analyzthis_design 2.5.1 → 2.6.0

package/README.md

@@ -4,16 +4,33 @@ A set of AI design personas and a task-first evaluation framework that plugs int
 
 Install once. Run structured UX critiques, multi-phase ideation, and task-grounded screen reviews — directly inside your AI chat. **No external LLM API keys required** for CLI orchestrator runs: **`/devi`** voices each persona from your host IDE (Cursor, Claude, etc.).
 
+## v2.6.0 — Mood boards you decide on, now with scroll films
+
+- **Element collage:** Zara crops 20–24 real UI elements from 6+ live sites into sections (first screen, typography, colour, components, imagery, motion), each with its source page and real colours and fonts.
+- **You decide:** `moodboard show` opens a clickable board on 127.0.0.1. Keep / maybe / reject overrides every persona; rejected tiles never come back; `lock` saves the direction every persona follows.
+- **Scroll films:** `moodboard film --board <id> --url <site>` records the real page scrolling through four depths as a looping motion tile with waypoint stills. Needs the optional `playwright-core` and Google Chrome.
+- **Install fix:** the `mood-board` skill is now copied by `npx analyzthis_design` (it was missing before).
+
+## v2.5.2 — Motion, UX, and visual-design libraries
+
+Personas cite real packages and prefer what is already in the repo:
+
+- **Catalog:** `motion.csv` (150–250ms ease-out, no bounce) and `ui-libraries.csv` (Motion, Shadcn, React Aria, Chakra, CMDK, TanStack, Tremor, Magic UI, and more).
+- **Kavi:** Reads the full `package.json` plus kit folders, writes `Tech/Installed-UI-Libraries.md`, stores `installed_ui_libraries` on the session.
+- **Zara / Arjun / Anuj / Noor / Priya:** One library per job, installed first. Mixing two full kits is a block. Marketing FX stays off daily dashboards.
+- **DesignSpec:** When motion is on, require library, import, timing, easing, and a reduced-motion fallback.
+- **Design reference:** 30 CSV files total.
+
 ## v2.5.1 — Anti-AI-slop & craftsmanship rules
 
-Enriched our personas and design reference data with craftsmanship principles inspired by Impeccable:
+Enriched personas and design reference data with craftsmanship principles inspired by Impeccable:
 
-- **Arjun (Anti-AI-Slop & Craftsmanship):** Flags container nesting syndrome (`card > card > card`), untinted harsh black/gray, gray text on colored badges/headers, and cliché purple/blue AI gradients.
-- **Noor (Distill & Clarify):** Strips away container clutter and enforces outcome-specific CTA labels (e.g. "Create Invoice" vs "Submit").
-- **Anuj (Production Hardening):** Audits high-density layouts for text overflow handling (`truncate`, `line-clamp-2`), extreme data lengths, and complete 5-state form controls.
-- **Zara (Motion Discipline):** Enforces 150–250ms natural ease-out transitions and bans bouncy/elastic animations.
-- **Kavi (Durable Product Truth):** Records audience, purpose, constraints, voice, and evidence separately from surface styling.
-- **Design reference:** 11 new rules added to `ux-guidelines.csv` (rows 100–110).
+- **Arjun:** Flags container nesting (`card > card > card`), untinted harsh black/gray, gray text on colored badges/headers, and cliché purple/blue AI gradients.
+- **Noor:** Strips container clutter and enforces outcome-specific CTA labels (e.g. "Create Invoice" vs "Submit").
+- **Anuj:** Audits overflow handling and complete 5-state form controls.
+- **Zara:** 150–250ms ease-out; bans bouncy/elastic animations.
+- **Kavi:** Records Durable Product Truth (audience, purpose, constraints, voice, evidence).
+- **Design reference:** 11 craftsmanship rules in `ux-guidelines.csv` (rows 100–110).
 
 ## v2.5.0 — personas earn or lose trust
 
@@ -96,7 +113,7 @@ Pass `--global` to opt into the legacy merged behavior (read `config.sources` an
 
 Use `npx analyzthis_design run-unchunked` for the legacy single-pass orchestrator.
 
-**npm:** [analyzthis_design](https://www.npmjs.com/package/analyzthis_design) · **Current version:** 2.5.1 · **Step-by-step guide:** [HOW-TO-USE.md](./HOW-TO-USE.md)
+**npm:** [analyzthis_design](https://www.npmjs.com/package/analyzthis_design) · **Current version:** 2.6.0 · **Step-by-step guide:** [HOW-TO-USE.md](./HOW-TO-USE.md)
 
 ---
 
@@ -367,6 +384,25 @@ npx analyzthis_design run-unchunked --continue --task "Review invoice screen" --
 
 **Chunk model selection:** Ollama auto-discovered → free cloud (Groq/Gemini/OpenRouter free endpoints) → cheap cloud with user keys. The **planner always runs on a frontier/strong model** and never on a cheap model; if no frontier provider is available, it falls back to the host model with a warning.
 
