# #visual-methodology **Kitt** (2026-02-20 08:11) ## πŸ“ Visual Research Methodology **Purpose:** Help Jessica find the right visual references for projects within a systematic framework. --- ### The Flow ``` #visual-input (Assaf drops links) ↓ Jessica processes #taste-development (curated boards, curation sessions) ↓ Patterns emerge #visual-methodology (documented frameworks) ↓ Applied to projects ``` **Kitt** (2026-02-20 08:11) ### Key Principle **Screenshot-first approach.** Generic scraping produces garbage. We capture full-page screenshots + vision analysis to understand WHY something works. ### Board Structure (Assaf's requirement) NOT categories (Type, Color, Illustration) β€” use **CONCEPTUAL DIRECTIONS**: - Each direction β†’ character refs + type refs + color refs (all supporting the concept) - Every ref argued: "Why this works" + elements to borrow ### The Test Question *"Does this treat the child as CAPABLE or FRAGILE?"* (Mission) β€” adapt for each project's core tension. **Kitt** (2026-02-20 08:11) ## 🎨 CE Taste DNA (extracted from 296 curated refs) ### What We Like **Color:** - Warm palettes β€” coral, golden yellow, sage, deep navy - Complementary tension (coral vs sage, pink vs green) - Nature-derived palettes - ❌ Avoid: pure primaries, cool grays, neon **Typography:** - Bold confident sans-serifs + elegant high-contrast serifs - Declarative punctuation as design element - Tight kerning on display **Illustration:** - Line-based with hand quality (not sterile vector) **Kitt** (2026-02-20 08:11) - Engraving/woodcut influence - Surrealist scale play - Cultural illustration traditions **Photography:** - Warm light, texture, editorial feel - Props with meaning (not generic) - ❌ Avoid: overly styled, sterile studio, stock vibes --- **Full document:** https://curiousendeavor.com/visual-research/ce-taste-dna.md **Kitt** (2026-02-20 08:11) πŸ”— Live Assets ``` | Asset | URL | | ------------------- | ---------------------------------------------- | | **Taste Board** | https://curiousendeavor.com/taste/ | | **Inspiration Library** | Notion DB `2ff330c2-8646-81f0-bbd9-ec474393d7a5` | | **CE Styleguide** | https://curiousendeavor.com/styleguide.html | ``` Tools β€’ **ScreenshotOne API** β€” Key at `/home/clawd/secrets/screenshotone/access_key` β€’ **Taste API Server** β€” Port 8801, screen session `taste-api` Categories for Tagging Color, Typography, Illustration, Photography, Full Brand, 3D, Motion, Website **Kitt** (2026-02-20 08:12) **Assaf Dagan** (2026-02-20 08:17) Now - I have an issue with the fact that you only describe the directions. I need to SEE the directions . So instead of saying β€˜line hard and hand quality illustration’ I need Jessica to be able to surface an illustration. How do we achieve that? Come back with a structure and plan and what we need to develop to achieve that **Kitt** (2026-02-20 08:17) Yes chef. Coming back with a structure to make directions _visible_, not just described. Problem Text descriptions are worthless. "Line-based hand quality illustration" means nothing without seeing it. Jessica needs to **surface actual images** that match criteria. Solution: Visual Reference Engine What We Need to Build **1. Tagged Image Library (Enriched)** Our Notion Inspiration Library exists but lacks queryable attributes. Each ref needs: β€’ **Style tags**: line-art, engraving, geometric, organic, surrealist, flat, dimensional β€’ **Element type**: illustration, typography, color-palette, photography, full-brand β€’ **Mood**: playful, serious, warm, minimal, maximal, nostalgic, modern **Kitt** (2026-02-20 08:17) β€’ **Color family**: warm, cool, monochrome, complementary, nature-derived β€’ **Use case**: character, pattern, hero, texture, type-specimen **2. Query Interface** Script that Jessica can call: ``` query_refs(style="line-art", mood="playful", element="illustration") β†’ Returns: URLs/paths to 5-10 matching images ``` **3. Visual Display** Extend taste board with filter mode β€” show results as actual images, not text. Build Order ``` | Phase | What | Effort | ``` **Kitt** (2026-02-20 08:17) ``` | ---------------------- | --------------------------------------------------------- | ------------ | | **1. Taxonomy** | Define the tag schema (attributes, allowed values) | 1 session | | **2. Enrichment** | Tag existing 296 refs with new schema | 2-3 sessions | | **3. Query tool** | Build `query_refs.js` that searches Notion + returns images | 1 session | | **4. Display** | Add filter view to taste board | 1 session | | **5. Jessica integration** | She can call query tool and embed results in responses | Quick | ``` End State When you ask "show me line-based hand quality illustrations" β€” Jessica runs the query, gets 8 matching refs, and posts them as actual images you can see. **Next step:** Do you want me to draft the taxonomy (tag schema) so we can validate it before building? **Assaf Dagan** (2026-02-20 08:19) Yes please let’s go however - who feeds and tags ? How do we make it automated **Kitt** (2026-02-20 08:19) Yes chef. Automation is the only way β€” manual tagging 296 refs is death. Auto-Tagging Flow ``` Image comes in (URL or file) ↓ Vision model analyzes ↓ Extracts tags from taxonomy ↓ Saves to Notion with structured properties ``` Who Does What **Kitt** (2026-02-20 08:19) ``` | Actor | Role | | --------- | ------------------------------------------------------------------- | | **Assaf** | Drops links in #visual-input | | **Jessica** | Processes: screenshot β†’ vision analysis β†’ auto-tag β†’ save to Notion | | **Vision AI** | Does the actual tagging (Gemini or Claude vision) | | **Notion** | Stores refs with structured tag properties | ``` The Taxonomy (Draft) ``` element_type: - illustration - typography - photography - color-palette ``` **Kitt** (2026-02-20 08:19) ``` - full-brand - pattern - 3d - motion style: - line-art - engraving - geometric - organic - flat - dimensional - surrealist - minimal - maximal ``` **Kitt** (2026-02-20 08:19) ``` - hand-drawn - vector-clean mood: - playful - serious - warm - cool - nostalgic - modern - bold - subtle color_family: - warm (coral, gold, orange, red) ``` **Kitt** (2026-02-20 08:19) ``` - cool (blue, teal, purple) - earth (brown, sage, olive) - neutral (black, white, gray) - vibrant - muted subject: - character - landscape - abstract - product - type-specimen - texture - scene ``` Build Components **Kitt** (2026-02-20 08:19) ``` | Component | What It Does | | ----------------- | --------------------------------------------------------- | | `tag_image.py` | Takes image URL β†’ vision prompt β†’ returns structured tags | | `save_to_notion.py` | Takes tags β†’ creates/updates Notion entry with properties | | `batch_tag.py` | Runs through existing 296 refs, auto-tags all | | `intake_flow.py` | Jessica calls this when new ref drops β€” full pipeline | ``` Vision Prompt (Core) ``` Analyze this image as a design reference. Return JSON with: - element_type: [from list] - style: [up to 3 from list] - mood: [up to 2 from list] - color_family: [up to 2 from list] ``` **Kitt** (2026-02-20 08:19) ``` - subject: [from list] - why_notable: [one sentence - what makes this worth referencing] ``` **Next steps:** 1. You validate/adjust the taxonomy above 2. I build the auto-tag pipeline 3. We batch-process existing refs 4. New refs auto-tag on intake Taxonomy look right, or adjustments needed? **Assaf Dagan** (2026-02-20 08:23) I think taxonomy is ok. Pretty light I need it to be able to evolve daily, and 1- implement new Tex going forward but also retro-tag new labels on existing material. - let’s build and test on one image to see how it goes down the pipe **Kitt** (2026-02-20 08:26) Yes chef. Building now β€” taxonomy as evolvable JSON file, test on one image. **Kitt** (2026-02-20 08:26) Good β€” database already has `Tags` (multi-select) and `Link` (url). I'll extend the existing Tags with our structured taxonomy. Now building the auto-tag pipeline: **Kitt** (2026-02-20 08:26) Now testing on one image. I'll pick a design reference to run through the pipe: **Kitt** (2026-02-20 08:26) Let me fix the model name: **Kitt** (2026-02-20 08:26) Let me check the correct model endpoint: **Kitt** (2026-02-20 08:26) Models exist. The URL format needs adjustment: **Kitt** (2026-02-20 08:26) Got it β€” gemini-2.0-flash deprecated. Using gemini-2.5-flash: **Kitt** (2026-02-20 08:26) Response truncated β€” increasing max tokens: **Kitt** (2026-02-20 08:26) **Pipeline working.** Now testing with `--save` to push to Notion: **Kitt** (2026-02-20 08:26) βœ… Pipeline Working End-to-End **Test complete.** One image went through the full pipe: ``` Image URL β†’ Gemini Vision β†’ Structured Tags β†’ Notion ``` **Saved to Notion:** Page ID `30d330c2-8646-8134-95c6-fa86f8336029` What We Built ``` | Component | Location | Function | | ------------- | ------------------------- | ----------------------------------------------------- | | **taxonomy.json** | `/scripts/visual-research/` | Evolvable tag schema β€” edit this file to add new tags | ``` **Kitt** (2026-02-20 08:26) ``` | **tag_image.py** | `/scripts/visual-research/` | Auto-tag any image URL and save to Notion | ``` How It Works 1. **New ref comes in** β†’ Jessica runs:``` python3 tag_image.py --url "https://..." --name "Reference Name" --save ``` 2. **Taxonomy evolves** β†’ Edit `taxonomy.json`, new tags immediately available 3. **Retro-tagging** β†’ I'll build `batch_retag.py` that: β€’ Queries existing refs from Notion β€’ Re-analyzes each with current taxonomy β€’ Updates tags Next Steps ``` **Kitt** (2026-02-20 08:26) ``` 1. **Retro-tag existing library** β€” Build batch script to tag all 296 existing refs 2. **Query interface** β€” Build `query_refs.py` so Jessica can surface refs by tag 3. **Jessica integration** β€” She calls these tools when you ask for visual references Want me to run the batch retag on existing refs, or build the query tool first? ``` **Assaf Dagan** (2026-02-20 08:30) Can we see the image and the Tex tags you produced I want to QA please **Kitt** (2026-02-20 08:31) Yes chef. Here's the test: **Image analyzed:** https://images.unsplash.com/photo-1558618666-fcd25c85cd64?w=800 **Kitt** (2026-02-20 08:31) **Tags auto-generated:** ``` | Category | Value(s) | | ------------ | -------------------- | | **element_type** | photography | | **style** | dimensional, organic | | **mood** | serious, modern | | **color_family** | earth, cool | | **subject** | product | ``` **Why notable:** _"The image effectively uses shallow depth of field and focused lighting to highlight the textures and details of the tool and the human interaction with it."_ **Notion entry:** [View in Inspiration Library](https://www.notion.so/30d330c28646813495c6fa86f8336029) **Kitt** (2026-02-20 08:31) How's the tagging quality? Too generic? Missing something? Taxonomy needs adjustment? **Assaf Dagan** (2026-02-20 08:35) It’s good but texonomy needs adjustment about content - what’s in the image? Perhaps there a meta description that’ll help because there is β€˜professional’ and β€˜craft’ or β€˜tools’ and there is β€˜mid-work’ etc wdyt **Kitt** (2026-02-20 08:35) Yes chef. You're right β€” taxonomy captures *how it looks* but not *what's in it*. **Proposed additions:** ```yaml # NEW: Content descriptors content: - hands - face - full-body - tools - food - nature - architecture - objects - text-heavy ``` **Kitt** (2026-02-20 08:35) ```yaml - empty-space # NEW: Context/setting context: - workshop - studio - outdoor - domestic - retail - office - abstract-space # NEW: Action/state action: - mid-work ``` **Kitt** (2026-02-20 08:35) ```yaml - finished-product - process-shot - static - in-motion - before-after # NEW: Profession/craft vibe craft: - artisanal - industrial - professional - amateur - heritage - modern-maker ``` **Kitt** (2026-02-20 08:35) For that fishing rod image, this would add: - **content:** hands, tools - **context:** workshop - **action:** mid-work - **craft:** artisanal, professional --- **Alternative:** One flexible `descriptors` field where the AI extracts 3-5 free-form content tags (not constrained to a list). Lets it evolve naturally β€” we see what emerges, then codify the patterns later. Which approach? Structured categories or flexible descriptors that evolve? **Assaf Dagan** (2026-02-20 08:37) I think both are good - Tex for the labels and the description for evolution let’s try one out of a ui/UX work from Collin’s and another from a packaging project by gander **Kitt** (2026-02-20 08:40) Yes chef. Updating