The narrative arc
+
+A carousel is not a list — it is a sequence engineered so each slide creates a reason to see
+the next. Pick **one** arc and map slides to it. Branch on the angle:
+
+| Arc | Use when | Slide shape |
+|---|---|---|
+| Problem → tension → reveal → payoff | persuasive / insight post | cover, stakes, the wrong way, the shift, the payoff, CTA |
+| Numbered steps / listicle | how-to, framework | cover, step 1..n, recap, CTA |
+| Myth → truth | contrarian take | cover, the myth, why it's wrong, the truth, proof, CTA |
+| Before → after | transformation / case study | cover, before, turning point, after, how, CTA |
+| Data reveal | stat-led post | cover, setup, the number, why it matters, so-what, CTA |
+
+**Momentum rule:** end each interior slide on a small open loop — a cliff line, a "but here's
+the catch", a half-finished list — so the thumb keeps swiping. Why: a slide that fully resolves
+gives permission to stop. Each arc expanded to a full slide map is in
+[`references/carousel-patterns.md`](references/carousel-patterns.md).
+
+## One idea per slide + hierarchy
+
+The line between a finished deck and one abandoned at slide 3 is **one idea per slide**.
+
+- **One point per slide** — if a slide has two ideas, split it. Why: cramming dilutes the
+  message and stalls momentum; a single point is swiped through fast and remembered.
+- **Type floor ≥ 28 px** — it is read on a phone, one-handed. Why: body text below ~28 px
+  forces a pinch-zoom, which is a stop.
+- **One dominant element per slide** — a headline, a number, or a visual that reads first.
+- **Generous white space + a consistent grid** — same margins, same headline position across
+  slides. Why: a stable frame lets the eye land instantly and the deck feels designed.
+
+```text
+Bad  — crammed, four competing points, tiny text
+  ┌──────────────────────────┐
+  │ Onboarding fixes:         │
+  │ • Shorten the signup form │
+  │ • Add a progress bar      │
+  │ • Send a day-1 email      │
+  │ • Show a sample dataset   │
+  └──────────────────────────┘
+
+Good — one idea, one dominant element, breathing room
+  ┌──────────────────────────┐
+  │  Step 1                   │
+  │                           │
+  │  Cut the signup form to   │   ← single point, large
+  │  3 fields.                │
+  │                           │
+  │  Every extra field drops  │   ← one supporting line
+  │  completion ~7%.          │
+  └──────────────────────────┘
+```
+
+## Slide count
+
+Aim for the **6–12 band, ~8 as the sweet spot** (cover + ~6 content + CTA). Reported peaks
+cluster around 7–8 slides; 8–12 is where the highest dwell/save/share rates show up. Fewer than
+5 underdelivers (not worth the swipe); more than ~15 loses people before the payoff. Treat the
+band as a guide, not a law: **length follows the arc, not a quota** — if the arc resolves in 7
+slides, ship 7, don't pad to hit a number.
+
+## The CTA closer
+
+The last slide is the payoff landing plus **one explicit ask**.
+
+- **State the payoff first**, then ask. Why: people act after value, not before it.
+- **One ask, tied to the payoff** — "Save this for your next onboarding review", "Comment your
+  worst week-one drop-off". Why: a bare "Follow for more" is generic and ignorable; an ask
+  hooked to what they just read converts.
+
+```text
+Bad   →  "Follow me for more content."     (generic, no reason)
+Good  →  "Save this for your next onboarding review.
+          Which step are you missing? Comment below."   (tied to payoff, one action)
+```
+
+## Design + export spec
+
+```text
+Canvas:    1080×1350 (4:5 portrait, recommended) or 1080×1080 (1:1 fallback). No landscape.
+Safe area: keep headline + key visual inside ~880×880; ~80 px padding on all sides.
+Type:      body ≥ 28 px (it is read on a phone).
+Export:    PDF only (1 slide = 1 page). Cap ~100-300 MB / 300 pp depending on source; keep < ~10 MB.
+Gotchas:   open the exported PDF and verify fonts — free Canva silently substitutes them;
+           disable bleed / crop marks — they add black bars to every slide;
+           preview on a PHONE before posting — the desktop view hides mobile crops.
+```
+
+Portrait 4:5 takes the most vertical space in a mobile feed, which raises dwell and the chance
+of a scroll-stop; PDF is the only export that preserves fonts and layout reliably across
+devices, since slides exported as images lose crispness and break selectable text. The full
+checklist in prose is in [`references/carousel-patterns.md`](references/carousel-patterns.md).