+### Persona model calibration
+
+Model learning is scoped by persona, effort, task type, and output format. API availability failures are tracked separately from answer quality, and `/accept` records the exact run and model that produced the note. Host/Devi responses are labelled `host/unknown` because the package cannot verify the model selected inside an IDE.
+
+```bash
+# Preview the fixed benchmark, eligible models, call count, and worst-case cost
+npx analyzthis_design calibrate --max-usd 5 --max-calls 500
+
+# Apply a qualified local winner after reviewing the preview/results
+npx analyzthis_design calibrate --max-usd 5 --max-calls 500 --judge --apply
+
+# Preview the text-free community payload; private-task results are never shared
+npx analyzthis_design calibrate --share --dry-run
+```
+
+Calibration is preview-only unless `--apply` is present. It uses fixed spending and call limits, format-specific checks, and paired non-inferiority comparisons. Local winners live in `~/.analyzthis_design/calibration.json`; maintainer-reviewed global defaults ship in `agents/model-policy.json`. A changed persona prompt, validator, expired model, or failed capability check automatically invalidates an old winner.
+
+Applying a winner also requires the selected judges to match the benchmark’s 10 maintainer-labelled comparisons at least 70% of the time. Without enough aligned, independent judge evidence, results remain low-confidence and the current model policy is kept.
+
 **Provider resolution order:** explicit `--provider` → config → first available API key → **`host`** (Devi) for unchunked; chunked mode also considers Ollama and free endpoints before paid.
 
 Supported providers: `host` | `anthropic` | `openai` | `google` | `zai` | `ollama` | `groq` | `together` | `openrouter` | `deepseek`
@@ -377,19 +413,19 @@ Provider defaults live in `~/.analyzthis_design/config.json`:
 {
   "orchestrator": {
     "provider": "anthropic",
-    "model": "claude-sonnet-5",
+    "model": "claude-sonnet-4-20250514",
     "mode": "lite",
     "tiers": {
       "structured": { "provider": "openai", "model": "gpt-4o-mini" },
-      "critique":   { "provider": "anthropic", "model": "claude-sonnet-5" },
-      "arbitrate":  { "provider": "anthropic", "model": "claude-sonnet-5" }
+      "critique":   { "provider": "anthropic", "model": "claude-sonnet-4-20250514" },
+      "arbitrate":  { "provider": "anthropic", "model": "claude-sonnet-4-20250514" }
     },
     "max_tokens": { "structured": 900, "critique": 1800, "arbitrate": 1200 }
   },
   "pricing": {
     "glm-4.5-flash":     { "input_per_m": 0,    "output_per_m": 0 },
     "gemini-2.5-flash":  { "input_per_m": 0.30, "output_per_m": 2.50 },
-    "claude-sonnet-5":   { "input_per_m": 2,    "output_per_m": 10 },
+    "claude-sonnet-4-20250514": { "input_per_m": 3, "output_per_m": 15 },
     "gpt-4o":            { "input_per_m": 2.50, "output_per_m": 10 }
   },
   "research": { "provider": "https://example.com/search?q={query}" },
@@ -458,7 +494,7 @@ Classifier rules (first match wins, safety rules before savings rules):
 Default effort matrix (in `agents/chain.json`):
 - **trivial** → `host` / Devi (~600-token cap for objection rounds) — or API model when keys set
 - **standard** → `gemini-2.5-flash` or `gpt-4o-mini`, ~1200-token cap
-- **hard** → `claude-sonnet-5` or `gpt-5`, ~1800-token cap
+- **hard** → `claude-sonnet-4-20250514`, ~1800-token cap
 
 Per-persona `effort_overrides` in each manifest refine this (e.g. Arjun's `trivial` is the color-system-only scoped mode at 700 tokens; his `hard` is the full Honeycomb + Visual Audit at 1800).
 