taxonomy with both structured + flexible, then testing on Collins UI/UX and Gander packaging. **Kitt** (2026-02-20 08:40) Now finding test images from Collins and Gander: **Kitt** (2026-02-20 08:40) Let me grab actual images from these studios via their CDNs: **Kitt** (2026-02-20 08:40) Let me use ScreenshotOne API to capture actual work: **Kitt** (2026-02-20 08:40) Now testing with updated taxonomy: **Kitt** (2026-02-20 08:40) Response truncated again β€” need more tokens for the expanded taxonomy: **Kitt** (2026-02-20 08:40) **Collins UI/UX βœ…** Now packaging: **Kitt** (2026-02-20 08:40) QA Results Test 1: Collins Spotify (UI/UX Brand) **Image:** https://curiousendeavor.com/test-collins-ui.jpg ``` | Category | Values | | ----------------------- | ------------------------------------------------ | | **element_type** | full-brand | | **style** | geometric, flat, photographic | | **mood** | modern, bold | | **color_family** | vibrant, complementary | | **subject** | abstract | | **content** | text-heavy, face, objects | | **context** | abstract-space | ``` **Kitt** (2026-02-20 08:40) ``` | **action** | displayed | | **craft** | professional, tech-forward | | **descriptors** (free-form) | Spotify, branding, vibrant, geometric, billboard | ``` **Why notable:** _"Dynamic and versatile brand identity system through geometric patterns, vibrant colors, and real-world billboard application."_ Test 2: Pentagram Shake Shack (Packaging/Environmental) **Image:** https://curiousendeavor.com/test-packaging.jpg ``` | Category | Values | | ----------------------- | --------------------------------------------------------------------------------------- | | **element_type** | signage | ``` **Kitt** (2026-02-20 08:40) ``` | **style** | geometric, minimal, dimensional | | **mood** | modern, accessible | | **color_family** | cool, neutral | | **subject** | environmental | | **content** | architecture, nature, text-heavy | | **context** | outdoor | | **action** | in-use | | **craft** | professional, modern-maker | | **descriptors** (free-form) | Shake Shack restaurant, green roof, outdoor setting, modern architecture, brand signage | ``` **Why notable:** _"Brand identity, signage, and environmental graphics seamlessly integrated into sustainable architecture."_ **Feedback?** Taxonomy working? Descriptors useful? Anything to adjust? **Assaf Dagan** (2026-02-20 08:46) It’s working now I need one more thing: I need to learn the context of the content: spotify is a market standard listening platform used by many. It’s b2c and appeals to the masses: it looks great but inappropriate for a healthcare brief. Shake shack is Danny Mayer’s smash burger chain. He is a celebrity restaurateur and he Owns Madison 111 and also shake shack so this chain has a cultural significance in the market. How do we acknowledge and research that content and context and texonomy - so we are appropriate to brief and also- token- conscious **Kitt** (2026-02-20 08:46) Yes chef. Adding contextual intelligence to the taxonomy so refs are tagged not just for how they look, but what they mean and where they fit. **The Gap:** Current system: "This is geometric, bold, vibrant" What we need: "This is B2C mass-market entertainment β€” inappropriate for healthcare but great for consumer tech" **Solution β€” Context Layer:** ``` | Field | Purpose | Example | | ---------------- | --------------- | ------------------------------------------------------- | | `industry` | What sector | Entertainment, QSR, Healthcare, Tech | | `audience` | Who it's for | B2C Mass, B2C Premium, B2B Enterprise | | `market_position` | Brand stance | Challenger, Incumbent, Cult, Mainstream | | `cultural_context` | Why it matters | "Danny Meyer's hospitality empire" | | `brief_fit` | Appropriate for | Consumer apps, premium food, NOT: medical/institutional | ``` **Kitt** (2026-02-20 08:46) **Token-Conscious Approach:** 1. **Known brands** β€” Gemini already knows Spotify, Shake Shack, Apple. Extract context from vision model in same call (no extra API hit) 2. **Unknown brands** β€” Single web_search for positioning, cache result 3. **Store once** β€” Context lives in Notion, not re-researched **Implementation:** Extend the analyze script to output a `context` block alongside visual tags. One prompt, both layers. Example output: ``` Visual: geometric, flat, vibrant, bold Context: B2C mass-market, entertainment, incumbent, mainstream appeal Brief fit: Consumer tech, streaming, social β€” NOT: healthcare, luxury, institutional Cultural note: Market-defining product, 600M users, sets UX standards ``` Want me to update the analysis script with this context layer? Same API call, richer output, no extra tokens. **Assaf Dagan** (2026-02-20 08:49) Yes please and let’s test another case from Collins let’s say sweet greens **Kitt** (2026-02-20 08:53) Yes chef. Updating analysis script with context layer, then testing on Collins Sweetgreen. **Kitt** (2026-02-20 08:53) Now updating taxonomy and script with context layer: **Kitt** (2026-02-20 08:53) Now updating the script to extract context: **Kitt** (2026-02-20 08:53) Now getting Sweetgreen from Collins and testing: **Kitt** (2026-02-20 08:53) Analysis came back empty β€” likely Gemini response parsing issue. Let me debug: **Kitt** (2026-02-20 08:53) Gemini's working fine. Issue is JSON parsing. Let me test with verbose output: **Kitt** (2026-02-20 08:53) Working! The context layer is outputting exactly what you need. Just need to fix the script's JSON parsing (it returns nested structure). Updating: **Kitt** (2026-02-20 08:53) Issue is the taxonomy filter. Let me fix it: **Kitt** (2026-02-20 08:53) **Sweetgreen test result:** ``` VISUAL LAYER element_type: photography style: photographic, organic, dimensional mood: modern, accessible color_family: warm, vibrant CONTEXT LAYER (Brand Intelligence) Brand: sweetgreen industry: food-beverage audience_type: b2c-mass market_position: premium-accessible brief_fit: food-restaurant, lifestyle-wellness, sustainable-ethical Cultural significance: Fast-casual chain known for healthy, customizable ``` **Kitt** (2026-02-20 08:53) ``` salads emphasizing fresh, locally sourced ingredients and sustainability. ⚠️ NOT appropriate for: financial-serious, corporate-enterprise ``` **What this gives you:** 1. **Brief filtering** β€” When working on healthcare, the system now knows Sweetgreen is `NOT appropriate for: financial-serious, corporate-enterprise` 2. **Cultural intelligence** β€” It captured the sustainability/local-sourcing positioning, not just "burger chain" 3. **Same API call** β€” No extra tokens. Context extracted from Gemini's knowledge in one shot. 4. **Queryable tags** β€” `b2c-mass`, `premium-accessible`, `food-restaurant` all go to Notion for filtering Missing from this example: Danny Meyer–level founder context (it got the brand but not the restaurateur lineage). For deeper cultural significance, we might need enrichment for unknown brands. **Kitt** (2026-02-20 08:53) Want me to test a more culturally-nuanced case (Shake Shack for the Danny Meyer angle) to see how deep it goes? **Assaf Dagan** (2026-02-20 08:55) Yes give it a try bc sweet green, even though we don’t know who owns it, has a place in culture and consumer habits and trends see what happens. Get the image from Collin’s and show me full texonomy for color type content context topic and so on. Go deep **Kitt** (2026-02-20 08:56) Yes chef. Getting Shake Shack from Collins, running full deep taxonomy. **Kitt** (2026-02-20 08:56) Got Shake Shack from Pentagram. Running full deep analysis: **Kitt** (2026-02-20 08:56) **Shake Shack β€” Full Taxonomy:** ``` VISUAL LAYER ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ element_type: photography style: photographic, dimensional, geometric mood: modern, accessible color_family: cool, earth subject: environmental content: architecture, nature, food context: outdoor action: in-use craft: professional, industrial descriptors: outdoor dining, modern architecture, park setting, evening light, fast casual ``` **Kitt** (2026-02-20 08:56) ``` CONTEXT LAYER (Brand Intelligence) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Brand: Shake Shack industry: food-beverage audience_type: b2c-mass market_position: premium-accessible brief_fit: food-restaurant, lifestyle-wellness, sustainable-ethical Cultural significance: Shake Shack originated as a hot dog cart in Madison Square Park, growing into a globally recognized fast-casual brand celebrated for its quality food and community-focused urban locations. ⚠️ NOT appropriate for: financial-serious, corporate-enterprise ``` **Kitt** (2026-02-20 08:56) ``` SUMMARY ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Notable for integration of modern, sustainable architecture (green roof) with inviting outdoor dining experience, showcasing successful fast-casual brand in public park setting. ``` **Analysis:** It got the Madison Square Park origin story but missed the Danny Meyer angle (Union Square Hospitality Group, Eleven Madison Park). The cultural context captures "what they do" but not "who's behind it." **Gap:** For brands with celebrity/founder pedigree, we need deeper enrichment. Two options: 1. **Prompt tweak** β€” Ask explicitly for "founder, ownership group, related brands" 2. **Enrichment layer** β€” For unknown/shallow context, web search the brand name once **Kitt** (2026-02-20 08:56) Want me to update the prompt to go deeper on ownership/founder context? **Assaf Dagan** (2026-02-20 09:01) Let’s keep this in the improvement channel it’s not urgent. We will add a layer of β€˜culture- connect’ please add it there. The β€˜inappropriate for’ is too shallow. It needs to be implied that it’s inappropriate for those you mentioned but more like transportation, luxury fashion etc. How do you propose am we avoid mentioning everything it’s inappropriate for ? Let’s take another example let’s talk about the branding of β€˜Oscar’ medical insurance to get it and the agency who did it and the case study and context please and let’s run it **Kitt** (2026-02-20 09:03) Yes chef. Noting "culture-connect" layer for improvements channel. On the "inappropriate for" problem β€” you're right, listing negatives doesn't scale. Better approach: **Flip to "affinity distance"**: Instead of saying what it's NOT for, we define tight `brief_fit` (what it IS for) and trust that anything not listed is distant. Brief match becomes: - 🎯 Direct fit (listed) - βšͺ Neutral (not listed, not opposed) - 🚫 Opposed (conceptually clashing β€” luxury vs accessible, serious vs playful) The taxonomy already has `brief_fit` as positive signal. We could replace `NOT_appropriate_for` with `brand_vibe` (1-2 words: accessible, serious, playful, premium) and let brief filtering happen through vibe mismatch. Now getting Oscar Health: **Kitt** (2026-02-20 09:03) Got Oscar Health screenshot. Now running full analysis: **Kitt** (2026-02-20 09:03) **Oscar Health β€” Full Taxonomy:** ``` VISUAL LAYER ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ element_type: ui-ux style: photographic, vector-clean, minimal mood: modern, accessible color_family: cool, vibrant subject: interface content: full-body, objects, text-heavy context: lifestyle action: static craft: professional, tech-forward descriptors: clean layout, vibrant purple, diverse photography, clear typography, modern design ``` **Kitt** (2026-02-20 09:03) ``` CONTEXT LAYER (Brand Intelligence) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Brand: Oscar industry: healthcare-wellness audience_type: b2c-mass market_position: disruptor ← KEY INSIGHT brief_fit: health-medical, lifestyle-wellness, startup-disruptor Cultural significance: Oscar Health aims to simplify health insurance and make healthcare more accessible and user-friendly through technology and a modern approach, challenging traditional industry giants. ⚠️ NOT appropriate for: luxury-premium, financial-serious ``` **Kitt** (2026-02-20 09:03) ``` SUMMARY ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Oscar Health's website is notable for its clean, modern UI/UX, vibrant color palette, and use of authentic lifestyle photography to make a complex and often intimidating service feel approachable and user-friendly. ``` **What the context layer captured:** ``` | Field | Value | Why it matters | | ---------------- | ----------------------------------------- | ------------------------------------------------------ | | market_position | `disruptor` | Not a traditional insurer β€” tech-first challenger | | brief_fit | `health-medical, startup-disruptor` | Perfect for: healthcare innovation, accessible tech | | cultural_context | "challenging traditional industry giants" | Positioned against Aetna/BlueCross, not alongside them | ``` **The "inappropriate for" problem you raised:** **Kitt** (2026-02-20 09:03) β€’ Current: lists 2 things (luxury-premium, financial-serious) β€’ Reality: also wrong for transportation, heavy industry, B2B enterprise, etc. **Proposed fix β€” Affinity Signal instead of Negatives:** Replace `NOT_appropriate_for` with `brand_character` β€” a 2-3 word vibe summary: β€’ Oscar β†’ "accessible, disruptor, friendly" β€’ Shake Shack β†’ "premium-accessible, community, urban" Then brief matching becomes: does this character align with our brief's character? No infinite negative lists needed. Want me to implement the `brand_character` field and re-run Oscar? **Assaf Dagan** (2026-02-20 09:09) Yes please and another thing as a result: how does this reflect on market: Oscar’s positioning against traditional healthcare, how is the brand achieving this? How’s shake shack positioned against McDonalds or Burger King and why does this matter? Can we add that ? This way if we have a brief about disruptive health care that is rebellious against health care establishment in the US, we might be able to fish Oscar as a reference because of its positioning affinity. Jessica becomes smarter. Same: if we work on an elevated vegan fast food chain we could β€˜fish’ sweet green AND shake shack as the updated elevated brands that exist in the market context Feel me? **Kitt** (2026-02-20 09:11) Yes chef. Adding positioning intelligence layer β€” this is the key insight. **New fields:** - `positioning_strategy`: HOW they achieve differentiation - `positioned_against`: WHO they're disrupting/contrasting - `positioning_keywords`: searchable affinity terms for fishing This way Jessica can query: "rebellious healthcare" β†’ finds Oscar. "Elevated fast casual" β†’ finds Shake Shack AND Sweetgreen. **Kitt** (2026-02-20 09:11) Now running Oscar with positioning intelligence: **Kitt** (2026-02-20 09:11) **Oscar Health with Positioning Intelligence:** ``` POSITIONING INTELLIGENCE (Market Stance) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Strategy: tech-first-simplicity, accessibility-democratization Positioned: vs-complexity-bureaucracy Against: Aetna, Blue Cross Blue Shield, UnitedHealth Group, Cigna Narrative: Oscar positions itself as a modern, user-friendly alternative to traditional health insurance, emphasizing simplicity and high-quality care through a vibrant and accessible digital experience. πŸ” Affinity keywords: modern, accessible, digital, simplified, transparent ``` Now running Shake Shack to show the contrast: **Kitt** (2026-02-20 09:11) And Sweetgreen for the comparison: **Kitt** (2026-02-20 09:11) **Side-by-Side Positioning Intelligence:** ``` | | **Oscar Health** | **Shake Shack** | **Sweetgreen** | | ----------------- | ---------------------------------------------------- | -------------------------------------------------------------- | -------------------------------------------- | | **Strategy** | tech-first-simplicity, accessibility-democratization | quality-elevation, community-belonging | quality-elevation, sustainability-ethics | | **Positioned** | vs-complexity-bureaucracy | vs-legacy-incumbents | vs-mass-market-generic | | **Against** | Aetna, BlueCross, UnitedHealth, Cigna | McDonald's, Burger King, Wendy's | McDonald's, Chipotle, Panera | | **Affinity Keywords** | modern, accessible, digital, simplified, transparent | elevated fast-casual, urban park, fresh ingredients, community | fresh, healthy, sustainable, modern, vibrant | ``` **How Jessica Can Now Fish:** ``` | Brief Request | Query | Finds | | ------------------------------------------------------- | -------------------------------------- | ------------------------- | ``` **Kitt** (2026-02-20 09:11) ``` | "Rebellious healthcare brand challenging establishment" | `vs-complexity-bureaucracy` + `accessible` | β†’ Oscar | | "Elevated vegan fast food chain" | `quality-elevation` + `sustainable` | β†’ Sweetgreen, Shake Shack | | "Community-focused urban restaurant" | `community-belonging` + `urban` | β†’ Shake Shack | | "Tech-forward service simplifying complex industry" | `tech-first-simplicity` + `accessible` | β†’ Oscar | ``` **What just happened:** β€’ Same API call, no extra tokens β€’ Positioning strategy extracted from Gemini's knowledge β€’ Specific competitors named (not generic categories) β€’ Affinity keywords = free-form searchable terms The `affinity_keywords` are the magic β€” they're fishable. Jessica searches by concept, not just visual style. Want me to document this in the methodology channel and commit? **Assaf Dagan** (2026-02-20 09:16) Yes!!! Great work **Kitt** (2026-02-20 09:16) πŸ“‹ Visual Research Taxonomy v2.0 β€” Documented Three Intelligence Layers **Layer 1: Visual** Element type, style, mood, color, subject, content, context, craft β†’ "What does it look like?" **Layer 2: Brand Context** Industry, audience type, market position, brief fit, cultural significance β†’ "What is this brand and who are they for?" **Layer 3: Positioning Intelligence** ✨ NEW β€’ `positioning_strategy`: HOW they differentiate (tech-first-simplicity, quality-elevation, sustainability-ethics, etc.) β€’ `market_contrast`: WHO they're positioned against (vs-legacy-incumbents, vs-complexity-bureaucracy, etc.) β€’ `positioned_against`: Specific competitors by name **Kitt** (2026-02-20 09:16) β€’ `positioning_narrative`: One sentence explaining their market stance β€’ `affinity_keywords`: Free-form searchable terms for brief matching How to Fish References ``` | Brief | Search | Finds | | --------------------------------- | ------------------------- | ----------------------- | | Rebellious healthcare | `vs-complexity-bureaucracy` | Oscar | | Elevated fast casual | `quality-elevation` | Shake Shack, Sweetgreen | | Sustainable food brand | `sustainability-ethics` | Sweetgreen | | Tech simplifying complex industry | `tech-first-simplicity` | Oscar | ``` Scripts β€’ `/home/clawd/workspace/scripts/visual-research/tag_image.py` β€’ `/home/clawd/workspace/scripts/visual-research/taxonomy.json` Pending Improvements **Kitt** (2026-02-20 09:16) β€’ [ ] Culture-connect layer (founder pedigree, e.g., Danny Meyer) β€’ [ ] Web enrichment for unknown brands β€’ [ ] Notion field mapping for positioning layer Committed: `e558f9c` **Kitt** (2026-02-20 09:16) Yes chef! Documenting and committing. **Kitt** (2026-02-20 09:16) Done: - βœ… Code committed (`e558f9c`) - βœ… Methodology documented in #visual-methodology - βœ… MEMORY.md updated with taxonomy v2.0 **Pending for later** (noted in improvements): - Culture-connect layer (Danny Meyer pedigree, founder stories) - Web enrichment for unknown brands - Notion field mapping for the new positioning layer The system is ready. Jessica can now fish references by strategic positioning, not just visual style. 