Add design intelligence loop

README.md

@@ -117,6 +117,7 @@ Community skills for UI craft. Claude consults these automatically during UI wor
 | Skill | What it does | Source |
 |-------|-------------|--------|
 | `emil-design-eng/` | Animation craft — easing curves, spring physics, polish | [emilkowalski/skill](https://github.com/emilkowalski/skill) |
+| `review-animations/` | Hard-nosed motion review — flags easing, duration, origin, interruptibility, performance, and a11y | [emilkowalski/skill](https://github.com/emilkowalski/skill) |
 | `make-interfaces-feel-better/` | 16 UI detail principles (optical alignment, shadows, stagger, hit areas) | [jakubkrehel](https://github.com/jakubkrehel/make-interfaces-feel-better) |
 | `taste-skill-suite/` | Anti-LLM-bias, typography calibration, color correction, layout diversification | [Leonxlnx/taste-skill](https://github.com/Leonxlnx/taste-skill) |
 | `impeccable/` | 17 design commands — dark mode mastery, polish, audit | [impeccable.style](https://impeccable.style) |
@@ -140,6 +141,7 @@ Run these after writing code to clean up before shipping.
 | `tdd/` | Test-driven development — red/green/refactor loop |
 | `fix-sentry-issues/` | Discover, triage, and fix production errors from Sentry |
 | `reclaude/` | Refactor bloated CLAUDE.md files using progressive disclosure |
+| `gemini-review/` | Read-only Google AI / Antigravity second-model review over a git diff |
 
 ---
 
@@ -186,6 +188,7 @@ Entry point: `skills/studio/_graph/studio.moc.md`
 │   ├── studio/                # Studio Skill Graph (90+ files)
 │   │   └── _graph/
 │   ├── emil-design-eng/
+│   ├── review-animations/
 │   ├── make-interfaces-feel-better/
 │   ├── taste-skill-suite/
 │   ├── impeccable/
@@ -201,7 +204,8 @@ Entry point: `skills/studio/_graph/studio.moc.md`
 │   ├── knip/
 │   ├── tdd/
 │   ├── fix-sentry-issues/
-│   └── reclaude/
+│   ├── reclaude/
+│   └── gemini-review/
 └── docs/
     └── setup-guide.md
 ```

docs/design-intelligence-loop.md

@@ -0,0 +1,105 @@
+# Design Intelligence Loop
+
+## Recommendation
+
+Rebuild the old OpenClaw weekly design digest as a Codex-owned automation that uses Stack as the durable skill source.
+
+Ownership split:
+- Stack owns the design-intelligence skill, graph references, templates, and promotion rules.
+- Codex owns runner scripts, eval bundles, install wiring, and eventual recurring automation.
+- Hermes and GBrain remain source/memory substrates: bookmark snapshots, imports, and backfill staging.
+- OpenClaw is historical evidence only; do not depend on its stale cron state for the rebuild.
+
+## Weekly Inputs
+
+Every weekly run must read:
+- Arc sidebar/bookmark state from `~/Library/Application Support/Arc/StorableSidebar.json`.
+- Arc History as a recency supplement only.
+- Field Theory/X bookmarks from `~/.ft-bookmarks/bookmarks.db`, JSONL, Markdown exports, and media metadata.
+- GBrain imported bookmark roots, especially deltas where Field Theory is ahead.
+- Curated web sources: Brian Lovin, Design Spells, Handheld Design, Featured Mobile.
+
+Intake is read-only. The run may write a source manifest, digest, and promotion packet, but it must not mutate Arc, Field Theory, X, GBrain roots, Vault, or Studio skills during collection.
+
+## Weekly Flow
+
+1. Build a source manifest with read status, candidate counts, and extraction failures.
+2. Synthesize the weekly digest using the preserved Output A/B/C structure.
+3. Produce a promotion packet for Zettelkasten and Studio/CDO skill candidates.
+4. Run the Codex design eval gate for any candidate skill update.
+5. Promote only through a reviewable Stack change after eval evidence exists.
