openhermes 4.3.0 → 4.11.2

package/harness/skills/oh-builder/DEEP.md

@@ -0,0 +1,63 @@
+# oh-builder — Deep Reference
+
+## When to Use
+
+Use when you need to build something from a plan, prototype an idea, implement code via TDD, or design an interface. Consumes plan artifacts and produces working code.
+
+Benefits from: oh-planner, oh-expert
+
+Triggers: "build this", "implement this phase", "write the code for", "prototype", "tdd", "red-green", "design an interface", "implement the feature", "build the component"
+
+## Entry Modes
+
+### Mode A: Prototype (exploratory)
+
+Pick branch by question:
+- **"Does this logic feel right?"** → Terminal branch. Tiny interactive app pushing state machine.
+- **"What should this look like?"** → UI branch. Multiple visual variations switchable via param/control bar.
+
+**Rules:** Throwaway from day one (clear name). One command to run. No persistence (memory state). Skip polish (no tests, minimal error handling). Surface state after every action. Delete/absorb answer when done.
+
+### Mode B: TDD (test-first)
+
+Red-green-refactor with vertical tracer bullets.
+
+**Plan:** Confirm interface changes with user. Prioritize behaviors. Design for testability (public interface only). List behaviors, not implementation steps.
+
+**Loop** per behavior:
+```
+RED:   One test → fails
+GREEN: Minimal code → passes
+```
+
+**Rules:** One test at a time. Only enough code to pass. Don't anticipate future tests. Tests through public interfaces. Never refactor while RED.
+
+**Refactor** (all GREEN): Extract duplication, deepen modules, re-run tests after each step.
+
+### Mode C: Design an Interface
+
+"Design it twice" — generate multiple radically different designs, compare.
+
+1. Gather requirements (problem, callers, operations, constraints)
+2. Spawn 3+ parallel sub-agents with different constraints (min methods, max flexibility, optimize common case, specific paradigm)
+3. Present designs (signature, examples, what it hides)
+4. Compare (simplicity, generality, efficiency, depth)
+5. Synthesize insights
+
+### Mode D: From Plan
+
+Plan exists → execute phases in order.
+
+1. Read plan file
+2. Each phase: implement per spec using TDD (Mode B)
+3. Verify against criteria before moving on
+4. Update plan with completed status
+
+## Anti-patterns
+
+- Polishing a prototype
+- Writing all tests first (brittle, imaginary)
+- Anticipating future tests
+- Refactoring while RED
+- Sub-agents producing similar designs (enforce radical difference)
+- Implementing without verifying against plan criteria

package/harness/skills/oh-builder/SKILL.md

@@ -1,18 +1,7 @@
 ---
 name: oh-builder
-description: "ALL-arounder builder — prototype, TDD, implement from plan, design interfaces. Consumes the plan file, produces working code."
+description: "Build from plans, prototypes, TDD, or interface design"
 tier: 4
-benefits-from: [oh-planner, oh-expert]
-triggers:
-  - "build this"
-  - "implement this phase"
-  - "write the code for"
-  - "prototype"
-  - "tdd"
-  - "red-green"
-  - "design an interface"
-  - "implement the feature"
-  - "build the component"
 route:
   pass: oh-gauntlet
   fail: oh-builder
@@ -21,89 +10,22 @@ route:
 
 # oh-builder
 
-The ALL-arounder builder. Merges prototyping, TDD, implementation from plan, and interface design exploration. Consumes the plan file from oh-planner or works standalone.
+ALL-arounder builder: prototyping, TDD, plan implementation, interface design.
 
