sheleg-design-skill 1.66.4 → 1.66.6

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,3 +1,41 @@
1
+ ## 1.66.6 (2026-11-11): the short path first, retired sources out
2
+
3
+ Family audit 2026-11-11 (SD1, SD2, SD4, SD5, SD7). The description is unchanged.
4
+
5
+ - **Small work reads one section, not the whole decision layer** (SD1). `SKILL.md` sends an
6
+ update or a bounded fix straight to [Focused visual work](plugins/sheleg-design/skills/sheleg-design/CREATIVE_DIRECTOR.md#focused-visual-work),
7
+ and that section now opens `CREATIVE_DIRECTOR.md` (it sat at line 444 of 524). Every rule is
8
+ kept; this is a reordering, plus one line saying new work goes on to Act 1.
9
+ - **Lazyweb is gone from the live text** (SD2): its MCP is retired. The reference servers are
10
+ Refero and Mobbin in `SKILL.md`, `CREATIVE_DIRECTOR.md`, `APPLYING.md`,
11
+ `DESIGN_SYNC_BRIDGE.md` §4 (heading and validator pin together), `MOBILE_SURFACES.md` (Refero
12
+ takes the table row), the director-record template, the Cursor rule (which named only Lazyweb),
13
+ the README, the live release scenarios and two sample records. Dated records keep their wording.
14
+ - **The Chromium fallback names runnable routes** (SD4): Chrome Dev or the agent's own headless
15
+ Chromium, with the Lighthouse CLI, instead of the chrome-devtools MCP. WebKit stays first;
16
+ `test/audit_regressions/safari-first.py` pins the new wording.
17
+ - **The component kit is recorded, not asked** (SD5): `shadcn/ui` is recorded once per project
18
+ with a one-line notice; the question is asked only when an existing component layer conflicts.
19
+ `COMPONENT_LAYER.md` says the same.
20
+ - **The director-record template carries no em dash** (SD7): 20 replaced by colons, commas and
21
+ parentheses; `--check-record` accepts both forms. The style-pack template is filed as B-147.
22
+ - Body 4,740 of 4,750 tokens (`audit_skill.py --house`). The Figma survey figures in *The craft
23
+ bar* were checked against the published report on 2026-11-11 (n=906; 58, 47, 36, 35, 15 per
24
+ cent) and are unchanged.
25
+
26
+ ## 1.66.5 (2026-10-27): route misses join the Triggers section
27
+
28
+ - **Route misses join `## Triggers`** (hub board row B-267), each inside its existing entry:
29
+ «переделай дизайн» beside "redesign", «тёмная тема, тёмную тему» beside "light/dark theme"
30
+ (the hub stemmer cannot reach «тёмную» from «тёмная»), "theme tokens" / «токены темы» beside
31
+ "design tokens", and "contrast" / «контрастность» beside "palette, colors" (the palette gate
32
+ computes contrast ratios, so contrast is this skill's ground). The description is unchanged.
33
+ Accessibility checks are not added: the family delegates that lane on the record (hub B-140),
34
+ and this skill's accessibility evidence section is scoped, not routed.
35
+ - **Body back inside the 4,750-token working limit** (4,744): the triggers cost 41 tokens, paid
36
+ by dropping a historical aside in *Choosing between packs* and two wordy clauses (Figma, the
37
+ unreachable `npx` case). No rule changed. The Cursor mirror is synced.
38
+
1
39
  ## 1.66.4 (2026-10-25): descriptions within a 300-character budget
2
40
 
3
41
  - **Description within a 300-character budget** (hub board row B-255 (a)): Claude Code lists every
package/README.md CHANGED
@@ -309,13 +309,13 @@ stops an agent inventing motion to fill the silence.
309
309
  The kits are not part of the install. `--kit` fetches one on demand, which is how
310
310
  the skill stays documentation while still having real components to hand.
311
311
 
312
- ## Reference sweeps (Refero, Lazyweb, Mobbin)
312
+ ## Reference sweeps (Refero, Mobbin)
313
313
 