+
+## Historical Backfill
+
+Initial backfill window: `2026-04-01` through the current run date.
+
+Chunk by week. Each chunk should produce:
+- `source-manifest.json`
+- `weekly-design-digest-YYYY-MM-DD.md`
+- `promotion-packet.md`
+
+Use Hermes tmp or another review staging area for generated packets until promotion is approved.
+
+## Commands
+
+Read-only source manifest:
+
+```bash
+/Users/maroun/Codex/scripts/design-intelligence-source-manifest.sh \
+  --since 2026-04-01 \
+  --until 2026-06-15 \
+  --out /Users/maroun/hermes/tmp/design-intelligence-backfill/source-manifest.json
+```
+
+Historical backfill packet:
+
+```bash
+/Users/maroun/Codex/scripts/run-design-intelligence-backfill.sh \
+  --since 2026-04-01 \
+  --until 2026-06-15 \
+  --out-root /Users/maroun/hermes/tmp/design-intelligence-backfill
+```
+
+Eval bundle dry run:
+
+```bash
+cd /Users/maroun/Codex
+DRY_RUN=1 FINAL_HTML_ONLY=1 IGNORE_USER_CONFIG=1 \
+  scripts/evaluate-design-skills.sh run-one design-intelligence-v1 001-operational-dashboard
+```
+
+Promotion gate:
+
+```bash
+cd /Users/maroun/Codex
+EVAL_BUNDLES="codex-current design-intelligence-v1" \
+EVAL_PROMPTS="001-operational-dashboard 002-productivity-app 003-landing-page-with-assets 004-existing-page-redesign 005-data-workflow" \
+FINAL_HTML_ONLY=1 IGNORE_USER_CONFIG=1 RUN_ID=design-intelligence-gate \
+  scripts/evaluate-design-skills.sh run-matrix
+```
+
+## Automation Gate
+
+Do not enable the recurring weekly automation until the operator approves the live job.
+
+Recommended live cadence:
+- Saturday morning: source manifest and digest.
+- Sunday evening: taste-compounding promotion packet after digest review.
+
+## Latest Candidate Eval
+
+Latest verified gate: `2026-06-17`.
+
+Baseline run:
+`/Users/maroun/Codex/eval/design-skills/runs/design-intelligence-gate-20260615T223548Z`
+
+Candidate run:
+`/Users/maroun/Codex/eval/design-skills/runs/design-intelligence-gate-final2-20260615T232621Z`
+
+Result: `design-intelligence-v1` beat `codex-current` on 5 of 5 fixtures with no candidate hard fails.
+
+Evidence saved with the candidate run:
+- `report.md`
+- `viewport-hard-gate.tsv`
+- `screenshots-final/contact-sheet-mobile.jpg`
+- `screenshots-final/contact-sheet-desktop.jpg`

skills/cdo/references/inspiration-sources.md

@@ -9,6 +9,12 @@ Curated by Maroun. These are the galleries, collections, and channels Studio sho
   - Curated by former GitHub/Meta designer. High bar. Real shipping products.
   - Best for: Overall site quality, navigation, layout, polish
 
+- **Recent — Design Inspiration** — https://recent.design/
+  - Daily-updated gallery of strong recent design found around the internet.
+  - Best for: fresh visual references, current product/site patterns, quick taste-mining before a design pass.
+  - Use when Studio/CDO needs contemporary examples without falling into stale Dribbble/Pinterest sludge.
+  - Added from Maroun's design inspo request, 2026-06-21.
+
 ### Micro-Interactions & Delight
 - **Design Spells** — https://www.designspells.com/
   - "Design details that feel like magic" by Chester & Duncan
@@ -25,6 +31,13 @@ Curated by Maroun. These are the galleries, collections, and channels Studio sho
   - Mobile app showcase
   - Best for: Mobile UI patterns, app design trends
 
+### Interaction Design Essays & Playgrounds
+- **Aresluna — Show your hands honor for the strange power they bring you** — https://aresluna.org/show-your-hands-honor/
+  - Marcin Wichary essay on designing finger-friendly interactions, with 38 playgrounds.