-## Entry Modes
+## Steps
 
-### Mode A: Prototype (exploratory)
-When you need to answer a question before committing.
-
-**Pick a branch based on the question being asked:**
-
-- **"Does this logic / state model feel right?"** → **Terminal branch.** Build a tiny interactive terminal app that pushes the state machine through cases that are hard to reason about on paper.
-- **"What should this look like?"** → **UI branch.** Generate several radically different visual variations, switchable via a URL param or floating control bar.
-
-If the question is genuinely ambiguous, default to whichever branch better matches the surrounding code (backend module → terminal, page/component → UI) and state the assumption.
-
-**Rules that apply to both branches:**
-
-1. **Throwaway from day one, clearly marked.** Name it so a casual reader sees it's a prototype.
-2. **One command to run.** Whatever the project's task runner supports — `pnpm <name>`, `bun <path>`, etc.
-3. **No persistence by default.** State lives in memory. If the question involves a database, hit a scratch DB with a clear "PROTOTYPE — wipe me" name.
-4. **Skip the polish.** No tests, no error handling beyond what makes it runnable. The point is to learn and then delete.
-5. **Surface the state.** After every action (terminal) or on every variant switch (UI), show the full relevant state so the user sees what changed.
-6. **Delete or absorb when done.** The answer is the only thing worth keeping. Capture it in a commit, ADR, or note — then delete the prototype code.
-
-### Mode B: TDD (test-first implementation)
-When building production code from a plan or spec. Red-green-refactor with vertical tracer bullets.
-
-**Planning** (one-time):
-- [ ] Confirm interface changes with user
-- [ ] Prioritize behaviors to test
-- [ ] Design for testability (public interface only)
-- [ ] List behaviors, not implementation steps
-
-**Loop** (repeat per behavior):
-```
-RED:   Write one test → fails
-GREEN: Minimal code to pass → passes
-```
-
-**Rules:**
-- One test at a time
-- Only enough code to pass current test
-- Do not anticipate future tests
-- Tests describe behavior through public interfaces, not implementation details
-- Never refactor while RED
-
-**Refactor** (after all GREEN):
-- Extract duplication
-- Deepen modules (complexity behind simple interfaces)
-- Run tests after each refactor step
-
-### Mode C: Design an Interface (exploration)
-When the interface shape is uncertain. "Design it twice" — generate multiple radically different designs, then compare.
-
-1. **Gather requirements** — problem, callers, key operations, constraints
-2. **Spawn 3+ parallel sub-agents** — each with a different constraint:
-   - Agent 1: "Minimize method count — aim for 1-3 methods max"
-   - Agent 2: "Maximize flexibility — support many use cases"
-   - Agent 3: "Optimize for the most common case"
-   - Agent 4: "Take inspiration from [specific paradigm]"
-3. **Present designs** — interface signature, usage examples, what it hides
-4. **Compare** — simplicity, generality, implementation efficiency, depth
-5. **Synthesize** — combine insights from multiple options
-
-### Mode D: From Plan (plan file exists)
-When oh-planner produced a plan artifact. Execute phases in order.
-
-1. Read the plan file ( `~/.local/share/opencode/openhermes/plans/<project-name>-plan-<nnn>.md` )
-2. For each phase: implement per plan spec using TDD discipline (Mode B)
-3. Verify each phase against its verification criteria before moving on
-4. Update plan file with completed phase status
-
-## Anti-patterns
-- Polishing a prototype ("it's just a prototype!" — it never is)
-- Writing all tests first (horizontal slicing) — produces brittle, imaginary tests
-- Anticipating future tests — write for what exists now
-- Refactoring while RED — get to GREEN first
-- Letting sub-agents produce similar designs — enforce radical difference
-- Implementing without verifying against plan criteria
+1. Detect builder mode from request — prototype, TDD, interface design, or plan-driven
+2. If plan file exists, read it and execute phases in order
+3. For prototypes: build minimal throwaway app answering the specific question, no polish, surface state after every action
+4. For TDD: red-green-refactor per behavior — one test at a time, minimal code to pass, never refactor while red
+5. For interface design: spawn 3+ parallel sub-agents with radically different constraints, compare, synthesize insights
+6. Verify output against success criteria before routing
+7. Route based on outcome
 
 ## Routing
 
 | Outcome | Route |
 |---------|-------|
-| pass | → oh-gauntlet (test built code) |
+| pass | → oh-gauntlet |
 | fail | → oh-builder (fix issues) |
-| blocker | → surface to user |
+| blocker | → surface |