314
314
  A style pack locks *how it looks*. It says nothing about what a good version of
315
315
  the screen you're about to build actually **contains**. For a new design or a
316
316
  redesign of a brand or flagship surface the sweep is not optional: through
317
- whichever reference servers are connected — Refero, [Lazyweb](https://www.lazyweb.com),
318
- Mobbin where the account is authorized — the skill looks at five or more shipped
317
+ whichever reference servers are connected (Refero, and Mobbin where the account
318
+ is authorized) the skill looks at five or more shipped
319
319
  products for the target screen (signup, onboarding, paywall, pricing, checkout,
320
320
  dashboard, settings) before laying anything out, and records what it takes from
321
321
  each. An empty sweep is recorded as `none found` with where it looked.
@@ -260,7 +260,8 @@ The cinematic layer is out of scope here; only the style half applies.
260
260
  nothing scroll-driven, transitions off under `prefers-reduced-motion`.
261
261
  - Honest state: no fake "connected", no optimistic spinner where live state
262
262
  exists; degraded renders visibly degraded.
263
- - If Lazyweb MCP tools (`mcp__lazyweb__*`) are available, sweep real-world
263
+ - If reference MCP tools are available (Refero `mcp__refero__*`, Mobbin
264
+ `mcp__mobbin__*` where the account is authorized), sweep real-world
264
265
  references for the target screen before laying it out — they inform layout,
265
266
  hierarchy and content order only; palette, type and motion stay the pack's.
266
267
  Optional: absent the tools, work from the pack alone.
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "sheleg-design-skill",
3
- "version": "1.66.4",
3
+ "version": "1.66.6",
4
4
  "description": "Design taste as an installable agent skill. Cinematic scroll-driven landing pages built on one scroll clock and layered degrade-to-calm motion, a motion doctrine that decides whether to animate before it decides how, three calibration dials, and thirty-nine locked style packs with ready-made design tokens \u2014 instrument-console, editorial-luxury, workbench, briefing-room, atrium, orchard, outrank, babylove, patchbay, nameplate, field-notes, cyclorama, showroom, blueprint, prism, maquette, scoreboard, datasheet, manpage, pigeonhole, roster, ora, tenor, paperclip, ledger, router, daylight, notation, almanac, vitrine, proscenium, awning, bulletin, rimlight, onionskin, deskmate, test-drive, surveyor and chorus. Colour, slop and fork-reciprocity gates run as scripts, not opinions. Works with Cursor, Claude Code and any agent that reads a SKILL.md.",
5
5
  "bin": {
6
6
  "sheleg-design-skill": "bin/cli.js"
@@ -3,7 +3,7 @@
3
3
  "name": "sheleg-design",
4
4
  "displayName": "SHELEG Design",
5
5
  "description": "SHELEG Design methodology: cinematic scroll-driven landing pages (single scroll clock, layered degrade-to-calm motion, WebGL particle formations), a motion doctrine that decides whether to animate before it decides how, and thirty-nine pluggable visual style packs, each with a ready-made token layer — instrument-console (near-black aerospace console; technical systems and infrastructure), editorial-luxury (cream and espresso; editorial, research, premium B2B), workbench (quiet light/dark UI; dashboards, admin, internal tools), briefing-room (dark 16:9 deck; investor and board presentations), atrium (cream daylight; consumer health and high-trust DTC), babylove (white with orange; SEO SaaS, long time-to-value, disconnected states), patchbay (black with mint-cyan; engines, buses, pipelines, OSS front doors), nameplate (cool slab with coral; press, certification and third-party trust), rimlight (white with blue light; studios, services and case studies), onionskin (white working sheet; developer and AI infrastructure), deskmate (warm beige with a dusk bleed; AI coworkers and chat-native agents), test-drive (warm paper with one coral; self-serve SaaS proven by the live product), surveyor (peach paper with teal and pink; visibility, monitoring and benchmark tools), chorus (warm paper under a crosshair grid, with a cut-corner bubble; AI-search visibility, brand monitoring and community marketing), outrank (white with violet; one brand across marketing and product), orchard (warm oat slabs; friendly biotech and wellness), field-notes (ruled green paper; auditable open-source developer tools), showroom (white gallery; product-led companies selling the app), blueprint (white technical stock; vector, storage and query infrastructure), prism (iridescent hard edge; command-first OSS infrastructure), maquette (near-black table; enterprise data infrastructure), cyclorama (looping pastel field; applied-AI consultancies), scoreboard (warm paper; growth, ads and accumulating metrics), datasheet (off-white specification; fraud, identity and device intelligence), manpage (cream manual; APIs, SDKs, CLIs and MCP servers), pigeonhole (filed white field; inbox, ticket and CRM triage), roster (faint square grid; platforms sold through who uses them), ora (warm coal; machine verdicts, audits and protocol traces), tenor (warm management paper; agent operations and autonomous back office), ledger (cream ruled console; analysts, BI and warehouse agents), paperclip (monochrome coal; orchestrators, schedulers and job runners), awning (white commerce forecourt; storefront, payroll and billing platforms), router (pale console with hairlines; dashboard, billing and inventory surfaces), daylight (bright portal with one shadow; client onboarding and service portals), notation (restrained technical hairlines; OSS, docs and developer workspaces), almanac (oatmeal editorial paper; manifestos and category-defining pages), vitrine (serif trust display; evaluated B2B, security and compliance), proscenium (white demo sequence; product tours and launch pages), bulletin (cheerful outlined bands; broad SMB and agency platforms). Ships the sheleg-design skill, the architecture reference, the motion doctrine, the Figma and Claude Design bridges, AI-surface patterns, the packs' token CSS, and the /sheleg-design command.",
6
- "version": "1.66.4",
6
+ "version": "1.66.6",
7
7
  "author": {
8
8
  "name": "ssheleg",
9
9
  "url": "https://x.com/sshlg93"
@@ -19,7 +19,7 @@ Read this when starting the build, and again before calling a surface done.
19
19
  the site-wide design tokens (color, type, spacing, components). For a new
20
20
  design or a redesign of a brand or flagship surface, the reference sweep has
21
21
  already run by this point — before any layout exists to defend — through
22
- whichever reference servers the session exposes (Refero, Lazyweb, Mobbin
22
+ whichever reference servers the session exposes (Refero, and Mobbin
23
23
  where authorized), and its result, an empty one included, is in the director
24
24
  record's `References` field ([`CREATIVE_DIRECTOR.md`](./CREATIVE_DIRECTOR.md),
25
25
  Act 2).
@@ -1,9 +1,9 @@
1
1
  # The component layer — how a kit composes with a pack
2
2
 