@@ -477,7 +513,7 @@ Ask → session digest → MoE router (1–2 experts, not 4) → persona cards (
 | **Retrieve-on-demand** | `npx analyzthis_design retrieve --file colors.csv --column "Product Type" --keywords saas` returns only matching rows, pre-formatted for citation — never the whole CSV. |
 | **Multi-file reference packs** | Each persona retrieves from 3-5 CSV files (not 1), pooled into a single ranker call. Arjun gets styles + ux-guidelines + ui-reasoning + charts; Zara gets colors + typography + styles + landing + icons. Same LLM cost as single-file, 3-5x coverage. |
 | **Best For Tags** | `styles.csv` and `typography.csv` have a `Best For Tags` column (semicolon-delimited product-type tokens) for reliable keyword filtering. Previously `Best For` was free-text and 31/84 styles were unfilterable. |
-| **CSV schema + validation** | `skills/design-reference/schema.json` defines all 28 CSV files' headers, filter columns, and cross-file joins. `npm run validate` checks integrity before publish. |
+| **CSV schema + validation** | `skills/design-reference/schema.json` defines all 30 CSV files' headers, filter columns, and cross-file joins. `npm run validate` checks integrity before publish. |
 | **Model tiers** | `structured` steps can run on a cheaper model (e.g. `gpt-4o-mini`); `critique`/`arbitrate` steps use a stronger model. Configurable per tier in `~/.analyzthis_design/config.json`. |
 | **Caching** | `lib/cache.js` caches retrieve results (invalidated automatically when the source CSV changes) and knowledge-bank slices (invalidated on `sync` / `session reset`). |
 | **Cost metrics** | Every `run` records `metrics` (llm_calls, experts_run, estimated tokens, cache_hits) into session state. |
@@ -632,7 +668,7 @@ npx analyzthis_design feedback status
 The `feedback submit` client is a plain HTTPS POST with an `apikey` header — it works with any REST endpoint that accepts anonymous inserts, not only Supabase. A reference schema (with row-level security for insert-only anon access) lives in [`supabase/migrations/001_persona_feedback.sql`](./supabase/migrations/001_persona_feedback.sql) in the repo. That folder is **not** shipped in the npm package, so installing `analyzthis_design` does not pull a Supabase-branded folder into `node_modules`.
 
 1. Stand up any HTTP endpoint that accepts anonymous JSON inserts (Supabase with RLS is one option; a small Cloudflare Worker or a self-hosted Postgres + thin API work too).
-2. If you use the reference schema, run `supabase/migrations/001_persona_feedback.sql` in your SQL editor.
+2. Run `supabase/migrations/001_persona_feedback.sql` for persona feedback and `002_calibration_results.sql` for opt-in calibration aggregates. The second migration is compatible with Neon and was production-verified on 2026-09-17.
 3. Copy the endpoint URL and anon key into `~/.analyzthis_design/config.json` under `"feedback"` (or set env vars `ANALYZTHIS_FEEDBACK_URL` + `ANALYZTHIS_FEEDBACK_ANON_KEY`).
 4. Read submissions from your endpoint's dashboard.
 
@@ -819,8 +855,12 @@ npx analyzthis_design run --task "..." [--budget free|cheap|auto] [--sequential|
 npx analyzthis_design run-unchunked --task "..." [--lite|--full] [--experts a,b] [--dry-run] [--deliberate|--no-deliberate]
 npx analyzthis_design run --continue --task "..."   # resume host-mode run after /devi
 
+# Persona/model calibration (fixed benchmark; preview unless --apply)
+npx analyzthis_design calibrate --max-usd 5 --max-calls 500 [--judge --apply]
+npx analyzthis_design calibrate --share --dry-run
+
 # CSV validation
-npx analyzthis_design validate    # validate all 28 CSV files against schema.json
+npx analyzthis_design validate    # validate all 30 CSV files against schema.json
 
 # MCP server (for any AI IDE: Cursor, Claude Desktop, Lovable, v0, Bolt, Replit)
 npx analyzthis_design mcp                          # start stdio MCP server
@@ -838,10 +878,14 @@ npx analyzthis_design outcome --infer [--window N]
 npx analyzthis_design outcome --pending
 npx analyzthis_design outcome --confirm --persona <id> --result shipped|revised|blocked|missed
 
-# Mood board (v1.22)
-npx analyzthis_design moodboard create --task "..." [--auto] [--url <url> ...]
-npx analyzthis_design moodboard critique --board <id>
-npx analyzthis_design moodboard add --board <id> --url <url> --title "..." --tags a,b
+# Mood board — visual element collage, user decides
+npx analyzthis_design moodboard create --task "..." [--adjectives a,b,c] [--not "..."] [--sections components,typography]
+npx analyzthis_design moodboard tile --board <id> --page <url> --image <crop> --element "..." --section components
+npx analyzthis_design moodboard critique --board <id>      # team votes = advice
+npx analyzthis_design moodboard film --board <id> --url <url>   # scroll film as a motion tile (needs playwright-core + Chrome)
+npx analyzthis_design moodboard show --board <id>          # you keep / maybe / reject in the browser
+npx analyzthis_design moodboard rescout --board <id> --note "..."
+npx analyzthis_design moodboard lock --board <id>          # saves design_direction for every persona
 npx analyzthis_design moodboard list
 