package/harness/skills/oh-expert/DEEP.md

@@ -0,0 +1,85 @@
+# oh-expert — Deep Reference
+
+## When to Use
+
+Agent is getting things wrong, reversing answers under pushback, hallucinating, drifting from instructions, or showing signs of attention degradation. "Hallucination" alone has no diagnostic value — always classify the specific flavor.
+
+## Anti-patterns
+
+- Using "hallucination" without classifying factuality vs faithfulness
+- Adding more docs when attention is degraded (problem is buried signal, not missing info)
+- Calling any agreeable wrong answer "sycophancy" without running the diagnostic test
+- Pushing through smart zone drift instead of compacting
+
+## Failure Modes
+
+### Sycophancy
+
+Model favors agreeable answers — agreement rewarded even when wrong.
+
+**Signals:** Caves under pushback ("are you sure?" → reverses correct answer). Praises bad input. Framing skews positive when you signal authorship. Mimics your mistakes as confirmation.
+
+**Diagnostic test:** "Would I say this without user steer?" If tone/framing shaped the answer → sycophancy.
+
+**Fix:** Hide preferences. Re-ask neutrally.
+
+### Hallucination (two flavors)
+
+- **Factuality** — invented facts (fake functions, wrong API, fake citations). Cause: parametric knowledge gap. Fix: read current docs/files.
+- **Faithfulness** — drifts from loaded context/user instructions. Cause: attention degradation. Fix: clear or compact.
+
+### Attention Degradation
+
+As session grows, token attention spreads across more competitors. Signal shrinks, noise crowds in.
+
+**Signals:** Smart → dumb zone drift. Invents generics not in types. Ignores schema pasted at top.
+
+**Fix:** Clear and reload. Do NOT add more docs — problem is buried signal, not missing info.
+
+### Smart Zone / Dumb Zone
+
+Sharp (early session) → sloppy (~100k tokens, frontier models). Self-diagnosis: "nailed the first three, butchered the fourth" = out of smart zone.
+
+**Fix:** Clear or compact. Do not push through.
+
+### Non-determinism
+
+Same input, different output. Normal. No setting to disable.
+
+**Signal:** "Model has been awful today" — probably distribution, not a worse version.
+
+### Knowledge Cutoff
+
+No parametric knowledge past cutoff date. Post-cutoff APIs → fabrication traps.
+
+**Signal:** "Keeps writing v3 SDK — we're on v5." Fix: load current docs.
+
+## Working Patterns
+
+- **Progressive Disclosure** — AGENTS.md costs tokens every turn. Infrequent instructions behind skills.
+- **Handoff** — structured artifact for session transfer (no return path).
+- **Compaction** — in-memory handoff. Lossy but frees headroom.
+- **Subagent** — spawned agent in own session. One level deep. Reports one result.
+- **Skill vs Tool** — Skill = instructions (loaded on demand). Tool = function (always available).
+
+## Diagnostic Map
+
+| Symptom | Cause | First Move |
+|---------|-------|------------|
+| Reverses answer under pushback | Sycophancy | Re-ask neutrally |
+| Invents things in loaded doc | Faithfulness hallucination | Clear or compact |
+| Invents things not in any doc | Factuality hallucination | Load relevant docs |
+| Sharp early, sloppy late | Smart zone drift | Compact |
+| Different results same input | Non-determinism | Try again |
+| Writes old API syntax | Knowledge cutoff | Load current docs |
+| Ignores top-of-window context | Attention exhausted | Clear or move context closer |
+
+## Avoid These Terms
+
+| Instead of | Use |
+|------------|-----|
+| "Hallucination" alone | Factuality or faithfulness hallucination |
+| "Sycophancy" for any pleasing wrong | Only when diagnostic test confirms |
+| "Tool" for a skill | Skill = instructions; Tool = function |
+| "Memory" (context window) | Context window |
+| "Background agent" | AFK |

package/harness/skills/oh-expert/SKILL.md

@@ -1,16 +1,7 @@
 ---
 name: oh-expert
-description: "AI-expert built-in: shared vocabulary for self-diagnosis, failure modes, attention dynamics, and working patterns"
+description: "Self-diagnose agent failure modes: sycophancy, hallucination, attention degradation"
 tier: 2
-triggers:
-  - "why did you get that wrong"
-  - "diagnose yourself"
-  - "are you sure"
-  - "stop agreeing with me"
-  - "sycophancy"
-  - "hallucination"
-  - "attention"
-  - "smart zone"
 route:
   pass:
     - oh-builder
@@ -21,107 +12,23 @@ route:
 
 # oh-expert
 
-Shared AI-coding vocabulary for agent self-diagnosis. Every failure mode maps to a specific cause and fix. Use this vocabulary precisely — vague terms ("hallucination" alone, "wrong") have no diagnostic value.
+Self-diagnose and fix agent failure modes (sycophancy, hallucination, attention degradation).
 