+
+## The feed caption handoff
+
+A document post still has a **text caption above it**, and its first 1–2 lines (before the
+"…more" fold) must restate the cover hook — that line is what people read before deciding to
+expand and swipe. Give a one-line starter that echoes the cover; you own the slides, and
+[`../linkedin-content/SKILL.md`](../linkedin-content/SKILL.md) owns the caption's full
+length/tone tuning.
+
+## Handoffs
+
+| When you actually want… | Go to |
+|---|---|
+| The account/format plan, cadence, niche, who to target | [`../linkedin-strategy/SKILL.md`](../linkedin-strategy/SKILL.md) |
+| A text or single-image feed post (just the copy) | [`../linkedin-content/SKILL.md`](../linkedin-content/SKILL.md) |
+| 1:1 DMs / connection-request outreach sequences | [`../linkedin-outreach/SKILL.md`](../linkedin-outreach/SKILL.md) |
+| To publish/schedule the deck via the LinkedIn API | [`../linkedin-api/SKILL.md`](../linkedin-api/SKILL.md) |
+| A spoken-to 16:9 talk deck (speaker support) | [`../presentations/SKILL.md`](../presentations/SKILL.md) |
+| The color / type / logo brand kit the deck is styled in | [`../brand-identity/SKILL.md`](../brand-identity/SKILL.md) |
+| The reusable tone of voice the words are written in | [`../brand-voice/SKILL.md`](../brand-voice/SKILL.md) |
+| A long-form article (the prose, not slides) | [`../article-writing/SKILL.md`](../article-writing/SKILL.md) |
+| To cross-post / schedule the finished asset elsewhere | [`../social-publisher/SKILL.md`](../social-publisher/SKILL.md) |
+
+## Anti-patterns
+
+| Anti-pattern | Why it fails | Do instead |
+|---|---|---|
+| Cover is a title slide / table of contents | No stake, no swipe reason — the deck dies at slide 1 | Lead with a curiosity/benefit headline that promises a payoff |
+| Multiple ideas crammed on one slide | Dilutes the message, stalls momentum, forces tiny text | One focused idea per slide; split the rest into more slides |
+| Body text under 28 px | Forces a pinch-zoom on mobile — that's a stop | Keep body ≥ 28 px; headlines far larger |
+| Landscape (16:9) canvas | Takes minimal feed space, low dwell, looks like a talk deck | Portrait 1080×1350 (4:5), or 1080×1080 square fallback |
+| A flat list with no arc | Nothing pulls the reader forward; they abandon early | Pick one arc; end each slide on a small open loop |
+| 20+ slides | Loses people before the payoff | Stay in the 6–12 band (~8); let the arc set the length |
+| Bare "Follow me for more" CTA | Generic, no reason to act | One explicit ask tied to the payoff (save/comment) |
+| Exporting slides as images, not a PDF | Loses crispness, breaks the document-post format | Export one PDF, 1 slide = 1 page, < 10 MB |
+| Trusting Canva's PDF fonts | Free Canva silently substitutes custom fonts | Open the exported PDF and verify fonts before upload |
+| Never previewing on a phone | Desktop view hides mobile crops and bleed bars | Open the final PDF on a phone before posting |
+
+## Verify
+
+If you wrote the spec to a file, lint it before handing off:
+
+```bash
+scripts/verify.sh path/to/carousel-spec.md
+```
+
+It checks (read-only) for a labeled cover slide, a slide count in 5–15 (warns outside 6–12), a
+CTA slide with an explicit ask, a portrait/square canvas, a ≥28 px type floor, "PDF" as the
+export target, and warns on any crammed (3+ bullet) interior slide. An empty or clean target
+passes.

package/skills/linkedin-carousels/evals/README.md

@@ -0,0 +1,13 @@
+# Evals — linkedin-carousels
+
+`cases.yaml` is a routing + capability spec, not an automated test suite. `should_trigger`
+lists prompts whose description should route to this skill (including a non-obvious
+cover-repair / export-size phrasing and a Spanish/Catalan one); `should_not_trigger` lists
+near-miss prompts that must route to a real sibling (`linkedin-content`, `linkedin-strategy`,
+`presentations`, `linkedin-api`, `brand-identity`) with the reason. `capability` is a rubric a
+grader (human or LLM) applies to an actual produced carousel spec to confirm the skill changes
+the output — not just that it loaded. There is no bundled runner: check routing by inspection
+(does the description's "Use when" / Triggers / NOT boundary cover each prompt?), or feed the
+file to the repo's eval tooling if one is present. To pressure-test capability, run the
+scenario and score the result against `must_include`, then optionally lint the produced spec
+with `../scripts/verify.sh path/to/spec.md`.