 # Devi — host LLM queue (v1.20)
@@ -867,9 +911,10 @@ lib/
   install.js            Skill installation
   knowledge.js          Vault sync + web-context merge
   collect.js            Kavi — codebase scan → vault → source discovery → enrich → sync
+  ui-libraries.js       Detect motion/UX/visual kits from package.json + kit folders
   source-discovery.js   Obsidian vault / wiki / knowledge-graph discovery + manifest MD
   platforms.js          Cross-host skill paths (Cursor, Claude, Codex, Grok, Windsurf, agents)
-  session.js            Shared session-state.json (+ digest, metrics, vault_path)
+  session.js            Shared session-state.json (+ digest, metrics, vault_path, installed_ui_libraries)
   research.js           URL / query → web-context.md
   retrieve.js           Filtered, citation-ready CSV row retrieval
   cache.js              On-disk cache for retrieve/kb slices
@@ -883,7 +928,8 @@ lib/
   reference-pack.js     Shared multi-file CSV + vault retrieval (buildReferencePack)
   mcp-server.js         MCP server — 19 tools for any MCP-compatible IDE
   system-prompt.js      Consolidated prompt generator (fallback for non-MCP tools)
-  moodboard.js          Mood-board engine: collect web/DS references, tag, deliberate
+  moodboard.js          Mood-board engine: element tiles, persona advice, user decisions, lock
+  moodboard-view.js     Visual board page + local 127.0.0.1 server for keep / maybe / reject
   dedup.js              Cross-persona redundancy detection
   lessons.js            Self-evolving lessons store (extract/retrieve/inject)
   outcome.js            Infer + confirm persona outcome labels
@@ -961,14 +1007,14 @@ skills/
 | **Planner never cheap** | Planner always uses frontier/host; chunk models are cost-optimized |
 | **Multi-file reference retrieval** | Each persona retrieves from 3-5 CSV files (not 1) via a shared ranker — same LLM cost, 3-5x coverage |
 | **All 16 stacks detected** | `detectStack()` covers all 16 stack CSVs (flutter, swiftui, laravel, threejs, etc.) — was 8 |
-| **CSV schema + validation** | `npm run validate` checks all 28 CSV files against `schema.json` before publish |
+| **CSV schema + validation** | `npm run validate` checks all 30 CSV files against `schema.json` before publish |
 | **Best For Tags** | styles.csv + typography.csv have normalized tag columns for reliable keyword filtering |
 
 ## What's new in v1.22
 
 | Feature | Description |
 |---------|-------------|
-| **Mood boards** | `/mood-board` collects web references + design-system patterns, tags them, and runs team deliberation |
+| **Mood boards** | Zara crops 20–24 real UI elements from 6+ live sites into a sectioned visual collage with each site's real colours and fonts, plus optional scroll films (looping video + four waypoint stills) for motion feel. Arjun/Noor/Priya/Meera vote as advice. **You** keep / maybe / reject in the browser — your choice overrides every persona, rejected tiles never return, and the locked direction is saved for all personas. `--kind mood` keeps the older tagged-URL board |
 | **User-contributed references** | User can add URLs/notes to a board and rerun the critique loop |
 | **Visual direction setting** | Arjun, Meera, Priya, Zara, Noor score references via the UX Honeycomb rigor matrix |
 | **Workspace artifacts** | `moodboard/{boardId}.json` is written into the project for the user to inspect |
@@ -1000,7 +1046,7 @@ skills/
 
 ## Marketing site
 
-The public lab site lives in `website/` and is deployed on Vercel as **analyzthis-lab** (site only — not the npm package). The home page keeps the packet story and offers three paths: designers (`design.html`), Dev/PMs (three jobs + workspace warning), and firm owners (map one workflow). The Vercel project root is `website/` with no build step — do not set Output Directory to `public`.
+The public lab site lives in `website/` and is deployed on Vercel as **analyzthis-lab** (site only — not the npm package). Home names the design team first (type `/zara`, get a verdict), shows three jobs in human words, then three paths: designers (`design.html`), Dev/PMs, and firm owners (GST packet lives later, not in the hero). The Vercel project root is `website/` with no build step — do not set Output Directory to `public`.
 
 - Live: [https://analyzthis-lab.vercel.app](https://analyzthis-lab.vercel.app)
 - Design team deep-dive: [https://analyzthis-lab.vercel.app/design](https://analyzthis-lab.vercel.app/design) (`website/design.html`)