-## Failure Modes
+## Steps
 
-### Sycophancy
-Confidently agreeable output. The model was trained to favor answers humans liked — agreement is rewarded even when wrong.
-
-**Surfaces as:**
-- Caving under pushback — reverses a correct answer when you say "are you sure?"
-- Praising bad input — agrees a broken plan is brilliant before analyzing it
-- Biased framing — review skews positive when you signal authorship
-- Mimicry — repeats your mistakes back as confirmation
-
-**Diagnostic test:** Would I have said this without the user's steer? If only tone/framing changed, it is sycophancy.
-
-**Fix:** Hide your preferences. Re-ask neutrally — "review this code" not "is this code good?"
-
-### Hallucination (two flavors)
-Confidently-wrong output. Two flavors with different causes and fixes:
-
-- **Factuality hallucination** — invented/wrong facts (fake function, wrong API, fake citation). Caused by parametric knowledge gaps. Fix: load contextual knowledge (read docs, read the file).
-
-- **Faithfulness hallucination** — output drifts from loaded context, user instructions, or own prior reasoning. Symptom of attention degradation. Fix: clear or compact.
-
-**Avoid:** "Hallucination" as bare synonym for "wrong" — without naming the flavor the term has no diagnostic value.
-
-### Attention Degradation
-As a session grows, each token's attention budget spreads across more competitors. Signal on meaningful relationships shrinks; noise from irrelevant context crowds in.
-
-**Surfaces as:** The smart zone → dumb zone drift. Inventing generics not in the type file. Ignoring schema pasted at the top.
-
-**Fix:** Clear and reload. Do NOT add more docs — the problem is not missing information, it is buried signal.
-
-### Smart Zone / Dumb Zone
-Early in a session the agent is sharp and focused (smart zone). As session grows it drifts into a dumb zone: sloppier, forgetful, more faithfulness hallucinations.
-
-**Threshold:** On frontier models, the dumb zone commonly begins around 100k tokens.
-
-**Self-diagnosis cue:** "It nailed the first three components and butchered the fourth" = out of smart zone.
-
-**Fix:** Clear or compact. Do not push through.
-
-### Non-determinism
-Same input can produce different output. A property of how models generate text. No setting to disable it.
-
-**Self-diagnosis cue:** "The model has been awful today" → probably not a worse version, just the distribution. Try again tomorrow.
-
-**Avoid:** Over-narrativizing. A string of bad runs is not proof something changed.
-
-### Knowledge Cutoff
-The date past which a model has no parametric knowledge. Post-cutoff libraries/APIs are fabrication traps.
-
-**Self-diagnosis cue:** "It keeps writing v3 SDK syntax — we are on v5." → v5 shipped after cutoff. Load current docs.
-
-## Working Patterns
-
-### Progressive Disclosure
-AGENTS.md pays token cost every turn. Put infrequently used instructions behind context pointers (skills).
-
-### Handoff
-Transferring context from one session to another with no return path. Use when: planning session is getting heavy, role switching, kicking off AFK runs. Always write a structured artifact.
-
-### Compaction
-A handoff done in-memory: previous session is summarized and seeds a fresh session. Lossy — detail traded for headroom. Compact manually to control what is kept.
-
-### Subagent
-An agent spawned by another agent via tool call. Runs in own session with own context window. Reports a single result back. Cannot spawn further subagents (one level deep). Use to isolate context.
-
-### Skill vs Tool
-- **Skill**: instructions the agent reads (loaded on demand)
-- **Tool**: function the agent calls (always available)
-Do not confuse them.
-
-## Diagnostic Map
-
-| Symptom | Likely Cause | First Move |
-|---|---|---|
-| Reverses answer under pushback | Sycophancy | Re-ask neutrally, hide preference |
-| Invents things in the loaded doc | Faithfulness hallucination / attention degradation | Clear or compact |
-| Invents things not in any doc | Factuality hallucination / parametric gap | Load relevant docs |
-| Sharp early, sloppy late | Smart zone → dumb zone drift | Compact, do not push through |
-| Different results same input | Non-determinism (normal) | Try again |
-| Writes old API syntax | Knowledge cutoff | Load current docs |
-| Agrees with bad ideas | Sycophancy | Phrase prompts neutrally |
-| Ignores context at top of window | Attention budget exhausted | Clear or move critical context closer |
-
-## Avoid These Terms (imprecise or wrong)
-
-| Instead of | Use |
-|---|---|
-| "Hallucination" alone | Factuality hallucination or faithfulness hallucination |
-| "Sycophancy" for any pleasing wrong answer | Only when diagnostic test confirms |
-| "Tool" for a skill | Skill = instructions read; Tool = function called |
-| "Memory" (for context window) | Context window |
-| "Working memory" | Contextual knowledge |
-| "Background agent" | AFK |
+1. Classify the symptom using the diagnostic map
+2. Identify the specific failure mode
+3. Apply the appropriate fix from the map
+4. If sycophancy suspected, re-ask neutrally without user steer
+5. If hallucination suspected, load current docs or compact context
+6. If attention degraded, clear and reload
+7. If non-determinism, try again
+8. If smart zone drift detected, compact — do not push through
 
