wand-decks-kit 0.8.6 → 0.9.0

package/PROMPT.md

@@ -10,6 +10,41 @@ is the workflow and the design system; the skill is the source of truth for comm
 
 ---
 
+
+## The guideline deck (0.9)
+
+Since 2026-09-23 the design source of truth is a real deck —
+`Vivicta_Session_Deck.pptx`, the standard Wand sales deck as it is actually sent
+— not the earlier reference PDFs. Everything below was **measured from it**, and
+`brand/BRAND.md` and the kit's `T` carry the counts:
+
+- **Tokens.** Dark ground is `240F48`, flat. Titles `FFFFFF`, body `F5F5F5`,
+  captions `BBBBBB`, lists in cards `C6BFE4`. Cards: solid `2A1F5C`, muted
+  `1B1738`, stroke `5B4BB0`. Accent text on dark `9481FF`. The old
+  `0F0C26 / B7B0DC / 2C2460` set was the reference PDFs' rendering, not the deck.
+- **Interior dark slides carry no art.** The gradient compositions belong to
+  the cover (`cover-art.png`, orb baked in) and the closing (`closing-art.png`
+  plus `orb-glow.png` centred). `background()` paints a flat colour.
+- **Chrome.** Tag at `0.42, 0.24`; mark at `9.40, 0.26`; footer band at
+  `y 5.29, h 0.34`, 7pt: `WAND · <deck>` left, `<confidential> · nn/NN` right.
+- **Cover.** Title at `0.42, 1.79`, 39pt shrinking to 30; body at `3.15`; a
+  `PRESENTED TO` block bottom-right (`presentedTo: [name, title]`).
+- **Closing.** Orb at `3.98, 0.69, 2.03²`; the line at 33pt shrinking to 22 —
+  a real ask is a sentence, not "Thank You".
+- **Five layouts the deck has and the kit did not:** `agenda`, `team` (with a
+  headshot slot), `fourCards`, `deliverColumns`, `coDelivery`. They are the
+  slides a sales session adds to the standard deck; a tester had to hand-build
+  all five.
+- **The brand gate.** `wand.js audit` runs `tools/brandlint.py`: every text run
+  in Geist, every colour a token, the 10 × 5.625 canvas, a footer on every
+  content slide. On the guideline deck itself it names exactly the slides the
+  tester retexted by hand (Arial on S14, Calibri on S07) — which is the point:
+  an edit-only job reports those and asks, rather than restyling unasked.
+
+`card()` variants are theme-aware now: `solid` on a light slide is a white
+card with a hairline, not a dark purple box hiding light ink — the agenda in
+the template rendered as four black boxes before that fix.
+
 ## 0. Ground truth: the reference decks
 
 Two decks in Project knowledge define the target output. **They are the specification.** When this

package/SKILL.md

@@ -4,7 +4,7 @@ description: Rebrand, merge, edit or author a PowerPoint deck on the Wand design
 compatibility: Requires Node.js 18+ and code execution enabled in claude.ai (Settings > Features). No network access needed - dependencies are bundled. Python 3 is used to read a source .pptx and to embed fonts; without it the kit can still author and check a deck, but the output will render in Arial on machines without Geist.
 license: Proprietary - Wand AI
 metadata:
-  version: "0.7.0"
+  version: "0.9.0"
 ---
 
 
@@ -61,7 +61,16 @@ node wand.js doctor
 
 If `node_modules` is somehow missing, `npm install` in this directory.
 
-## The four jobs
+## Before any build: show the plan, then wait
+
+Write the slide-by-slide mapping in the chat **before** writing a line of JSON:
+`S07 (timeline, 6 phases) → phaseCards`, one line per slide, with the reason
+where it is not obvious. Then build. A field run skipped this, built the wrong
+thing, and paid a full second pass. This is not optional for a rebuild, a merge
+or an adaptation; only "author from a brief" with fewer than eight slides may
+go straight to the spec.
+
+## The six jobs
 
 ### Rebrand a deck they attached
 
@@ -89,6 +98,45 @@ to cover — and it says so.
 spec. Ask which deck leads if they have not said. Flag duplicates rather than
 silently dropping them, and list every merge in the review note.
 