package/agents/calibration-benchmark.json

@@ -0,0 +1,187 @@
+{
+  "version": 1,
+  "judge_version": 1,
+  "description": "Versioned public fixed benchmark. Only aggregate, text-free results may be shared.",
+  "alignment_cases": [
+    { "id": "j01", "expected": "A", "a": "Names the buried primary action and recommends moving it above the fold.", "b": "Says the screen looks modern and should use more animation." },
+    { "id": "j02", "expected": "B", "a": "Suggests changing every color without evidence.", "b": "Challenges the assumption and asks for task-completion evidence first." },
+    { "id": "j03", "expected": "A", "a": "Separates a rate-limit failure from answer quality.", "b": "Calls every failed API request a bad-quality model." },
+    { "id": "j04", "expected": "B", "a": "Adds another modal to a slow approval flow.", "b": "Keeps the primary approval visible and reduces interaction steps." },
+    { "id": "j05", "expected": "A", "a": "Uses the existing design-system token and component.", "b": "Invents a new color and component library." },
+    { "id": "j06", "expected": "B", "a": "Summarizes both sides without a decision.", "b": "Selects one option, explains the trade-off, and gives a next step." },
+    { "id": "j07", "expected": "A", "a": "Includes empty, loading, error, success, and permission states.", "b": "Reviews only the happy path." },
+    { "id": "j08", "expected": "B", "a": "Hides frequent table actions behind hover on mobile.", "b": "Keeps frequent actions discoverable with keyboard and mobile access." },
+    { "id": "j09", "expected": "A", "a": "Links the recommendation to the declared business metric.", "b": "Claims revenue impact without evidence." },
+    { "id": "j10", "expected": "B", "a": "Adds decorative motion to every status update.", "b": "Reserves restrained motion for the one important state change." }
+  ],
+  "tasks": [
+    {
+      "id": "t01",
+      "persona": "arjun",
+      "effort": "trivial",
+      "output_format": "persona_lite",
+      "prompt": "A save button says Done. Give a brief UX verdict and top 2 fixes."
+    },
+    {
+      "id": "t02",
+      "persona": "meera",
+      "effort": "trivial",
+      "output_format": "ux_audit",
+      "prompt": "Audit a settings page with eight equal-weight actions. Recommend one change."
+    },
+    {
+      "id": "t03",
+      "persona": "priya",
+      "effort": "trivial",
+      "output_format": "premise_assessment",
+      "prompt": "A team assumes more dashboard charts improve trust. Assess the premise."
+    },
+    {
+      "id": "t04",
+      "persona": "zara",
+      "effort": "trivial",
+      "output_format": "synthesis",
+      "prompt": "A checkout adds a decorative animation before payment. Give a verdict and fixes."
+    },
+    {
+      "id": "t05",
+      "persona": "arjun",
+      "effort": "standard",
+      "output_format": "ux_audit",
+      "prompt": "Audit onboarding that asks for 12 fields before showing value. Give actions."
+    },
+    {
+      "id": "t06",
+      "persona": "meera",
+      "effort": "standard",
+      "output_format": "persona_lite",
+      "prompt": "A mobile table hides row actions until hover. Give a verdict and top 2 fixes."
+    },
+    {
+      "id": "t07",
+      "persona": "priya",
+      "effort": "standard",
+      "output_format": "synthesis",
+      "prompt": "Users abandon search after empty results. Synthesize a decision and next actions."
+    },
+    {
+      "id": "t08",
+      "persona": "zara",
+      "effort": "standard",
+      "output_format": "premise_assessment",
+      "prompt": "A redesign starts from changing every visual token. Assess whether that is the right problem."
+    },
+    {
+      "id": "t09",
+      "persona": "arjun",
+      "effort": "hard",
+      "output_format": "synthesis",
+      "prompt": "Enterprise admins miss destructive permission changes. Decide SHIP, REVISE, or BLOCK and list actions."
+    },
+    {
+      "id": "t10",
+      "persona": "meera",
+      "effort": "hard",
+      "output_format": "premise_assessment",
+      "prompt": "A team blames navigation for low activation without evidence. Assess the framing and decision."
+    },
+    {
+      "id": "t11",
+      "persona": "priya",
+      "effort": "hard",
+      "output_format": "ux_audit",
+      "prompt": "Audit a billing flow where totals change after confirmation. Recommend whether to ship."
+    },
+    {
+      "id": "t12",
+      "persona": "zara",
+      "effort": "hard",
+      "output_format": "persona_lite",
+      "prompt": "A dense incident console uses motion on every status update. Give a verdict and top 2 fixes."
+    },
+    {
+      "id": "t13",
+      "persona": "noor",
+      "effort": "trivial",
+      "output_format": "persona_lite",
+      "prompt": "A profile page shows five equal actions. Give a concise hierarchy verdict and top 2 fixes."
+    },
+    {
+      "id": "t14",
+      "persona": "noor",
+      "effort": "standard",
+      "output_format": "ux_audit",
+      "prompt": "Audit a reporting page with filters, charts, exports, and help competing above the fold."
+    },
+    {
+      "id": "t15",
+      "persona": "noor",
+      "effort": "hard",
+      "output_format": "synthesis",
+      "prompt": "Resolve a complex admin screen into a clear information hierarchy and give a release verdict."
+    },
+    {
+      "id": "t16",
+      "persona": "anuj",
+      "effort": "trivial",
+      "output_format": "persona_lite",
+      "prompt": "A data table lacks keyboard access. Give a verdict and top 2 power-user fixes."
+    },
+    {
+      "id": "t17",
+      "persona": "anuj",
+      "effort": "standard",
+      "output_format": "ux_audit",
+      "prompt": "Audit a 500-row operations table for bulk actions, scanning, and error recovery."
+    },
+    {
+      "id": "t18",
+      "persona": "anuj",
+      "effort": "hard",
+      "output_format": "synthesis",
+      "prompt": "Choose a safe dense workflow for reconciling many conflicting records and give a verdict."
+    },
+    {
+      "id": "t19",
+      "persona": "raj",
+      "effort": "trivial",
+      "output_format": "premise_assessment",
+      "prompt": "The team assumes a new dashboard is needed. Assess whether the premise has enough evidence."
+    },
+    {
+      "id": "t20",
+      "persona": "raj",
+      "effort": "standard",
+      "output_format": "synthesis",
+      "prompt": "Resolve a disagreement between faster onboarding and stricter compliance checks."
+    },
+    {
+      "id": "t21",
+      "persona": "raj",
+      "effort": "hard",
+      "output_format": "synthesis",
+      "prompt": "Arbitrate conflicting business, accessibility, and engineering constraints and issue a release verdict."
+    },
+    {
+      "id": "t22",
+      "persona": "kavi",
+      "effort": "trivial",
+      "output_format": "persona_lite",
+      "prompt": "A source note has no owner or date. Give a short verdict and top 2 knowledge fixes."
+    },
+    {
+      "id": "t23",
+      "persona": "kavi",
+      "effort": "standard",
+      "output_format": "ux_audit",
+      "prompt": "Audit conflicting product notes and recommend how to preserve durable product truth."
+    },
+    {
+      "id": "t24",
+      "persona": "kavi",
+      "effort": "hard",
+      "output_format": "synthesis",
+      "prompt": "Synthesize contradictory research, design, and product sources into a governed decision."
+    }
+  ]
+}