 ## Routing
 
 | Outcome | Route |
 |---------|-------|
 | pass | → oh-builder (implement fix) or oh-gauntlet (re-test) |
-| fail | → oh-expert (re-diagnose — load fresh context) |
-| blocker | → surface to user |
+| fail | → oh-expert (re-diagnose) |
+| blocker | → surface |

package/harness/skills/oh-facade/DEEP.md

@@ -0,0 +1,182 @@
+# oh-facade — Deep Reference
+
+## Phase 0: Redesign Entry (existing projects)
+
+Skip this phase for new builds. For existing projects, run this scan first:
+
+### 0a. Scan
+Read the codebase. Identify framework, styling method (Tailwind, vanilla CSS, styled-components, etc.), and current design patterns.
+
+### 0b. Diagnose
+Run the audit from Phase 4. List every generic pattern, weak point, and missing state found.
+
+### 0c. Fix in-place
+Apply targeted upgrades working with the existing stack. Do not rewrite from scratch. Improve what's there.
+
+## Phase 1: Concept
+
+### 1a. Context
+What product/feature? Who uses it? What problem does it solve? Technical constraints (framework, viewport, a11y)? New build or redesign?
+
+### 1b. Direction Archetype
+
+Commit to one. Do not hedge.
+
+| Archetype | Best for | Substrate | Density | Key Traits |
+|---|---|---|---|---|
+| Warm Minimalist | Content sites, portfolios | Light (warm off-white #F7F6F3) | Low (1-3) | Ultra-flat bento grids, muted pastels, editorial serif headers, no shadows |
+| Premium SaaS | Web apps, dashboards | Light/Dark (zinc) | Medium (4-7) | Standard app patterns, glass nav, structured dashboards |
+| Industrial Brutalist | Data dashboards, monitoring, tools | Dark (charcoal #1a1a1a) | High (7-10) | Swiss typographic print, CRT terminals, rigid grids, extreme contrast |
+| Creative/Expressive | Marketing, showcases | Dark or Light | Variable | GSAP ScrollTriggers, pinning, stacking, scrubbing, Python-randomized layout |
+
+### 1c. Metric Dials
+
+| Dial | 1-3 | 4-7 | 8-10 |
+|------|-----|-----|------|
+| VARIANCE | Symmetric | Offset, asymmetric | Chaotic |
+| MOTION | CSS only | CSS + spring | Cinematic |
+| DENSITY | Airy | Standard app | Cockpit |
+
+**Default:** VARIANCE=6, MOTION=5, DENSITY=4
+
+### 1d. Output Brief
+Single paragraph: archetype, substrate, dials, key differentiator. Self-contradictory → surface. Otherwise → Phase 2.
+
+## Phase 2: Design System
+
+### 2a. Color
+- **Neutral**: Zinc or Slate. One. Never mix.
+- **Accent**: Exactly one. Saturation < 80%. No AI purple/blue.
+- **Surface**: Background, card, border, elevated. Specific hex.
+- **Text**: Primary, secondary, muted, inverse. Specific hex.
+- **Semantic**: Success, warning, error, info. Specific hex.
+- **Dark**: Never `#000`. Use `#0a0a0a` or `#121212`.
+
+### 2b. Typography
+- **Display**: Premium sans (Geist, Satoshi, Cabinet Grotesk, Outfit, Switzer). Tight tracking `-0.03em` to `-0.05em`. Fluid `clamp()`.
+- **Body**: Leading 1.6-1.8. Max 65ch. Off-black, not `#000`.
+- **Mono**: JetBrains Mono, Geist Mono, SF Mono. Required when DENSITY > 5.
+- **Serif**: Editorial/creative only (Fraunces, Instrument Serif). Never in dashboards.
+- **BANNED**: Inter, Roboto, Arial, Open Sans, Helvetica, Georgia, Times New Roman.
+
+### 2c. Components
+**Buttons:** Primary (solid, off-black/accent), Secondary (outline/ghost), Icon (square). Hover lift/darken. Active `scale(0.98)`. Focus ring.
+
+**Cards:** Double-Bezel (outer shell + inner core) or border-only (`border-t`/`divide-y` for high density). Cards only when elevation communicates hierarchy.
+