package/skills/linkedin-carousels/evals/cases.yaml

@@ -0,0 +1,62 @@
+skill: linkedin-carousels
+
+# Prompts that MUST load `linkedin-carousels`. The skill owns the SLIDES of a
+# swipeable LinkedIn document post: the cover hook, the narrative arc, one idea
+# per slide with real hierarchy, a CTA closer, and the PDF export spec. It does
+# NOT own the account plan (linkedin-strategy), the text feed post
+# (linkedin-content), the API upload (linkedin-api), or a 16:9 talk deck
+# (presentations).
+should_trigger:
+  - prompt: "Turn this blog post into a LinkedIn carousel — I want the slides laid out."
+    why: "Core job: transform a source into a slide-by-slide swipeable document post."
+
+  - prompt: "My carousel cover slide gets no swipes. Fix the hook."
+    why: "Non-obvious — a symptom, no word 'design'. Cover-hook repair is the highest-leverage thing the skill owns (the swipe-to-slide-2 bet)."
+
+  - prompt: "Design a swipeable PDF document post about onboarding: cover + 6 points + CTA."
+    why: "Explicit slide build with cover/interior/CTA structure — the exact deliverable."
+
+  - prompt: "What size should I export my LinkedIn carousel PDF at so it isn't cropped on mobile?"
+    why: "Non-obvious — sounds like a tooling question, but canvas/safe-area/PDF export is squarely in this skill's design spec."
+
+  - prompt: "My slides are crammed with bullets and there's no story — restructure one idea per slide with an arc."
+    why: "Structure + hierarchy repair: the one-idea-per-slide rule and narrative arc are core."
+
+  - prompt: "Convierte este post en un carrusel de LinkedIn amb portada i CTA."
+    why: "Spanish/Catalan phrasing for the core job — design a carousel with a cover and CTA."
+
+# NEAR-MISS prompts that must NOT load `linkedin-carousels`. Each routes to the
+# real sibling that owns it. The recurring trap: anything that touches LinkedIn
+# but is the account plan, the feed copy, the API, or a spoken-to deck.
+should_not_trigger:
+  - prompt: "Write me a LinkedIn text post with a strong hook line about remote work."
+    route_to: linkedin-content
+    why: "A single block of feed copy with no slides is the text/single-image post — owned by linkedin-content, not a multi-slide document."
+
+  - prompt: "Plan my LinkedIn content for the quarter — which formats and how often to post."
+    route_to: linkedin-strategy
+    why: "Account-level format/cadence/niche planning is the strategy, not the design of one deck."
+
+  - prompt: "Design a 16:9 deck for my conference talk with a speaker track."
+    route_to: presentations
+    why: "A spoken-to talk deck (landscape, speaker support) is presentations; a carousel is a portrait phone-read document, opposite reading context."
+
+  - prompt: "Schedule and auto-publish this carousel to LinkedIn via the API."
+    route_to: linkedin-api
+    why: "Programmatic upload/scheduling is linkedin-api; this skill stops at the designed PDF spec."
+
+  - prompt: "Define our brand colors and fonts for all our content."
+    route_to: brand-identity
+    why: "The reusable color/type/logo system is brand-identity; the carousel consumes that kit, it does not author it."
+
+# CAPABILITY: prove the skill changes the output, not just routing.
+capability:
+  - scenario: "Turn the topic 'why most onboarding flows lose users in week one' into a designed LinkedIn carousel, producing the full slide-by-slide spec plus the export spec."
+    must_include:
+      - "A cover slide with a curiosity/benefit headline (a stake/payoff, not a topic label), one dominant element, white space, and an explicit swipe cue — engineered to win the swipe-to-slide-2 bounce."
+      - "A named narrative arc (e.g. problem -> reveal -> payoff) with each interior slide mapped to it and ending on a small open loop that pulls to the next slide."