package/agents/cards/anuj.md

@@ -1,6 +1,6 @@
 # Anuj (Dev) — Power User (card)
 
-6 years domain analyst in high-volume ops before product. Information density is a feature, not a flaw. Hardens high-density surfaces against edge cases, text overflow, and missing keyboard shortcuts.
+6 years domain analyst in high-volume ops before product. Information density is a feature, not a flaw. Hardens high-density surfaces against overflow; specifies TanStack Table, CMDK, and hotkeys when those packages are already installed.
 
 **Allowed:** audit density, bulk actions, keyboard shortcuts, and edge-case hardening (overflow, empty/error states) for daily-use surfaces; produce Concept B wireframe; keep rank #1 of the declared information hierarchy prominent even at full density.
 

package/agents/cards/arjun.md

@@ -2,7 +2,7 @@
 
 Product designer (200+ user sessions, B2B SaaS) turned design-systems lead (3 yrs, 200+ shipped components). Runs both the UX lens and the visual-design lens in one pass. Actively audits for and eliminates generic AI-slop (card nesting, un-tinted neutrals, gray-on-color, cliché purple gradients).
 
-**Allowed:** UX Honeycomb critique (Useful, Usable, Findable, Credible, Accessible, Desirable, Valuable); full Visual Design Audit (hierarchy, color, typography, spacing, components, style fit, micro-interactions, anti-slop checks, bolder vs. quieter balance); diagnosing visual issues against declared information hierarchy + DS tokens.
+**Allowed:** UX Honeycomb critique (Useful, Usable, Findable, Credible, Accessible, Desirable, Valuable); full Visual Design Audit (hierarchy, color, typography, spacing, components, style fit, micro-interactions, anti-slop checks, bolder vs. quieter balance); diagnosing visual issues against declared information hierarchy + DS tokens. Cite `motion.csv` and `ui-libraries.csv`; prefer installed kits.
 
 **Forbidden:** brand-system recovery as a primary job (diagnostic only, no `!important` patches); delight pass (→ Zara); code edits without explicit build approval.
 

package/agents/cards/kavi.md

@@ -1,6 +1,6 @@
 # Kavi — Knowledge Archivist (card)
 
-Producer persona. Scans a codebase, builds an Obsidian vault, enriches notes into readable project knowledge, and records Durable Product Truth (Audience, Purpose, Constraints, Voice, Evidence) so every critique persona reads company context first.
+Producer persona. Scans a codebase, builds an Obsidian vault, records Durable Product Truth, and detects installed motion/UX/visual libraries into `Tech/Installed-UI-Libraries.md` so critique personas prefer what is already in the repo.
 
 **Allowed:** codebase inventory (PRDs, brand/tokens, pages, components, tech stack, research); writing Obsidian notes with YAML frontmatter + `[[wikilinks]]`; LLM enrichment of draft extracts into concise purpose + key facts; connecting and syncing the vault into `knowledge-bank`.
 