3
3
  Split out of `SKILL.md` on 2026-09-10, when its body measured 5703 tokens against the
4
- house budget of 5000. The DECISION stays in `SKILL.md` — the default is `shadcn/ui`,
5
- asked once per project, recorded where the pack is recorded, and it does not apply to a
6
- cinematic scroll surface. What moved here is the MECHANICS: why a kit composes with a
4
+ house budget of 5000. The DECISION stays in `SKILL.md`: the default is `shadcn/ui`,
5
+ recorded once per project where the pack is recorded (asked only when an existing component
6
+ layer conflicts), and it does not apply to a cinematic scroll surface. What moved here is the MECHANICS: why a kit composes with a
7
7
  token layer rather than competing with it, the two vocabularies that must be mapped, the
8
8
  six axes of the adapter contract, and who does the work.
9
9
 
@@ -15,15 +15,74 @@ the middle three worth doing.
15
15
 
16
16
  ## Contents
17
17
 
18
+ - Focused visual work
18
19
  - Act 1 — The brief, and the sentence that could falsify it
19
20
  - Act 2 — Cast the tools, by measuring rather than recalling
20
21
  - Act 3 — Fork, but only when the fork is real — and write the rubric first
21
22
  - Act 4 — Judge, then graft, then record what the cast produced
22
23
  - Act 5 — Validate: alignment first, then quality
23
24
  - First time right — decisions before pixels
24
- - Focused visual work
25
25
  - The record the director owes
26
26
 
27
+ ## Focused visual work
28
+
29
+ An update or a bounded fix (one token, one screen's type or spacing, a list of
30
+ known defects) needs only this section: reuse the existing brief and select the
31
+ relevant lens below. New work and redesigns go on to Act 1. These lenses are
32
+ modes inside this skill, not new global routers or an external command
33
+ dependency. Do not repeat direction selection for an already chosen direction,
34
+ or run every lens for a single local correction.
35
+
36
+ | Lens | Question and concrete work | Evidence to keep |
37
+ |---|---|---|
38
+ | **Type** | List the text roles and the faces/weights actually in use. Improve distinctions between those roles through size, weight, spacing and contrast. Keep a single family when it does the job. Check long labels, the target languages, zoom and fallback fonts. | Representative rendered roles; loaded/fallback font status; the specific reading or hierarchy defect fixed. A font name or a universal line-length target is not a quality verdict. |
39
+ | **Composition** | Establish the leading element, supporting elements and meaningful groups. Adjust proximity, rhythm and density to the task. Inspect the relevant supported sizes and content extremes; preserve a sensible reading and focus order. | Before/after of the same state and content; evidence of wrapping, overflow and order. Optical adjustments follow inspection rather than guessed pixel offsets. |
40
+ | **State adaptation** | Walk empty, long-content, error, pending, keyboard and recovery states relevant to the scenario. Separate browser prototype behavior from native platform expectations. | State/viewport identifiers, declared versus walked coverage and explicit simulations. No invented latency or claim that a browser preview proves native behavior. |
41
+ | **Finish** | Revisit the already identified defects and check what the changes introduced. Keep useful details that were already working. Stop when no actionable finding remains or the agreed budget is reached. | Each previous finding is resolved, partial or unresolved, with current captures and any regressions. The number of edits is not a success metric. |
42
+
43
+ At the copy/design seam, use the [rendered-copy review](VISUAL_REVIEW.md#rendered-copy)
44
+ for visible text roles, project policy, counterexamples and coverage receipts.
45
+
46
+ Before visual review, check that each capture belongs to the expected build,
47
+ screen, state, viewport and content. A stale screenshot or one taken before an
48
+ overlay opened cannot prove that overlay correct. If the capture cannot be
49
+ verified, mark the visual claim unverified and continue only the checks the
50
+ available evidence supports.
51
+
52
+ Judge the render against the brief before reading an explanation of how hard it
53
+ was to build. Separate observed defects from preferences and hypotheses. A fresh
54
+ reviewer is useful when available and authorized; an explicit separate review
55
+ pass by one agent is a valid fallback. Neither requires an external detector,
56
+ binary, service, API key, font download or new framework. Reuse local assets and
57
+ existing inspection tools; missing inspection remains NOT_RUN.
58
+
59
+ These local procedures were informed by a comparison of Impeccable's
60
+ [type](https://github.com/pbakaus/impeccable/blob/12b25ae25848202ce7a9092442198ada5c4fd984/plugin/skills/impeccable/reference/typeset.md),
61
+ [layout](https://github.com/pbakaus/impeccable/blob/12b25ae25848202ce7a9092442198ada5c4fd984/plugin/skills/impeccable/reference/layout.md)
62
+ and [finish-review methods](https://github.com/pbakaus/impeccable/blob/12b25ae25848202ce7a9092442198ada5c4fd984/plugin/skills/impeccable/reference/degraded/finish-reviewer.md) (Apache-2.0). They use this family's contracts; no
63
+ upstream runtime, assets or automatic workflow are included. The links record
64
+ provenance, not instructions to fetch during every run.
65
+
66
+ **Markers, not fixed bans (decided 2026-10-07).** This section used to refuse
67
+ impeccable's fixed style bans outright, and the refusal was right about one
68
+ thing: a ban that cannot lose an argument contradicts *an explicit user
69
+ requirement beats the baseline* ([`SKILL.md`](./SKILL.md), Calibration). What it
70
+ left was no floor at all, and generated defaults kept shipping through reviews
71
+ that had nothing to point at. The floor is now
72
+ [`SLOP_MARKERS.md`](./SLOP_MARKERS.md): each pattern is a marker with an id, a
73
+ severity, the reason it reads as a default, what to do instead and **the
74
+ exception under which it is not slop**. It holds with `Style pack: none`; a pack
75
+ may tighten it and may not loosen it; a brief that explicitly asks for a marked
76
+ pattern wins, once the record below carries the requirement on its `Brief` or
77
+ `Open` line with the marker id. The machine half runs as
78
+ `npx sheleg-design-skill --lint <dir>` and blocks only on S1; the `review` rows
79
+ belong to the render critique in Act 5. Material from impeccable and taste-skill
80
+ enters the catalogue paraphrased, with provenance rows, never as copied rule
81
+ text, and impeccable's detector stays an optional cross-check that is `NOT_RUN`
82
+ when absent.
83
+
84
+ ---
85
+
27
86
  ## Act 1 — The brief, and the sentence that could falsify it
28
87
 