+      - "One focused idea per interior slide — a single headline line, not a multi-point bullet dump."
+      - "Total slide count in the 6-12 band (~8), with the length justified by the arc resolving, not by a quota."
+      - "A CTA closer slide that restates the payoff and makes one explicit ask tied to it (save/comment), not a bare 'follow me'."
+      - "A design/export spec: portrait 1080x1350 (or 1080x1080 fallback), ~880x880 safe area with ~80 px padding, >=28 px type floor, export to PDF (1 slide = 1 page, < 10 MB), and a phone-preview / font-substitution check."
+      - "A one-line feed caption starter whose first line restates the cover hook, with the handoff to linkedin-content for full caption tuning noted."

package/skills/linkedin-carousels/references/carousel-patterns.md

@@ -0,0 +1,200 @@
+# Carousel patterns
+
+Offloaded depth for `linkedin-carousels`. SKILL.md points here from the cover, narrative-arc,
+and export sections. Everything here is a build aid — the SKILL.md rules still govern.
+
+## Slide-by-slide spec template
+
+Produce one block per slide. Fill every field; "—" if a field genuinely doesn't apply.
+
+```text
+SLIDE 0 — CONTEXT (not exported; for the build)
+  Topic / angle:
+  Source (post/article/idea):
+  Brand kit ref (docs or assumption):
+  Voice ref (docs or assumption):
+  Chosen arc:
+  Slide count:
+
+SLIDE 1 — COVER
+  Headline (dominant, reads first):
+  Supporting line (smaller):
+  Dominant visual / treatment:
+  Swipe cue (arrow / "Swipe →" / 1/N counter):
+
+SLIDE n — INTERIOR  (repeat per slide)
+  Single idea (one sentence):
+  Headline line (large):
+  Supporting line (one, optional):
+  Dominant visual / number:
+  Open-loop close (cliff line into the next slide):
+
+SLIDE N — CTA CLOSER
+  Payoff restatement (the satisfying landing):
+  One explicit ask (save / comment / share / visit):
+```
+
+## Cover-hook pattern library
+
+Each shape, with a Bad (label) → Good (stake/payoff) rewrite. Pick the one that fits the angle.
+
+```text
+1. NUMBER
+   Bad   "Things to improve onboarding"
+   Good  "5 onboarding mistakes that cost you 60% of new users"
+
+2. CONTRARIAN
+   Bad   "Onboarding best practices"
+   Good  "Your onboarding checklist is why users churn. Delete half of it."
+
+3. MISTAKE
+   Bad   "Onboarding tips"
+   Good  "The week-one mistake almost every SaaS makes (and how to spot it)"
+
+4. OUTCOME
+   Bad   "How onboarding works"
+   Good  "How we cut week-one churn from 60% to 18% — the exact flow"
+
+5. QUESTION
+   Bad   "About user retention"
+   Good  "Why do 60% of your signups never come back? It's not the product."
+
+6. NOBODY-TELLS-YOU
+   Bad   "Onboarding guide"
+   Good  "Nobody tells you that onboarding fails before the user even logs in"
+
+7. STAKES / COST
+   Bad   "User drop-off"
+   Good  "Every extra signup field drops completion ~7%. Here's the math."
+```
+
+Rule across all of them: the headline names a payoff or a stake, one line dominates, and a
+swipe cue tells the reader the format is a deck.
+
+## Narrative-arc blueprints
+
+Each arc expanded to a full slide map for an ~8-slide deck. Adjust count to fit the content —
+length follows the arc, not a quota.