+  - Best for: touch ergonomics, motor memory, keyboard/finger interaction details, perceived speed, and explaining why tiny interaction choices matter.
+  - Use when Studio/CDO needs principled backing for mobile gestures, physical affordances, latency polish, or “this detail feels small but changes the whole interaction” arguments.
+  - Added from Maroun's design inspo request, 2026-06-20.
+
 ## Tier 2 — Deep Aesthetic Research
 
 ### Are.na Channels
@@ -60,6 +73,10 @@ Another curated platform — higher signal than Dribbble.
 - **Refero** — https://refero.design/ — Real product screenshots
 - **Product Hunt** — Design quality of new launches
 
+### Product Presentation & Mockup Tooling
+
+- **Ultramock** — https://ultramock.io/ — Turns product screens into premium customizable visuals, videos, and 3D mockups. Use when portfolio/case-study/launch assets need more polish than raw screenshots or stock device frames. Added from Maroun's saved-link request, 2026-06-19.
+
 ## How to Use These
 
 ### During `/tastemining` or `/research`

skills/design-intelligence/SKILL.md

@@ -0,0 +1,79 @@
+---
+name: design-intelligence
+description: "Run the weekly design intelligence loop: scan curated sources, Arc bookmarks, Field Theory/X bookmarks, and GBrain deltas; synthesize a design digest; propose review-gated skill updates; and require eval evidence before promotion."
+---
+
+# Design Intelligence Loop
+
+Use this skill for the rebuilt weekly design digest and taste-compounding workflow.
+
+The loop has two jobs:
+- Find high-signal design inspiration from curated sources and Maroun's saved links.
+- Convert repeated lessons into reviewable Studio/CDO skill updates only after evidence and eval.
+
+## Operating Rule
+
+Run as a read-only intake by default. Do not mutate Arc, Field Theory, X/Twitter, GBrain source roots, Vault, or Studio skill files during source collection.
+
+Promotion is a separate review step.
+
+## Load Order
+
+1. Read `references/source-adapters.md` for the source surfaces and safe access rules.
+2. Read `references/output-contract.md` for digest and backfill deliverables.
+3. Read `references/promotion-rules.md` before proposing any skill update.
+4. For the 2026 OpenClaw outage recovery candidate, read `references/backfill-candidate-guidance-2026-06-15.md`.
+5. Read `eval/checklist.md` before claiming a run is complete.
+6. Use `templates/weekly-digest.md` for the report body.
+
+## Weekly Run
+
+Default cadence: Saturday morning, after bookmark sync and before any taste-compounding promotion.
+
+Use a seven-day window unless the operator passes an explicit range.
+
+Required source lanes:
+- Curated sources: Brian Lovin `/sites` or writing, Design Spells, Handheld Design, Featured Mobile.
+- Arc bookmarks/sidebar: current sidebar plus Arc history as a date supplement.
+- Field Theory/X bookmarks: live Field Theory SQLite/JSONL/Markdown export.
+- GBrain: imported `x-bookmarks` and saved-link roots, especially deltas where Field Theory is ahead.
+
+## Historical Backfill
+
+For the OpenClaw outage recovery, run a catch-up window before resuming normal weekly cadence.
+
+Default catch-up window: `2026-04-01` through the current run date.
+
+Chunk the backfill by week. Each chunk should produce its own source manifest and digest so weak or unreachable sources are visible instead of hidden inside one giant report.
+
+## Outputs
+
+Every run must produce:
+- A source manifest with counts, paths used, fetch/read status, and candidate samples.
+- A weekly design digest with Output A, Output B, and Output C.
+- A promotion packet with proposed Studio/CDO changes, evidence links, and the eval gate required before promotion.
+
+The digest may suggest Zettelkasten notes and skill changes, but it must not directly write them.
+
+## Evaluation Gate
+
+Before a proposed skill update becomes the default runtime behavior:
+- Install or refresh the candidate skill into Codex.
+- Run the design-skill eval matrix against `codex-current`.