29
88
  **Nothing is designed until this exists in writing.** Four lines, and the fourth
@@ -120,7 +179,7 @@ measured** rather than presenting it as though it had been.
120
179
  brand or flagship surface.** Before `style`, look at what has shipped: at least
121
180
  five real products, each with what this surface takes from it — a structural
122
181
  trait, never its palette. Use whichever reference servers the session actually
123
- exposes — Refero, Lazyweb, and Mobbin where the account is authorized — and,
182
+ exposes (Refero, and Mobbin where the account is authorized) and,
124
183
  with none of them present, shipped products found on the open web, saying the
125
184
  servers were absent. Structure crosses and identity does not
126
185
  ([`DESIGN_SYNC_BRIDGE.md`](./DESIGN_SYNC_BRIDGE.md) §4). **An empty sweep is a
@@ -439,63 +498,6 @@ So the order of work for a new surface is fixed:
439
498
  not score it — a separate agent or a separate pass does
440
499
  ([`ADA_RUBRIC.md`](./ADA_RUBRIC.md), the judge).
441
500
 
442
- ---
443
-
444
- ## Focused visual work
445
-
446
- For a bounded request, reuse the existing brief and select the relevant lens
447
- below. These are modes inside this skill, not new global routers or an external
448
- command dependency. Do not repeat direction selection for an already chosen
449
- direction, or run every lens for a single local correction.
450
-
451
- | Lens | Question and concrete work | Evidence to keep |
452
- |---|---|---|
453
- | **Type** | List the text roles and the faces/weights actually in use. Improve distinctions between those roles through size, weight, spacing and contrast. Keep a single family when it does the job. Check long labels, the target languages, zoom and fallback fonts. | Representative rendered roles; loaded/fallback font status; the specific reading or hierarchy defect fixed. A font name or a universal line-length target is not a quality verdict. |
454
- | **Composition** | Establish the leading element, supporting elements and meaningful groups. Adjust proximity, rhythm and density to the task. Inspect the relevant supported sizes and content extremes; preserve a sensible reading and focus order. | Before/after of the same state and content; evidence of wrapping, overflow and order. Optical adjustments follow inspection rather than guessed pixel offsets. |
455
- | **State adaptation** | Walk empty, long-content, error, pending, keyboard and recovery states relevant to the scenario. Separate browser prototype behavior from native platform expectations. | State/viewport identifiers, declared versus walked coverage and explicit simulations. No invented latency or claim that a browser preview proves native behavior. |
456
- | **Finish** | Revisit the already identified defects and check what the changes introduced. Keep useful details that were already working. Stop when no actionable finding remains or the agreed budget is reached. | Each previous finding is resolved, partial or unresolved, with current captures and any regressions. The number of edits is not a success metric. |
457
-
458
- At the copy/design seam, use the [rendered-copy review](VISUAL_REVIEW.md#rendered-copy)
459
- for visible text roles, project policy, counterexamples and coverage receipts.
460
-
461
- Before visual review, check that each capture belongs to the expected build,
462
- screen, state, viewport and content. A stale screenshot or one taken before an
463
- overlay opened cannot prove that overlay correct. If the capture cannot be
464
- verified, mark the visual claim unverified and continue only the checks the
465
- available evidence supports.
466
-
467
- Judge the render against the brief before reading an explanation of how hard it
468
- was to build. Separate observed defects from preferences and hypotheses. A fresh
469
- reviewer is useful when available and authorized; an explicit separate review
470
- pass by one agent is a valid fallback. Neither requires an external detector,
471
- binary, service, API key, font download or new framework. Reuse local assets and
472
- existing inspection tools; missing inspection remains NOT_RUN.
473
-
474
- These local procedures were informed by a comparison of Impeccable's
475
- [type](https://github.com/pbakaus/impeccable/blob/12b25ae25848202ce7a9092442198ada5c4fd984/plugin/skills/impeccable/reference/typeset.md),
476
- [layout](https://github.com/pbakaus/impeccable/blob/12b25ae25848202ce7a9092442198ada5c4fd984/plugin/skills/impeccable/reference/layout.md)