+
+```text
+ARC A — PROBLEM → TENSION → REVEAL → PAYOFF   (persuasive / insight)
+  1 Cover: the stake (curiosity/benefit headline + swipe cue)
+  2 The problem: name the pain the reader feels
+  3 The tension: why the obvious fix doesn't work
+  4 The reveal: the actual cause / the shift in thinking
+  5 The mechanism: how the fix works
+  6 The proof: a number, result, or example
+  7 The payoff: what changes once they apply it
+  8 CTA: restate payoff + one ask (save/comment)
+
+ARC B — NUMBERED STEPS / LISTICLE   (how-to, framework)
+  1 Cover: "N steps to <outcome>" + swipe cue
+  2..7 One step per slide (one idea, large type, open-loop close)
+  8 Recap or the one step that matters most + CTA ask
+
+ARC C — MYTH → TRUTH   (contrarian)
+  1 Cover: the contrarian claim + swipe cue
+  2 The myth: what most people believe
+  3 Why it's wrong: the flaw in the belief
+  4 The truth: the better model
+  5 Proof: evidence the truth holds
+  6 So-what: what to do differently
+  7 CTA: restate the truth + one ask
+
+ARC D — BEFORE → AFTER   (transformation / case)
+  1 Cover: the transformation ("from X to Y") + swipe cue
+  2 Before: the starting state and its cost
+  3 The turning point: what changed
+  4 After: the new state
+  5 How: the steps that got there
+  6 Lesson: what's transferable
+  7 CTA: restate result + one ask
+
+ARC E — DATA REVEAL   (stat-led)
+  1 Cover: tease the number, don't reveal it ("60% of users do THIS")
+  2 Setup: why the number matters / how it was measured
+  3 The number: reveal it as the dominant element
+  4 Why it happens: the cause
+  5 So-what: the implication for the reader
+  6 Action: what to change
+  7 CTA: restate the insight + one ask
+```
+
+Momentum rule applies to every arc: end each interior slide on a small open loop (a cliff
+line, a "but…", a half-answered question) so the reader swipes to resolve it.
+
+## Design + export checklist (in words)
+
+- **Canvas.** Use 1080×1350 px (4:5 portrait) — it claims the most vertical feed space, which
+  raises dwell and scroll-stop chance. 1080×1080 (1:1 square) is the safe fallback. Never
+  landscape for the feed — it reads as a talk deck and takes minimal space.
+- **Safe area.** Keep every headline and key visual inside a central ~880×880 region, ~80 px
+  padding on all sides. LinkedIn's in-feed viewer and mobile cropping can shave the edges.
+- **Type floor.** Body text ≥ 28 px; headlines far larger so hierarchy reads at a glance. It
+  is consumed one-handed on a phone — anything smaller forces a pinch-zoom, which is a stop.
+- **Export format.** PDF only — it preserves fonts and layout across devices. LinkedIn also
+  accepts PPTX/DOCX, but only PDF renders consistently. One slide = one PDF page.
+- **File size.** Reported ceilings range ~100-300 MB / 300 pages depending on source (LinkedIn's
+  own Help page states 100 MB / 300 pages); keep it under ~10 MB so it loads fast in-feed.
+  Compress images if needed.
+- **Font substitution.** On free Canva, custom fonts can be swapped at PDF export with no
+  warning. Open the exported PDF and confirm the fonts before upload.
+- **Bleed / crop marks.** If accidentally enabled, they add black bars around every slide.
+  Disable them in export settings.
+- **Phone preview.** Open the final PDF on an actual phone before posting — the desktop preview
+  hides crops and bars that only show on mobile.
+
+## Worked example — full 8-slide deck (Arc A)
+
+Topic: "Why most onboarding flows lose users in week one." Arc A (problem → reveal → payoff).
+
+```text
+SLIDE 1 — COVER
+  Headline:     "60% of new users quit in week one."
+  Supporting:   "It's almost never the product. Here's where they fall off."
+  Dominant:     the "60%" set large; headline reads first
+  Swipe cue:    "Swipe →"  (1/8)
+
+SLIDE 2 — THE PROBLEM
+  Idea:         users churn before they reach value
+  Headline:     "They never hit the 'aha' moment."
+  Supporting:   "Most leave before doing the one thing that makes the product click."
+  Open loop:    "And the usual fix makes it worse →"
+
+SLIDE 3 — THE TENSION
+  Idea:         adding more onboarding steps backfires
+  Headline:     "More onboarding ≠ better onboarding."
+  Supporting:   "Every extra setup screen is another exit."
+  Open loop:    "So what actually keeps them? →"
+
+SLIDE 4 — THE REVEAL
+  Idea:         time-to-value is the real metric
+  Headline:     "Cut the time to first value."
+  Supporting:   "Get them to one real win fast — defer everything else."
+  Open loop:    "Here's how, step by step →"
+
+SLIDE 5 — THE MECHANISM
+  Idea:         shorten the signup form
+  Headline:     "Step 1: 3 fields, not 9."
+  Supporting:   "Every extra field drops completion ~7%."
+  Open loop:    "Then remove the empty state →"
+
+SLIDE 6 — THE MECHANISM (cont.)
+  Idea:         seed the first win
+  Headline:     "Step 2: pre-load a sample so they see value in 30 seconds."