+- Promote only when the candidate beats `codex-current` on at least 4 fixtures and has no hard fails.
+
+If the eval harness is unavailable, mark the promotion packet as blocked rather than approving it from taste alone.
+
+## Candidate Guidance From Historical Backfill
+
+Status: candidate from the 2026-04-01..2026-06-15 OpenClaw outage backfill. Use this guidance for `design-intelligence-v1` eval runs; do not treat it as default runtime taste until the eval gate passes and the Stack change is reviewed.
+
+- Ground taste in concrete references before inventing. Use saved references, pattern libraries, side-by-side comparisons, and explicit visual vocabulary as design infrastructure; do not let the model free-associate a generic style when evidence exists.
+- Treat component registries as raw ingredients, not an aesthetic. When using shadcn/ui or a registry source, borrow coverage of states, layout primitives, and component anatomy, then deliberately customize density, tokens, radii, shadows, and interaction states so the result does not read like a default registry demo.
+- Use named, state-driven motion instead of generic decoration. Pick the smallest transition that clarifies a real state change, document its trigger, timing, interruption behavior, and reduced-motion fallback, and avoid `transition-all`, perpetual ambient movement, or motion that hides missing hierarchy.
+- For AI-assisted design or builder workflows, expose the design tree. The UI should show the brief/prompt, generated directions, comparison criteria, selected branch, iteration history, and eval/QA state as first-class product surfaces rather than collapsing the experience into a chat box.
+- Generate design candidates in multiples when exploring taste, compare them side by side against the rubric, then refine the selected branch. Do not promote a single attractive artifact without critique/eval evidence.
+- For operational and data-heavy tools, favor dense comparative workspaces: tables, filters, timelines, side panels, validation states, batch controls, and specific deltas. Do not let portfolio, launch-page, or configurator inspiration override the user's primary workflow.
+- Dense mobile tables must recompose, not widen the page. At narrow viewports, switch to priority columns, stacked row cards, or an internal scroll region whose parent still keeps `documentElement.scrollWidth <= innerWidth`; do not rely on a large table `min-width` that creates page-level horizontal overflow.
+- Before/after redesign artifacts should open on the improved experience. If a legacy "before" view is included for comparison, constrain it inside the viewport and never let the old layout define the page width or primary mobile experience.
+- Mobile hard gates apply to the whole shell, not just content tables. Top-level app wrappers, sidebars, sticky headers, nav rows, filters, panels, charts, and dialogs need `min-width:0`, `max-width:100%`, and wrapping/contained overflow so no child can expand the document past the viewport.

skills/design-intelligence/eval/checklist.md

@@ -0,0 +1,33 @@
+---
+id: design-intelligence.eval-checklist
+name: Design Intelligence Eval Checklist
+description: Completion checklist for weekly digest and skill promotion runs.
+---
+
+# Eval Checklist
+
+## Intake Complete
+
+- [ ] Curated web sources checked or marked unreachable.
+- [ ] Arc sidebar scanned read-only.
+- [ ] Arc history used only as a recency supplement.
+- [ ] Field Theory/X bookmarks scanned from SQLite or JSONL.
+- [ ] GBrain import delta checked.
+- [ ] Source manifest written with counts, paths, and failures.
+
+## Digest Complete
+
+- [ ] Output A includes 3-5 findings with source links and apply-this notes.
+- [ ] Output B includes at most 3 Zettelkasten candidates and uses wikilinks.
+- [ ] Output C includes at most 3 Studio/CDO skill candidates.
+- [ ] No digest claim says a file was saved unless the file exists.
+- [ ] Failures are visible instead of hidden.
+
+## Promotion Complete
+
+- [ ] Each skill candidate has evidence and an idempotency check.
+- [ ] No source corpus or Vault note was mutated during intake.
+- [ ] Candidate changes are reviewable as a patch/PR.
+- [ ] Codex design eval was run or the packet is marked blocked on eval.
+- [ ] Candidate beats `codex-current` on at least 4 fixtures with no hard fails before default promotion.
+