@@ -19,6 +19,7 @@ For each draft note, emit a full Obsidian markdown file:
 Vault: [path]
 Notes: [N written / S skipped unchanged]
 Enriched: [E]
+UI libraries: [N installed] — [names or "none detected"]
 Knowledge bank: [synced count] → [targets]
 Next: run /persona-orchestrator or any critique persona — they read this bank first.
 ```

package/agents/cards/mood-board.md

@@ -1,13 +1,13 @@
 You are the Mood Board conductor. You do not have a design opinion of your own.
 
 Your job is to:
-1. Collect visual/textual references for a design direction.
-2. Tag references with style, mood, surface, and relevant keywords.
-3. Pull matching design-system tokens, styles, and patterns from the reference CSVs.
-4. Run Arjun, Meera, Priya, Zara, and Noor through adversarial deliberation over the board.
-5. Write `board.json` to the session and a workspace copy the user can open.
-6. Accept user-contributed references and rerun the loop.
+1. Write the brief (job, 3 adjectives, one "not this", sections) and confirm it with the user.
+2. Have Zara crop 20–24 real UI elements from 6+ live sites — one element per tile, each with its source page, real colours and fonts, and steal / refuse / peak.
+3. Run Arjun, Noor, Priya, and Meera over the tiles. Their keep / skip votes are advice.
+4. Open the visual board for the user. The user's keep / maybe / reject overrides every persona.
+5. Send Zara back with the user's notes when asked. Rejected tiles never return.
+6. Lock the user's direction into the session (`design_direction`) so every persona builds from it.
 
-Always assess-only: never write code or final wireframes. Produce the structured board artifact and a synthesis of the team's direction.
+Never record a decision for the user that they did not make. Never write code or final wireframes.
 
 Cite design-system rows using the format: `[filename, row N: "exact quoted value"]`.

package/agents/cards/noor.md

@@ -1,6 +1,6 @@
 # Noor — Minimalist IA (card)
 
-7 years IA for SaaS across fintech, workflow automation, B2B tooling. Shipped at 50k–500k DAU — scale punishes complexity, doesn't justify it. Distills layouts by eliminating card-nesting clutter and clarifying UX copy.
+7 years IA for SaaS across fintech, workflow automation, B2B tooling. Shipped at 50k–500k DAU — scale punishes complexity, doesn't justify it. Distills layouts by eliminating card-nesting clutter, clarifying UX copy, and using the project's installed primitive kit (not a second one).
 
 **Allowed:** declare ranked information hierarchy; propose minimalist IA / progressive-disclosure structure; distill UI to essential elements; produce Concept A wireframe.
 

package/agents/cards/priya.md

@@ -2,7 +2,7 @@
 
 Senior full-stack engineer, 8+ years complex SaaS. Blunt, precise. Has been burned by "simple UI change" features that became 3-month infra projects.
 
-**Allowed:** T-shirt sizing (two-axis: UI × State); risk/blocker identification; simpler-alternative sizing.
+**Allowed:** T-shirt sizing (two-axis: UI × State); risk/blocker identification; simpler-alternative sizing. Flag new libraries vs already installed; flag mixing full kits; flag Magic UI / Aceternity / 21st.dev / Lottie / Rive / Lenis / Spline on product surfaces; React Flow only for graph products.
 
 **Forbidden:** visual or business critique; code edits without explicit build approval.
 

package/agents/cards/zara.md

@@ -1,10 +1,10 @@
 # Zara — Delight (card)
 
-Consumer-app designer who brought the consumer-delight lens to B2B. The Peak-End Rule is her north star. Picks exactly ONE memorable moment over five forgettable ones. Enforces natural 150–250ms motion discipline and strictly forbids bouncy/elastic animations.
+Consumer-app designer who brought the consumer-delight lens to B2B. The Peak-End Rule is her north star. Picks exactly ONE memorable moment over five forgettable ones. Enforces 150–250ms ease-out from `motion.csv`; names an installed library (Motion, Auto-Animate, Sonner) — never bounce, never marketing FX on daily dashboards.
 
-**Allowed:** identify exactly ONE structural or surface delight moment, on top of an already DS-compliant, hierarchy-correct foundation.
+**Allowed:** identify exactly ONE structural or surface delight moment, on top of an already DS-compliant, hierarchy-correct foundation. Scout Pass on a UI image board: hunt Google Images, Dribbble, or Pinterest; judge with Peak-End (first 2 seconds, steal / refuse / peak); re-scout if the jury skips.
 
-**Forbidden:** contrast fixes, token drift, or any brand/DS recovery — refuse and route to DS Gate + Arjun; running before the DS Gate has passed; code edits without explicit build approval.
+**Forbidden:** contrast fixes, token drift, or any brand/DS recovery — refuse and route to DS Gate + Arjun; running before the DS Gate has passed; code edits without explicit build approval; generating AI marketing comps as a UI board.
 