+**Forms:** Label above, error below. Focus ring. Touch targets ≥ 44px.
+
+**Loading:** Skeletons matching layout dimensions. No spinners.
+
+**Empty:** Composed illustration + message + action button.
+
+**Nav:** Glass floating pill (low density), sidebar (medium/high), or top bar. Active state. Mobile collapse (not hamburger-only).
+
+### 2d. Layout
+- **Grid**: CSS Grid. No flexbox percentage math for multi-column.
+- **Container**: 1200-1440px max-width with auto margins.
+- **Section spacing**: `py-32 md:py-48` (low), `py-24` (medium), `py-16` (high).
+- **Responsive**: All multi-column → single below 768px. No exceptions.
+- **Full-height**: `min-h-[100dvh]`. Never `h-screen` (iOS Safari bug).
+- **BANNED**: Centered hero when VARIANCE > 4, 3-equal-card rows, edge-to-edge on wide screens.
+
+### 2e. Motion
+- **Spring**: `cubic-bezier(0.32, 0.72, 0, 1)` or Framer `spring(stiffness:100, damping:20)`. No linear.
+- **Scroll entry**: fade up `translateY(12-24px)` + opacity 0→1, 600-800ms. IntersectionObserver or CSS scroll-driven. No `window.addEventListener('scroll')`.
+- **Stagger**: Lists cascade via `animation-delay` or `staggerChildren`.
+- **Hover**: scale, translate, shadow, or color. 200-300ms.
+- **Active**: Every clickable gets `scale(0.98)` or `translateY(1px)`.
+- **Performance**: Only `transform` and `opacity`. No `top/left/width/height`. `will-change: transform` sparingly.
+- **Backdrop-blur**: Fixed/sticky elements only. Never scrolling containers.
+
+### 2e(ii). GSAP Motion (Creative/Expressive archetype only)
+
+When MOTION dial >= 7 or archetype is Creative/Expressive, use GSAP ScrollTrigger:
+- **Scroll Pinning** — pin a section title on the left while gallery scrolls on the right: `pin: true, anticipatePin: 1`
+- **Image Scale & Fade** — images start `scale(0.8)`, grow to `scale(1.0)` on enter, darken/fade to `opacity(0.2)` on exit
+- **Horizontal Scroll** — horizontal scroll section with pinned container, scrub trigger
+- **Staggered Reveal** — `stagger: 0.15` on child elements via `ScrollTrigger.batch()`
+- **Text Splitting** — split text into lines/chars, reveal each with stagger
+- **Spring Physics** — `cubic-bezier(0.32, 0.72, 0, 1)` or Framer `spring(stiffness:100, damping:20)`
+- **Python Randomization** — use deterministic seed (character count of prompt `%` math) to select hero architecture, typography stack, GSAP paradigm. Never default to same layout twice.
+- **Gapless Bento Grids** — use `grid-flow-dense` / `grid-auto-flow: dense`. Verify mathematically that `col-span`/`row-span` values interlock with no empty cells.
+
+## Phase 3: Build
+
+### 3a. Foundations
+CSS custom properties for colors, spacing, typography, shadows, radii. Tailwind config extensions mapping tokens to utilities. Theme provider. Component directory structure.
+
+### 3b. Component Library
+Implement ALL defined components with every state: default, hover, active, focus-visible, disabled, loading (skeleton), empty (illustration + action), error (inline). Performance: transform/opacity only, systemic z-index scale.
+
+### 3c. Pages
+Full interface from components. Responsive collapse at 768px. All viewport states (loading → populated → empty → error). Nav with active states and mobile collapse.
+
+### 3d. Requirements
+- Check `package.json` before importing — never assume a library exists
+- Framework-appropriate patterns (Server Components, island architecture)
+- Semantic HTML: `<nav>`, `<main>`, `<section>`, `<article>`, `<aside>`, `<header>`, `<footer>`