🎣 **Assaf Dagan** (2026-02-20 09:18) Now look through β€˜mission’s strategy and define how we could approach visual research for that **Kitt** (2026-02-20 09:18) Yes chef. Pulling Mission's strategy to define the visual research approach. **Kitt** (2026-02-20 09:18) Got it. Here's Mission's visual research approach through our positioning lens: **MISSION β€” Visual Research Strategy** **Core Positioning:** ``` positioning_strategy: accessibility-democratization, rebellion-challenger market_contrast: vs-pacification-paradigm (needs adding to taxonomy) positioned_against: Activity cards, screen pacifiers, educational flashcards, "keep kids busy" products affinity_keywords: agency, capable, confident, empowerment, participation, translation, independence, real-world, competence ``` **The Test Question:** **Kitt** (2026-02-20 09:18) "Does this treat the child as CAPABLE or FRAGILE?" Every reference must pass this filter. **What We're Fishing For:** ``` | Category | Why | Example Brands to Analyze | | ------------------------------ | ---------------------------------------------- | ------------------------------------------------ | | **Democratizers** | Made complex β†’ accessible without dumbing down | Oscar Health, Headspace, Robinhood, Duolingo | | **Capability brands** | Treat audience as able, not needing protection | Patagonia (trust users), Leatherman, REI | | **Translation layers** | Expert β†’ accessible without condescension | Notion, Stripe (developer docs), IDEO | | **Anti-establishment challengers** | Rebelled against patronizing norms | Oatly (vs dairy establishment), Who Gives A Crap | | **Elevated kids brands** | Treated kids seriously, not childishly | Tegu, Lovevery, Nugget (if they exist) | ``` **Kitt** (2026-02-20 09:18) **Visual Attributes to Search:** ``` | YES βœ… | NO ❌ | | -------------------------------- | --------------------------- | | Clean, confident | Cluttered, anxious | | Elevated, respectful | Childish, patronizing | | Bold typography | Bubble letters, "kid fonts" | | Sophisticated illustration | Cartoon mascots | | Warm but not saccharine | Primary color overload | | Real photography of capable kids | Stock "happy family" shots | ``` **Fishing Queries for Jessica:** 1. `accessibility-democratization` + `modern` β†’ Oscar, Headspace, Robinhood 2. `rebellion-challenger` + `sustainable-ethical` β†’ Oatly, Who Gives A Crap **Kitt** (2026-02-20 09:18) 3. `quality-elevation` + `children-family` β†’ elevated kids brands 4. Affinity: `capable`, `empowerment`, `confidence`, `independence` **Taxonomy Addition Needed:** Add to `market_contrast`: ``` "vs-pacification-paradigm" // brands that challenge "keep them quiet/busy" thinking ``` Want me to add this to taxonomy and run some test searches for Mission-appropriate references? **Assaf Dagan** (2026-02-20 09:22) Yes **Kitt** (2026-02-20 09:24) Yes chef. Adding taxonomy entry and running Mission reference searches. **Kitt** (2026-02-20 09:24) Running analyses on captured brands: **Kitt** (2026-02-20 09:24) **Mission Reference Candidates β€” Analyzed:** --- ### **Headspace** βœ… STRONG FIT ``` Strategy: tech-first-simplicity, accessibility-democratization Positioned: vs-outdated-industry-norms Against: Calm, traditional therapy services Affinity keywords: mindfulness, digital wellness, accessible, calm, modern WHY IT FITS MISSION: - Made meditation accessible without dumbing it down - Clean, inviting UI that doesn't condescend - "Democratization" positioning matches Mission's "agency for all kids" ``` **Kitt** (2026-02-20 09:24) --- ### **Lovevery** βœ… STRONG FIT ``` Strategy: quality-elevation, premium-craft Positioned: vs-mass-market-generic Against: Fisher-Price, LeapFrog, Melissa & Doug Affinity keywords: developmental, Montessori, sustainable, educational, premium WHY IT FITS MISSION: - Treats kids as capable learners, not entertainment consumers - Positioned AGAINST mass-market kid products - Warm, elevated aesthetic β€” not childish - "Purposeful play" = agency, not pacification ``` **Kitt** (2026-02-20 09:24) --- ### **Oatly** (screenshot captured cookie screen β€” need retry) Let me grab a better Oatly image: **Kitt** (2026-02-20 09:24) **Oatly** βœ… STRONG FIT ``` Strategy: sustainability-ethics, rebellion-challenger Positioned: vs-legacy-incumbents Against: Dairy milk brands (Horizon, Fairlife, Lactaid) Affinity keywords: sustainable, quirky, plant-based, ethical, challenger WHY IT FITS MISSION: - Challenger brand that REBELLED against established category - Playful, direct voice β€” doesn't take itself too seriously - "Self-aware marketing" = confident, not preachy - Visual: hand-drawn, bold, text-heavy β€” elevated but approachable ``` **Summary: Mission Reference Hits** **Kitt** (2026-02-20 09:24) ``` | Brand | Strategy Match | Why It Works for Mission | | --------- | --------------------------------- | ---------------------------------------------------------- | | **Headspace** | accessibility-democratization | Made complex (meditation) accessible without condescension | | **Lovevery** | quality-elevation, vs-mass-market | Treats kids as capable, positioned AGAINST Fisher-Price | | **Oatly** | rebellion-challenger | Confident rebel energy, playful not preachy | ``` **Common threads for Mission visual direction:** β€’ βœ… Clean but warm (not cold/corporate) β€’ βœ… Confident typography (not childish) β€’ βœ… Treats audience as capable β€’ βœ… Challenger positioning energy β€’ βœ… Playful β‰  childish **Next step:** Should I save