+### Adapt a deck and keep its graphics
+
+The ask sounds like a rebrand but is not: *"take our standard deck, drop in an
+agenda, a team slide and a demo placeholder, keep everything else as is."* They
+want the source's own slides and art preserved, with a few slides retexted or
+added. **This kit re-authors from a spec; it cannot copy a source slide's
+shapes.** Say that in one line, then edit the source `.pptx` directly with the
+pptx tools — and still run this kit's gates on the result:
+
+```bash
+node wand.js audit <edited-deck>.pptx              # file, pdf text, renders + contact sheet
+node wand.js audit <edited-deck>.pptx --dump work/visible_dump.txt   # + coverage
+```
+
+`audit` runs every gate that does not need a spec. A deck delivered without it
+has been checked by nobody. Say which slides were lifted unchanged, which were
+retexted, and that the source's fonts (not Geist) carry through.
+
+### Edit a deck they attached — not built here
+
+"Change the date on the cover", "swap slide 4's title", "add our logo to the
+partner slide". They want *that*, and nothing else touched.
+
+1. Do exactly the edit, directly in the `.pptx` with the pptx tools. Do not
+   rebuild, do not restyle, do not "tidy" anything nearby.
+2. Then run the guideline over the whole deck:
+
+   ```bash
+   node wand.js audit <their-deck>.pptx
+   ```
+
+   The **brand** gate lists every slide where a text run is not Geist, a colour
+   is not a token, the canvas is not the 10 × 5.625 grid, or a content slide
+   has no footer.
+3. Deliver the edited deck first. Then report the inconsistencies the gate
+   found — per slide, in one line each — and **ask whether to fix them**. Do not
+   fix them unasked: an edit-only job that comes back restyled is a broken
+   promise, even when the restyle is right.
+
 ### Edit a deck this kit built
 
 ```bash
@@ -124,6 +172,13 @@ optional — **sweep the entire spec for that pattern before re-running anything
 Paying a full cycle per instance is the slow way to learn the same fact
 repeatedly.
 
+## One skill at a time
+
+If `wand-decks` or `wand-docs` is installed beside `wand-design`, both may
+fire on the same file. **`wand-design` leads; this file yields.** Do not run
+two setups, and tell the user once: remove the older per-family skill, the
+combined one covers it.
+
 ## Rules that decide whether the output is usable
 
 1. **Faithful means verbatim.** No consolidating slides, no paraphrasing to
@@ -139,6 +194,14 @@ repeatedly.
 4. **Icons are named.** `node wand.js glyphs` lists all 75. Pick by meaning.
    They recolour to the theme, so there is no polarity to get wrong.
 5. **Never invent content.** If the source is ambiguous, ask.
+6. **A FAIL in the gate table is not shippable.** Not "acceptable two-line
+   wraps", not "probably fine" — either fix it and re-run, or list each failing
+   item with its measurement in the review note and let the user decide. A
+   field run shipped on `FAIL fit + bounds` with a reassurance instead of a
+   list; the reassurance was not checked against anything.
+7. **Verbatim applies to rebuilds, merges and lifted slides.** When the job is
+   to author or retext a slide, say which slides were retexted — the review
+   note separates *lifted verbatim* from *rewritten*.
 
 ## Useful commands
 
@@ -150,6 +213,7 @@ repeatedly.
 | `node wand.js check <spec>` | fit in real Geist, bounds, assets |
 | `node wand.js preview <spec>` | writes `preview/trace.js` for the browser preview |
 | `node wand.js inspect <deck>` | parts, relationships, embedded fonts |
+| `node wand.js audit <deck> [--dump f]` | every spec-free gate on any .pptx, including the **brand** gate (fonts, colours, canvas, footer) — use on decks edited outside the kit |
 
 ## Deliver
 

package/docs/LAYOUTS.md

@@ -24,6 +24,11 @@ full-bleed three) `deck` / `n` / `total` for the footer.
 
 | Layout | Params | Use for |
 |---|---|---|
+| `agenda` | `items[{title,body}]` | What we'll cover — 2–5 numbered cards (guideline S03) |
+| `team` | `items[{name,role,initials}]` | Who's in the room — headshot slot per person (guideline S02) |
+| `fourCards` | `items[{title,body}]` | Demo objectives, pillars — four titled cards under a lede (S12) |
+| `deliverColumns` | `columns[{heading,lead,body}]` | How we deliver — three flat columns with an accent bar (S14) |
+| `coDelivery` | `items[{title,body}]` | Numbered rows in two columns, up to six (S13) |
 | `threeUp` | `items[{eyebrow,title,pill,body,emph}], arrows` | 3 comparable things. `arrows: true` for a progression |
 | `fourUpIcon` | `items[{icon,title,body}], kickerText` | 4 attributes or capabilities |
 | `sixUpIcon` | `items[{icon,title,body}]` | 6 features. Don't use it three slides running |