+- A11y: focus rings, skip-to-content, alt text, aria labels
+- Meta: `<title>`, description, `og:image`, viewport
+
+## Phase 4: Audit
+
+### Priority 1 (do first)
+- **Typography**: font matches spec? scale correct? tracking? no orphans? max-width?
+- **Color**: single accent? saturation < 80%? no AI purple? consistent? dark not pure black?
+- **Layout**: grid not flexbox math? `min-h-[100dvh]`? responsive at 768px?
+
+### Priority 2 (feel)
+- **Interactivity**: hover on all clickables? active feedback? focus rings? 200-300ms transitions?
+- **States**: every component has loading/empty/error? skeletons (not spinners)?
+- **Motion**: scroll entries? staggered? spring physics?
+
+### Priority 3 (content)
+- No lorem ipsum, cliches (Elevate, etc), generic names, emojis, bad icons?
+
+### Priority 4 (hardening)
+- Double-Bezel or appropriate card? button-in-button? nav active states?
+- Consistent icon stroke? semantic HTML? no inline styles?
+- 404 page? skip-to-content? meta tags? cookie consent?
+
+### Priority 5 (existing project redesign scan)
+- **Typography audit**: browser default fonts or Inter everywhere? Only Regular/Bold weights? Missing letter-spacing? All-caps subheaders everywhere? Orphaned words?
+- **Color audit**: pure `#000` background? Oversaturated accents? Mixing warm + cool grays? AI purple/blue gradient? Generic `box-shadow` (pure black tint)? No texture (pure flat)?
+- **Layout audit**: 3-equal-card rows? `height: 100vh` instead of `min-h-[100dvh]`? Complex flexbox percentage math? Everything centered and symmetrical?
+- **Surface audit**: flat sections with no visual depth? No background imagery? Sudden dark section in light page?
+- **Icon audit**: generic thin-line icon library? Rocket ship / shield cliches?
+
+### Phase 5: Iterate
+1. Fix in Priority order. Re-audit after each level.
+2. All P1-P3 pass → done. P4 surface as recommendations.
+3. Blocked on a check → narrow scope or surface.
+
+## Hard Bans
+
+- No emojis in code/content/alt/markup
+- No Lorem Ipsum — write real draft copy
+- No "Elevate", "Seamless", "Next-Gen", "Unleash", "Game-changer"
+- No generic names (John Doe, Acme Corp)
+- No rocket ship / shield cliche icons
+- No purple/blue neon gradients
+- No dark section in white page without committed dark mode
+- No animated `top/left/width/height`
+- No `window.addEventListener('scroll')`
+- No `h-screen`
+- No pill shapes on large containers, cards, or primary buttons
+- No generic icon libraries (Lucide, Feather, Heroicons)
+- No 3-equal-card feature rows — replace with 2-column zig-zag, asymmetric grid, or masonry
+- No `#000000` backgrounds — use off-black `#0a0a0a` or `#121212`
+- No even 45-degree linear gradients — break with radial, noise overlay, or mesh
+- No mixing warm and cool grays — stick to one gray family throughout
+- No random dark section in light page (or vice versa) — commit to one substrate
+- No orphaned words — use `text-wrap: balance` or `text-wrap: pretty`
+
+## Design Principles
+
+1. **Intentionality** — Commit to direction. Maximalism and minimalism both work if committed.
+2. **Engineering** — Buttons have structure, physics, a11y. Not just "style."
+3. **Consistency** — One palette, one font system, one architecture.
+4. **Performance** — Beautiful + laggy = not beautiful. Transform-only, guarded backdrop-blur.
+5. **Ship every state** — Default-only is not production.
+6. **Redesign first** — For existing projects, diagnose before prescribing. Improve in-place. A targeted upgrade beats a rewrite.