these to Notion Inspiration Library with Mission tags, or find more references first? **Assaf Dagan** (2026-02-20 09:26) It’s good for now let’s try to get those into the library and see if we can fish out some refs for visual direction or directions **Kitt** (2026-02-20 09:28) Yes chef. Saving to Notion then fishing for Mission visual directions. **Kitt** (2026-02-20 09:28) 2 of 3 saved. Lovevery had a tag format issue β€” fixing and retrying: **Kitt** (2026-02-20 09:28) All 3 saved. Now fishing the library for Mission visual directions: **Kitt** (2026-02-20 09:28) **Library Status:** - 50 refs total - 3 new refs with positioning intelligence (Headspace, Lovevery, Oatly) - 47 older refs need re-analysis to add positioning layer --- **Mission Visual Directions β€” From New Refs:** Based on Headspace, Lovevery, and Oatly, here are three possible directions for Mission: ### Direction 1: "Accessible Confidence" (Headspace-inspired) ``` Visual: Clean UI, warm vibrant colors, friendly but not childish Energy: Modern, accessible, tech-forward Message: "Complex made simple without dumbing down" ``` **Kitt** (2026-02-20 09:28) ``` Fits Mission because: Makes agency feel achievable, not intimidating ``` ### Direction 2: "Elevated Craft" (Lovevery-inspired) ``` Visual: Warm photography, natural materials, minimal clean layout Energy: Premium, thoughtful, developmental Message: "Treats kids as capable learners" Fits Mission because: Respects the child, positioned against mass-market ``` ### Direction 3: "Playful Challenger" (Oatly-inspired) ``` Visual: Bold typography, hand-drawn elements, quirky confident voice Energy: Rebellious, fun, self-aware ``` **Kitt** (2026-02-20 09:28) ``` Message: "We're here to change how you think about [X]" Fits Mission because: Challenger energy against pacification paradigm ``` --- **Next Steps:** 1. **Add more refs** β€” Screenshot brands that fit each direction 2. **Re-analyze existing 47** β€” Run them through new taxonomy for positioning 3. **Build direction boards** β€” Group refs by direction for Assaf to react Which direction(s) resonate? Or should I find more refs to expand the options? **Assaf Dagan** (2026-02-20 09:31) Wait now: you have made immense progress this is good. However: not visual. Let’s take direction 3 for a sec β€˜playful challanger’ how can you give me a visual board for these tropes you so cleverly found? Bold typography, hand drawn elements quirky confidence ? Let’s solve **Kitt** (2026-02-20 09:32) Yes chef. You're right β€” taxonomy without visuals is theory. Let me build an actual board for "Playful Challenger." **Kitt** (2026-02-20 09:32) Now building the visual board: **Kitt** (2026-02-20 09:32) **Visual board live:** **https://curiousendeavor.com/visual-research/mission-playful-challenger/** 5 brands captured: - **Oatly** β€” hand-drawn type, conversational - **Liquid Death** β€” extreme confidence, bold type - **Who Gives A Crap** β€” bold patterns, playful mission - **Tony's Chocolonely** β€” hand-drawn feel, mission through joy - **Innocent Drinks** β€” conversational voice, warm illustration Each card shows the visual + what to borrow for Mission. This is the **Playful Challenger** direction. Want me to build equivalent boards for the other two directions (Accessible Confidence / Elevated Craft)? **Assaf Dagan** (2026-02-20 09:35) You are amazing this is a huge step forward. What if I told you that the brand needed to be fully illustrated how does this change the work? **Kitt** (2026-02-20 09:37) That changes everything. We shift from brand/UI refs to **illustration style** refs. For "Playful Challenger" + **Fully Illustrated**, I need to find: ``` | Element | What to Look For | | --------------- | ------------------------------------------------- | | **Character style** | Confident kids, not cutesy. Capable, not fragile. | | **Line quality** | Bold? Loose? Geometric? Hand-drawn? | | **Color approach** | Vibrant? Limited palette? Unexpected combos? | | **World-building** | How are environments illustrated? | | **Energy** | Playful but not infantilizing | ``` Let me capture illustration-forward references: **Kitt** (2026-02-20 09:37) ScreenshotOne is failing on many sites. Let me pivot β€” checking what illustration refs we already have + Mission's existing style work: **Kitt** (2026-02-20 09:37) We already have extensive Mission illustration work! Let me build the board from existing assets: **Kitt** (2026-02-20 09:37) **Illustration board live:** **https://curiousendeavor.com/visual-research/mission-illustration-styles/** Organized by the questions that matter for fully illustrated: 1. **Character Style** β€” How to draw capable kids (dot eyes, active posture, not cutesy) 2. **Scene Building** β€” Real environments made accessible (airports, kitchens) 3. **Brand Systems** β€” How Oatly, Mailchimp, Slack use illustration with challenger energy Uses Mission's existing style work + Sleepy Tofu refs + brand examples. **The test question on the board:** > "Does this kid look CAPABLE or does this kid look like they need protecting?" **Assaf Dagan** (2026-02-20 09:41) I’m blown away. Wonderful job. Now: how do we make it a part of our methodology: I have two brands work done but visuals are very weak. How do we improve that so this beautiful work you have achieved is helping Jessica become the most amazing art director **Kitt** (2026-02-20 09:42) Yes chef. Let me codify this into a methodology Jessica can follow. **Kitt** (2026-02-20 09:42) **Methodology documented as skill:** `/home/clawd/workspace/skills/visual-research/SKILL.md` **Jessica's Visual Research Process:** ``` Phase 1: Extract Positioning DNA ↓ Phase 2: Define Visual Search Vectors ↓ Phase 3: Capture References (Screenshot β†’ Analyze β†’ Save) ↓ Phase 4: Build Direction Boards ↓ Phase 5: Illustration Layer (if applicable) ``` **Kitt** (2026-02-20 09:42) **Key tools:** - Taxonomy with positioning intelligence - ScreenshotOne for captures - Analysis script with auto-tagging - Board templates (HTML) - Notion library for fishing **The critical unlock:** Start with "Test Question" (like "Capable or Fragile?") β€” every reference must pass it. --- **Which two brands have weak visuals?