477
- and [finish-review methods](https://github.com/pbakaus/impeccable/blob/12b25ae25848202ce7a9092442198ada5c4fd984/plugin/skills/impeccable/reference/degraded/finish-reviewer.md) (Apache-2.0). They use this family's contracts; no
478
- upstream runtime, assets or automatic workflow are included. The links record
479
- provenance, not instructions to fetch during every run.
480
-
481
- **Markers, not fixed bans (decided 2026-10-07).** This section used to refuse
482
- impeccable's fixed style bans outright, and the refusal was right about one
483
- thing: a ban that cannot lose an argument contradicts *an explicit user
484
- requirement beats the baseline* ([`SKILL.md`](./SKILL.md), Calibration). What it
485
- left was no floor at all, and generated defaults kept shipping through reviews
486
- that had nothing to point at. The floor is now
487
- [`SLOP_MARKERS.md`](./SLOP_MARKERS.md): each pattern is a marker with an id, a
488
- severity, the reason it reads as a default, what to do instead and **the
489
- exception under which it is not slop**. It holds with `Style pack: none`; a pack
490
- may tighten it and may not loosen it; a brief that explicitly asks for a marked
491
- pattern wins, once the record below carries the requirement on its `Brief` or
492
- `Open` line with the marker id. The machine half runs as
493
- `npx sheleg-design-skill --lint <dir>` and blocks only on S1; the `review` rows
494
- belong to the render critique above. Material from impeccable and taste-skill
495
- enters the catalogue paraphrased, with provenance rows, never as copied rule
496
- text, and impeccable's detector stays an optional cross-check that is `NOT_RUN`
497
- when absent.
498
-
499
501
  ## The record the director owes
500
502
 
501
503
  One record, every time, whether the work took ten minutes or a day — a file in
@@ -20,7 +20,7 @@ agent that does not have it — the pack stands on its own and nothing below app
20
20
  - 1. What crosses, and in what shape
21
21
  - 2. Style packs — the pack is the source of truth
22
22
  - 3. Figma — one border at a time
23
- - 4. Reference sweeps (Lazyweb, Mobbin, Refero) — layout crosses, identity does not
23
+ - 4. Reference sweeps (Mobbin, Refero): layout crosses, identity does not
24
24
  - 5. Live-site extraction — the pack first, the sync second
25
25
  - 6. What cannot cross
26
26
  - 7. Round-trip discipline
@@ -121,13 +121,12 @@ Practically: a value only ever enters the system **at the pack**. Figma variable
121
121
  Claude Design components are both *outputs* of it. When they disagree, the pack wins
122
122
  and the other side is stale.
123
123
 
124
- ## 4. Reference sweeps (Lazyweb, Mobbin, Refero) — layout crosses, identity does not
124
+ ## 4. Reference sweeps (Mobbin, Refero): layout crosses, identity does not
125
125
 
126
126
  A reference sweep answers *what a good version of this screen contains* — sections,
127
127
  hierarchy, content order. It never answers what it looks like.
128
128
 
129
- Three servers can fill the slot and they are not interchangeable. **Lazyweb** is
130
- web-product screens and growth mechanics. **Mobbin** searches screens, **flows** and
129
+ Two servers can fill the slot and they are not interchangeable. **Mobbin** searches screens, **flows** and
131
130
  web **sections** (hero, pricing, footer), strongest on native iOS and also carrying
132
131
  web; its flows come back as evenly-spaced *preview images* per step, and its own tool
133
132
  description says to look at them rather than trust the metadata. **Refero** searches
@@ -135,8 +134,8 @@ screens, returns visually and functionally *similar* screens for one you already
135
134
  have, and returns flows as *structure* — a goal, an action and a system response per
136
135
  step.
137
136
 
138
- **That is the split worth knowing: two of the three answer "flow", and they answer it
139
- in different media.** Mobbin shows you what each step looked like; Refero tells you
137
+ **That is the split worth knowing: both answer "flow", and they answer it in
138
+ different media.** Mobbin shows you what each step looked like; Refero tells you
140
139
  what each step did. Drawing a diagram with decision points and recovery paths reads
141
140
  Refero; judging whether a step actually works on a phone looks at Mobbin. Sweep
142
141
  whichever are present; with more than one, sweep them all and say which answered
@@ -204,7 +204,7 @@ having two sources instead of one.
204
204
 