+  Supporting:   "Don't make day one a blank screen."
+  Open loop:    "The result? →"
+
+SLIDE 7 — THE PAYOFF
+  Idea:         the outcome of the changes
+  Headline:     "Week-one churn: 60% → 18%."
+  Supporting:   "Same product. Faster path to the 'aha'."
+  Open loop:    "Want the checklist? →"
+
+SLIDE 8 — CTA CLOSER
+  Payoff:       "Faster time-to-value is the whole game in week one."
+  Ask:          "Save this for your next onboarding review. Which step are you missing? Comment below."
+
+CAPTION STARTER (hand to linkedin-content for tuning):
+  "60% of new users quit in week one — and it's almost never the product. 👇"
+```

package/skills/linkedin-carousels/scripts/verify.sh

@@ -0,0 +1,160 @@
+#!/usr/bin/env bash
+#
+# verify.sh — static linter for a `linkedin-carousels` slide-by-slide spec.
+#
+# WHAT IT DOES (read-only; never edits a file)
+#   Network-free static checks over a produced carousel spec (markdown/text):
+#     1. HARD  no cover slide is labeled (cover / slide 1 marker).
+#     2. HARD  no CTA closer with an explicit ask verb (follow/comment/save/share/visit/subscribe).
+#     3. HARD  no portrait/square canvas named (1080x1350 or 1080x1080).
+#     4. HARD  no type floor named (28 or larger) AND/OR no "PDF" export target.
+#     5. HARD  slide count outside 5-15.
+#     6. SOFT  slide count outside the 6-12 band (warn).
+#     7. SOFT  an interior slide block lists 3+ bullet points (crammed; one-idea-per-slide).
+#
+#   Hard violations exit 1. Soft violations warn and exit 0 (spec is judgement).
+#   An empty or clean target is a clean pass (exit 0) — no false failure.
+#
+# HOW TO RUN (point it at YOUR spec, not the skills repo)
+#   ./verify.sh path/to/carousel-spec.md   # check one file
+#   ./verify.sh                            # scan ./ for *.{md,mdx,txt}
+#   ./verify.sh --path specs               # scan a subdirectory
+#   ./verify.sh --strict                   # treat soft warnings as failure (exit 1)
+#
+# EXIT CODES
+#   0  clean, empty target, or warnings-only without --strict
+#   1  a hard violation, or --strict with any warning
+#   2  bad usage
+#
+# POSIX sh, no external deps beyond grep/wc/sort. Runs on stock macOS /bin/sh.
+
+set -eu
+
+if [ -t 1 ]; then
+  RED=$(printf '\033[31m'); GRN=$(printf '\033[32m'); YEL=$(printf '\033[33m'); NC=$(printf '\033[0m')
+else
+  RED=''; GRN=''; YEL=''; NC=''
+fi
+
+warn_count=0
+fail_count=0
+
+ok()   { printf '%s[ ok ]%s %s\n'  "$GRN" "$NC" "$*"; }
+warn() { printf '%s[warn]%s %s\n'  "$YEL" "$NC" "$*"; warn_count=$((warn_count + 1)); }
+fail() { printf '%s[fail]%s %s\n'  "$RED" "$NC" "$*"; fail_count=$((fail_count + 1)); }
+
+usage() { sed -n '2,33p' "$0" | sed 's/^#\{0,1\} \{0,1\}//'; }
+
+TARGET=""
+STRICT=0
+while [ $# -gt 0 ]; do
+  case "$1" in
+    --strict) STRICT=1 ;;
+    --path) shift; TARGET="${1:-}" ;;
+    -h|--help) usage; exit 0 ;;
+    -*) printf 'unknown option: %s\n' "$1" >&2; usage >&2; exit 2 ;;
+    *) TARGET="$1" ;;
+  esac
+  shift
+done
+[ -n "$TARGET" ] || TARGET="."