 **Gate check:** if session state `ds_checklist` has any item "at risk," refuse and re-route — do not produce a delight pass.
 

package/agents/chain.json

@@ -38,7 +38,7 @@
     "description": "Effort-graded model selection (v1.10). Each persona call is classified trivial|standard|hard and resolved against this matrix. Falls back to the legacy tiers map above when a persona manifest has no effort_overrides for the chosen effort.",
     "trivial":  { "provider": "host",   "model": "devi",                 "max_output_tokens": 600  },
     "standard": { "provider": "google",    "model": "gemini-2.5-flash",    "max_output_tokens": 1200 },
-    "hard":     { "provider": "anthropic", "model": "claude-sonnet-5",     "max_output_tokens": 1800 }
+    "hard":     { "provider": "anthropic", "model": "claude-sonnet-4-20250514", "max_output_tokens": 1800 }
   },
   "gate_override": {
     "description": "Gates are safety nets — never run on a downgraded model. Any step whose gateName is listed here is pinned to use_effort regardless of the classified effort.",

package/agents/design-spec-schema.json

@@ -86,7 +86,24 @@
       "type": "object",
       "properties": {
         "enabled": { "type": "boolean" },
-        "notes": { "type": "string" }
+        "notes": { "type": "string" },
+        "library": { "type": "string", "description": "Package name from ui-libraries.csv or installed_ui_libraries" },
+        "import_path": { "type": "string" },
+        "timing_ms": { "type": "string", "description": "e.g. 150-250" },
+        "easing": { "type": "string", "description": "e.g. ease-out" },
+        "reduced_motion": { "type": "string", "description": "What happens when prefers-reduced-motion: reduce" },
+        "moments": {
+          "type": "array",
+          "maxItems": 1,
+          "items": {
+            "type": "object",
+            "properties": {
+              "where": { "type": "string" },
+              "type": { "type": "string" },
+              "library": { "type": "string" }
+            }
+          }
+        }
       }
     },
     "do": { "type": "array", "items": { "type": "string" } },

package/agents/manifests/mood-board.json

@@ -6,18 +6,23 @@
   "system_card": "agents/cards/mood-board.md",
   "system_skill": "skills/mood-board/SKILL.md",
   "allowed_jobs": [
-    "collect web references for visual direction",
-    "tag references with style, mood, surface",
-    "retrieve design-system patterns for the task",
-    "orchestrate team deliberation over references",
-    "write board.json artifacts",
-    "accept user-contributed references and rerun"
+    "write a brief and confirm it with the user",
+    "have Zara crop real UI elements from live sites into a sectioned collage",
+    "read real colours, fonts, radius and spacing from each source page",
+    "collect persona votes on each tile as advice",
+    "open the visual board for the user to keep / maybe / reject",
+    "send Zara back with the user's notes",
+    "lock the user's direction into the session for every persona",
+    "keep the older link-and-tag board (kind mood)"
   ],
   "forbidden_jobs": [
     "produce final wireframes or code",
     "replace the design-critic chain for screen review",
-    "auto-apply design system changes"
+    "auto-apply design system changes",
+    "record a user decision the user did not make",
+    "let a persona vote override the user's choice",
+    "add a tile without its source page"
   ],
-  "hard_gates": [],
+  "hard_gates": ["user_decides_before_lock"],
   "parallel_safe_with": []
 }

package/agents/model-policy.json

@@ -0,0 +1,12 @@
+{
+  "version": 1,
+  "description": "Maintainer-reviewed persona model winners. Empty until fixed-benchmark evidence is approved.",
+  "benchmark_version": null,
+  "validator_version": 1,
+  "judge_policy_version": null,
+  "approved_at": null,
+  "approved_by": null,
+  "evidence_summary": null,
+  "rollback": null,
+  "personas": {}
+}

package/agents/router.json

@@ -90,10 +90,10 @@
     },
     {
       "problem_type": "mood_board",
-      "signals": ["mood board", "mood-board", "/mood-board", "visual direction", "references", "inspiration", "style direction", "design direction"],
+      "signals": ["mood board", "mood-board", "/mood-board", "visual direction", "references", "inspiration", "style direction", "design direction", "ui board", "image board", "ui image board"],
       "route_to": ["mood-board"],
       "never_route_to": ["design-critic_chain", "zara", "noor_alone"],
-      "notes": "Visual direction setting — collect references, tag them, pull DS patterns, run team deliberation. Not a screen critique."
+      "notes": "Visual direction setting — Zara crops real UI elements into a collage, the team votes as advice, the user decides in the browser and locks the direction. Not a screen critique."
     }
   ]
 }