205
205
  | Server | Tools | Best at |
206
206
  |---|---|---|
207
- | **Lazyweb** | `mcp__lazyweb__*` | web product screens, flows, paywalls, growth mechanics |
207
+ | **Refero** | `mcp__refero__*` | screens and flows on web and mobile; similar screens for one you already have; flows as structure (goal, action and system response per step) |
208
208
  | **Mobbin** | `mcp__mobbin__*` | shipped app screens and flows by category — strongest on native iOS and Android, and it carries web products too, so it is worth a sweep on a website as well |
209
209
 
210
210
  **Gate on the tools, never on the config.** A server can be registered and still
@@ -4,7 +4,7 @@ description: 'Use when deciding how a surface looks or moves: landing, particle/
4
4
  license: MIT
5
5
  compatibility: Kit/lane commands need node>=16, npx, network. Optional siblings — dataviz, shadcn, migrate-radix-to-base. Each has an in-text fallback when absent or unreachable.
6
6
  metadata:
7
- version: 1.66.4
7
+ version: 1.66.6
8
8
  ---
9
9
 
10
10
  # SHELEG Design
@@ -18,7 +18,9 @@ read it per frame and react in their own language. Nothing crossfades — things
18
18
  *redeploy*. Every layer degrades to a calm static state.
19
19
 
20
20
  **READ FIRST, BEFORE ANY OF THE CRAFT BELOW:**
21
- [`CREATIVE_DIRECTOR.md`](./CREATIVE_DIRECTOR.md) — the decision layer: what the
21
+ [`CREATIVE_DIRECTOR.md`](./CREATIVE_DIRECTOR.md). An update or a bounded fix reads
22
+ only its first section, [Focused visual work](./CREATIVE_DIRECTOR.md#focused-visual-work).
23
+ New work reads it whole: the decision layer, what the
22
24
  surface is for and **what would prove it failed**; the lane and the mode (a
23
25
  redesign starts by measuring what it replaces, an update changes a token); which
24
26
  tools to cast, measured with `npx sshlg-skills pack design --lane <lane>`; when to
@@ -58,7 +60,7 @@ cannot produce a conformance row — a receipt where it did is refused, because
58
60
 
59
61
  ## Triggers
60
62
 
61
- "build a landing page" / "сделай лендинг", "design a landing" / "дизайн лендинга", "mobile screen" / "мобильный экран", "dashboard style" / "стиль дашборда, сделай дашборд, админка", "scroll animation" / "скролл-анимация, анимации", "design tokens, style pack" / "дизайн-токены", "light/dark theme" / "светлая/тёмная тема", "figma variables" / "переменные фигмы, фигма в код, макеты", "palette, colors" / "палитра, цвета", "typography, font" / "типографика, шрифт", "how it looks, make it prettier" / "выглядит, красиво, красивее", "visual reference" / "визуальные референсы", "investor deck as a web page" / "веб-презентация", "redesign" / "редизайн, свёрстай, вёрстка".
63
+ "build a landing page" / "сделай лендинг", "design a landing" / "дизайн лендинга", "mobile screen" / "мобильный экран", "dashboard style" / "стиль дашборда, сделай дашборд, админка", "scroll animation" / "скролл-анимация, анимации", "design tokens, theme tokens, style pack" / "дизайн-токены, токены темы", "light/dark theme" / "светлая/тёмная тема, тёмная тема, тёмную тему", "figma variables" / "переменные фигмы, фигма в код, макеты", "palette, colors, contrast" / "палитра, цвета, контрастность", "typography, font" / "типографика, шрифт", "how it looks, make it prettier" / "выглядит, красиво, красивее", "visual reference" / "визуальные референсы", "investor deck as a web page" / "веб-презентация", "redesign" / "редизайн, переделай дизайн, свёрстай, вёрстка".
62
64
 
63
65
  ## When to Use
64
66
 
@@ -109,7 +111,7 @@ npx sheleg-design-skill --kit <pack>
109
111
 
110
112
  The generated `src/styles.css` is authoritative for states a core pack leaves
111
113
  open. A widened pack's kit and `## Components` section must agree. Where `npx`
112
- is unreachable — no node, no registry — nothing is blocked: the chosen pack and
114
+ is unreachable, nothing is blocked: the chosen pack and
113
115
  its token layer in `styles/tokens/<pack>.css` still carry the rules by reading,
114
116
  but the component states only a kit materializes are then **unverified** — say
115
117
  so, rather than presenting a read-off state as a kit-checked one. To author a
@@ -258,8 +260,7 @@ populated page yet, the choice is premature — build the page in the pack the
258
260
  brief's register points at, and revisit once there is something real to switch.
259
261
  If the brief gives no signal at all, the defaults are **`workbench`** for product
260
262
  UI and **`showroom`** for a marketing page: both are quiet enough that switching
261
- away later costs layout, not identity. (This used to say "the default pack" and
262
- name none, which is not a fallback.)
263
+ away later costs layout, not identity.
263
264
 