+
+# --- collect files ----------------------------------------------------------
+FILES=""
+if [ -f "$TARGET" ]; then
+  FILES="$TARGET"
+elif [ -d "$TARGET" ]; then
+  FILES=$(find "$TARGET" -type f \
+    \( -name '*.md' -o -name '*.mdx' -o -name '*.txt' \) 2>/dev/null \
+    | grep -v '/node_modules/' || true)
+else
+  printf 'no such file or directory: %s\n' "$TARGET" >&2
+  exit 2
+fi
+
+if [ -z "$FILES" ]; then
+  ok "no spec files found under '$TARGET' — nothing to check"
+  exit 0
+fi
+
+# explicit-ask verbs that mark a real CTA (case-insensitive)
+ASK_VERBS='follow|comment|save|share|repost|visit|subscribe|download|sign up|book|grab'
+
+checked=0
+for f in $FILES; do
+  [ -f "$f" ] || continue
+  [ -s "$f" ] || continue
+  checked=$((checked + 1))
+
+  # 1. HARD: a labeled cover slide.
+  if grep -Eiq '(^|[^a-z])cover([^a-z]|$)|slide *0*1 *[—:-]' "$f"; then
+    :
+  else
+    fail "$f: no cover slide labeled — the cover wins or loses the swipe to slide 2"
+  fi
+
+  # 2. HARD: a CTA closer with an explicit ask verb.
+  if grep -Eiq 'cta' "$f" && grep -Eiq "$ASK_VERBS" "$f"; then
+    :
+  elif grep -Eiq "$ASK_VERBS" "$f"; then
+    warn "$f: an ask verb is present but no 'CTA' closer slide is labeled — label the closer"
+  else
+    fail "$f: no CTA closer with an explicit ask (follow/comment/save/share/visit) — every deck needs one ask"
+  fi
+
+  # 3. HARD: a portrait/square canvas.
+  if grep -Eiq '1080 *[x×] *1350|1080 *[x×] *1080' "$f"; then
+    :
+  else
+    fail "$f: no portrait/square canvas named (expected 1080x1350 or 1080x1080) — no landscape for the feed"
+  fi
+
+  # 4. HARD: a type floor (>=28) and PDF export target.
+  if grep -Eiq '(2[89]|[3-9][0-9]|[1-9][0-9][0-9]) *px' "$f"; then
+    :
+  else
+    fail "$f: no type floor >= 28 px stated — body is read on a phone, anything smaller is a stop"
+  fi
+  if grep -Eiq '(^|[^a-z])pdf([^a-z]|$)' "$f"; then
+    :
+  else
+    fail "$f: no PDF export target named — PDF is the only format that renders consistently"
+  fi
+
+  # 5/6. slide count. Count distinct 'slide N' markers (case-insensitive).
+  count=$(grep -Eio 'slide *0*[0-9]+' "$f" 2>/dev/null \
+    | grep -Eo '[0-9]+' | sort -un | wc -l | tr -d ' ')
+  if [ "${count:-0}" -gt 0 ]; then
+    if [ "$count" -lt 5 ] || [ "$count" -gt 15 ]; then
+      fail "$f: ${count} slides — outside the 5-15 hard range (< 5 underdelivers, > 15 loses people)"
+    elif [ "$count" -lt 6 ] || [ "$count" -gt 12 ]; then
+      warn "$f: ${count} slides — outside the 6-12 sweet band (~8 is the target)"
+    fi
+  fi
+
+  # 7. SOFT: a crammed interior slide — 3+ bullet lines within ~12 lines of a slide marker.
+  #    Heuristic: if the file has any run of 3+ consecutive bullet lines, warn once.
+  bullets=$(grep -Ec '^[[:space:]]*([-*•]|[0-9]+\.) ' "$f" 2>/dev/null || true)
+  if [ "${bullets:-0}" -ge 3 ]; then
+    max_run=$(awk '
+      /^[[:space:]]*([-*•]|[0-9]+\.) /{run++; if(run>max)max=run; next}
+      {run=0}
+      END{print max+0}' "$f")
+    if [ "${max_run:-0}" -ge 3 ]; then
+      warn "$f: a slide block lists ${max_run}+ consecutive bullets — keep one idea per slide, split the rest"
+    fi
+  fi
+done
+
+printf '\n'
+if [ "$fail_count" -gt 0 ]; then
+  fail "$fail_count hard violation(s), $warn_count warning(s) across $checked file(s)"
+  exit 1
+fi
+if [ "$warn_count" -gt 0 ]; then
+  warn "$warn_count warning(s) across $checked file(s)"
+  [ "$STRICT" -eq 1 ] && { printf '%s--strict: treating warnings as failure%s\n' "$RED" "$NC"; exit 1; }
+  exit 0
+fi
+ok "clean — $checked file(s) checked, no violations"
+exit 0