264
265
  **The mounting rides the semantic token contract.** Token names are not
265
266
  uniform across the packs, so the comparison never swaps raw CSS: each pack's
@@ -293,13 +294,11 @@ A style pack is a **token layer and a set of rules**. It does not ship a button.
293
294
  Every product UI therefore needs a second decision the packs deliberately do not
294
295
  make: which component kit draws the controls.
295
296
 
296
- **The default is `shadcn/ui`, and the question is asked once per project.** Ask it
297
- the way `ux-foundation` asks the Figma question — once, at the point design work
298
- starts, never per screen — and record the answer wherever the style pack is
299
- recorded: with `super-ux` installed that is the screens record's *Design system*
300
- line, and without it, whatever file names the pack. Default **yes**; a project
301
- that already has a component layer has already answered, and migrating one on
302
- taste is not a design decision.
297
+ **The default is `shadcn/ui`: record it once per project and say so, do not ask.**
298
+ Record it where the style pack is recorded (with `super-ux`, the screens record's
299
+ *Design system* line; without it, whatever file names the pack). Ask only when an
300
+ existing component layer conflicts: that project has already answered, and migrating
301
+ one on taste is not a design decision.
303
302
 
304
303
  **It composes rather than competes**, and the mapping is where it goes wrong: an
305
304
  undefined custom property falls back silently instead of erroring, and the token remap
@@ -317,7 +316,7 @@ how a hero ends up looking like a settings screen.
317
316
 
318
317
  Touching a Figma file — publishing the pack as variables, or building from a
319
318
  design — starts at [`FIGMA_BRIDGE.md`](./FIGMA_BRIDGE.md). It owns the contract
320
- and the one rule to know before opening the tool: **a frame is read
319
+ and its first rule: **a frame is read
321
320
  by designers and stakeholders, so drawing in one is publishing**, and creating a
322
321
  file needs a named destination rather than a guess.
323
322
 
@@ -327,7 +326,7 @@ Where `/design-sync` is available, a pack can be pushed to claude.ai/design so t
327
326
  design agent builds from **this pack's** tokens, not its own defaults. The contract
328
327
  and its gaps: [`DESIGN_SYNC_BRIDGE.md`](./DESIGN_SYNC_BRIDGE.md).
329
328
 
330
- ## Real-world references (Refero, Lazyweb, Mobbin)
329
+ ## Real-world references (Refero, Mobbin)
331
330
 
332
331
  A pack fixes *how it looks*, not what a good screen contains. A new design or brand
333
332
  redesign sweeps real products first — **content and structure**, never a visual to
@@ -93,9 +93,9 @@ narrowest of the three big engines, so a pass there is the stricter claim.
93
93
  | an automation library (browser-use and the like) | its WebKit/Safari backend, where it has one |
94
94
 
95
95
  **Chromium is a stated fallback, never the default.** Use it only when the task
96
- needs Chrome itself — a Chrome-only API, Lighthouse, a CDP trace, the
97
- chrome-devtools MCP — and say which and why in the record. Prefer the Chrome Dev
98
- build for that run. When you launch your own Chrome or Chromium for a UI run,
96
+ needs Chrome itself (a Chrome-only API, a Lighthouse audit through the
97
+ Lighthouse CLI, a CDP trace) and say which and why in the record. Run it in the Chrome Dev
98
+ build or in your own headless Chromium. When you launch your own Chrome or Chromium for a UI run,
99
99
  give it a temporary profile (never the user's), record what you started, and
100
100
  close every process you started before you finish.
101
101
 
@@ -4,7 +4,7 @@ surface_class: <flagship | product | internal | ad>
4
4
  revision: <commit or date this record describes>
5
5
  ---
6
6
 
7
- # Director record — <surface>
7
+ # Director record: <surface>
8
8
 
9
9
  <!--
10
10
  HOW TO USE THIS TEMPLATE
@@ -46,8 +46,8 @@ marker id here or under Open.
46
46
 
47
47
  <!--
48
48
  One word first: new | redesign | update | audit | declined
49
- then where you entered, e.g. "redesign — entered at verify + visual-qa".
50
- Declined: "declined — <the reason, in a sentence>".
49
+ then where you entered, e.g. "redesign: entered at verify + visual-qa".
50
+ Declined: "declined: <the reason, in a sentence>".
51
51
  -->
52
52
 
53
53
  ## Taste
@@ -60,27 +60,27 @@ decide nothing. Name materials, references, type, rhythm, the subject's own
60
60
  vernacular.
61
61
  Take: e.g. "ledger-paper rhythm from the subject; tabular figures; one warm accent"
62
62
  Ban: e.g. "violet gradients; three equal feature cards; stock 3D blobs"
63
- Anti-patterns: optional — SLOP_MARKERS ids this surface is most at risk of
64
- Dials: optional — VARIANCE / MOTION / DENSITY with one anchor each (SKILL.md, Calibration)
65
- Profile: optional — the standing taste profile this was read from, if any
63
+ Anti-patterns: (optional) SLOP_MARKERS ids this surface is most at risk of
64
+ Dials: (optional) VARIANCE / MOTION / DENSITY with one anchor each (SKILL.md, Calibration)
65
+ Profile: (optional) the standing taste profile this was read from, if any
66
66
  -->
67
67
 
68
68
  ## References
69
69
 
70
70
  <!--
71
- At least five real products, each with its URL and what we take from it — the
71
+ At least five real products, each with its URL and what we take from it: the
72
72
  structural trait, not its palette. One line each:
73
- - <Product> — https://… — take: <the trait>
73
+ - <Product>, https://…, take: <the trait>
74
74
  Or, when the sweep came back empty, say so and say where you looked:
75
75
  none found
76
- Searched: <the reference servers and queries, e.g. Refero "pricing table", Lazyweb "usage-based pricing">
76
+ Searched: <the reference servers and queries, e.g. Refero "pricing table", Mobbin "usage-based pricing">
77
77
  Mandatory for new design and redesign of a brand or flagship surface.
78
78
  -->
79
79
 
80
80
  ## Cast
81
81
 
82
82
  <!--
83
- Each skill or tool, what it is for, and the lane it serves — one line each.
83
+ Each skill or tool, what it is for, and the lane it serves, one line each.
84
84
  Say whether the roster was measured (`npx sshlg-skills pack design`) or read
85
85
  from the harness list.
86
86
  -->
@@ -89,11 +89,11 @@ from the harness list.
89
89
 
90
90
  <!--
91
91
  Start with yes or no.
92
- no — <why: the brief already determines the answer, a locked system, a fix>
93
- yes — rubric below written before either direction; then
92
+ no: <why, e.g. the brief already determines the answer, a locked system, a fix>
93
+ yes: rubric below written before either direction; then
94
94
  A: <cast / direction>
95
95
  B: <cast / direction>
96
- Winner: <A or B> — grafted from the other: <the trait>
96
+ Winner: <A or B>, grafted from the other: <the trait>
97
97
  -->
98
98
 
99
99
  ## Rubric
@@ -119,9 +119,9 @@ A clean render is a valid result: write "clean render" and stop. No defect quota
119
119
  <!--
120
120
  The `--lint` run this record stands on: the command, the revision it ran at and
121
121
  the counts by severity, copied from its summary line.
122
- npx sheleg-design-skill --lint src @ 3f2a9c1 — 0 S1, 2 S2, 1 S3, 1 waived
122
+ npx sheleg-design-skill --lint src @ 3f2a9c1: 0 S1, 2 S2, 1 S3, 1 waived
123
123
  Each waived marker: its id and the Brief or Open line that earned it. An S1
124
- finding blocks — fix it, or waive it in source with `sheleg-lint-allow: V0NN <reason>`.
124
+ finding blocks: fix it, or waive it in source with `sheleg-lint-allow: V0NN <reason>`.
125
125
  -->
126
126
 
127
127
  ## Alignment
@@ -141,7 +141,7 @@ ratios, type sizes counted, durations measured, NOT_APPLICABLE with its reason.
141
141
  ## Signature
142
142
 
143
143
  <!--
144
- Exactly one signature moment for the flow — what it is, where it happens, and
144
+ Exactly one signature moment for the flow: what it is, where it happens, and
145
145
  why it comes from the subject rather than from a catalogue.
146
146
  What: <the moment>
147
147
  Where: <screen / state / transition>
@@ -151,15 +151,15 @@ why it comes from the subject rather than from a catalogue.
151
151
  ## Surfaces
152
152
 
153
153
  <!--
154
- Where the job also lives outside the app or page — widget, Live Activity, App
155
- Intent, Watch, PWA install, share card — or why none is needed.
154
+ Where the job also lives outside the app or page (widget, Live Activity, App
155
+ Intent, Watch, PWA install, share card), or why none is needed.
156
156
  -->
157
157
 
158
158
  ## Haptics
159
159
 
160
160
  <!--
161
161
  Native only: a table of event → pattern, each pattern used for its documented
162
- meaning and switchable off. Otherwise "n/a — <reason, e.g. web surface>".
162
+ meaning and switchable off. Otherwise "n/a: <reason, e.g. web surface>".
163
163
  -->
164
164
 
165
165
  ## ADA
@@ -171,11 +171,11 @@ are NOT_ASSESSED until a labelled set exists.
171
171
  Profile: flagship
172
172
  Targets: Interaction, Inclusivity
173
173
  G: 15 PASS / 0 FAIL / 2 N/A
174
- J: NOT_ASSESSED — no labelled set yet
174
+ J: NOT_ASSESSED (no labelled set yet)
175
175
  H: contact sheet reviewed by <who>, <date>
176
176
  FAIL triples: region → defect → change
177
177
  ad: the same lines on the ad profile, plus "Safe zones: <placement and result>".
178
- Any other class: "n/a — <reason>".
178
+ Any other class: "n/a: <reason>".
179
179
  -->
180
180
 
181
181
  ## Open