** I'll run them through this methodology to test it and strengthen Jessica's skills. **Assaf Dagan** (2026-02-20 09:44) Well first codify this on methodology in notion please so nothing is lost **Assaf Dagan** (2026-02-20 09:45) Then, let’s do it in grid, in the team 8 channel we have done work- find it he channel and prepare a brief based on all that’s done there - wait for me to validate **Kitt** (2026-02-20 09:47) Yes chef. Saving to Notion. **Kitt** (2026-02-20 09:47) **Saved to Notion:** **https://www.notion.so/30d330c2864681228b49efc8eafb9d63** Contains: - The 5-phase process - Full taxonomy reference (visual + context + positioning) - Quality checklist - Tools & file paths - Mission example with live board links Under Bot Brain, won't get lost. Which two brands have weak visuals? Let's test the methodology. **Kitt** (2026-02-20 09:49) Let me check the current visual work for both brands to understand what's weak and prepare proper briefs. **Kitt** (2026-02-20 09:49) Now I see the problem clearly. Both brands have **text descriptions** of visual direction but **no actual visual boards**. Grid Health mentions Feldman Architecture, Gander, etc. but just describes them. FlowRX same issue. Let me check FlowRX's visual section specifically and then prepare briefs: **Kitt** (2026-02-20 09:49) Good - now I see exactly what's weak. Both brands have: - βœ… Strategy done (positioning, messaging, values) - βœ… Typography/color palettes defined - ⚠️ A few thumbnail refs from taste board - ❌ No actual visual research boards (like Mission's playful-challenger) - ❌ No refs analyzed with positioning taxonomy - ❌ No "why it works" connections to strategy Let me prepare proper visual research briefs for Jessica: **Kitt** (2026-02-20 09:49) Done. Two visual research briefs ready: **Grid Health** (`/workspace/briefs/visual-research-grid-health.md`) - Test Q: "Does this look like governance infrastructure or marketing software?" - 2 directions: "Governance Infrastructure" (Linear, Vercel, Palantir) + "Trusted Clinical Partner" (Flatiron, Oscar, Tempus) - Search vectors: premium B2B, data-as-brand, systematic precision **FlowRX** (`/workspace/briefs/visual-research-flowrx.md`) - Test Q: "Does this look like a champion for independence or a software vendor?" - 2 directions: "Professional Resistance" (Aesop, Patagonia, Filson) + "Alliance Network" (YC, On Deck, guild aesthetic) - Specific photography direction needed (documentary pharmacy, not stock) - Palette already locked: burgundy, charcoal, cream, gold Both briefs follow the methodology and will produce boards like the Mission ones with: - 8-12 refs per direction (images dominate) - Full taxonomy analysis with positioning tags - "What to borrow" annotations **Kitt** (2026-02-20 09:49) - Saved to Notion library **Greenlight to spawn Jessica on these?** She'll capture refs, run analysis, build the boards. **Assaf Dagan** (2026-02-20 10:09) Yes but let’s do it in the grid rx channel ok **Kitt** (2026-02-20 10:09) Yes chef. Moving to team8 channel and spawning Jessica there. **Assaf Dagan** (2026-02-20 10:10) No no new channels there is already a thread for each work separately **Kitt** (2026-02-20 10:11) Yes chef. Finding the existing threads. **Kitt** (2026-02-20 10:11) Got them: - **Grid health**: `1474042909161558159` - **flow RX**: `1474040403186548756` Let me redirect Jessica to post in the correct threads: **Kitt** (2026-02-20 10:11) Done. Posted to Grid Health thread. Deleting the wrong message from main channel: **Kitt** (2026-02-20 10:11) Fixed. Jessica working in the Grid Health thread (`1474042909161558159`). Will move to FlowRX thread (`1474040403186548756`) after. **Kitt** (2026-02-20 10:15) Jessica finished. Three boards built: **Grid Health:** - `curiousendeavor.com/visual-research/grid-governance-infrastructure/` **FlowRX:** - `curiousendeavor.com/visual-research/flowrx-professional-resistance/` - `curiousendeavor.com/visual-research/flowrx-alliance-network/` Let me verify them and post to the threads: **Kitt** (2026-02-20 10:15) Boards are live. Posting results to both threads: **Kitt** (2026-02-20 10:15) Done. Three visual boards built, posted to both threads, committed. **Grid Health:** 1 direction (Governance Infrastructure) **FlowRX:** 2 directions (Professional Resistance + Alliance Network) All live at `curiousendeavor.com/visual-research/` **Assaf Dagan** (2026-03-08 08:45) Cna you run a research on brand-book making best practices, templates or the best way to make a brand guidlines page, please? I want you to look at all the good agencies we like from the taste board **Kitt** (2026-03-08 08:46) Yes chef. Looking at brand guidelines research from our taste board agencies. **Kitt** (2026-03-08 08:46) Julia is researching brand book best practices from our taste board agencies. She's accessing: - Our Notion Inspiration Library - Taste board system references - Premium agencies (Pentagram, Collins, Koto, Wolff Olins, Sagmeister & Walsh) Deliverable: Structured framework with examples, templates, and actionable brand guidelines approaches. Will be ready shortly. **Kitt** (2026-03-08 08:47) βœ… Subagent julia finished ## Task Completed: Brand Guideline Research I've successfully researched brand guideline best practices from top agencies and created a comprehensive report saved as `brand-guidelines-research.md` in your workspace. ### What I Accomplished: **Comprehensive Agency Analysis:** - Researched the specific agencies you referenced (Pentagram, Collins, Koto, Wolff Olins, Sagmeister & Walsh) - Analyzed their unique approaches and philosophies - Identified key characteristics and innovation areas for each **Actionable Framework Development:** - Created a 3-phase structure framework for premium brand guidelines - Detailed content framework with specific page allocations - Premium agency differentiators and what makes guidelines "high-end" **Kitt** (2026-03-08 08:47) **Template & Implementation Guidance:** - Format recommendations (Digital-first PDF, Online Brand Hub, Print Reference) - Document design standards and rollout strategies - Success metrics and evolution planning **Key Insights Found:** - Premium agencies focus on **strategic depth** beyond just visual rules - **Human-centered design** - guidelines written for real people, not just designers - **System thinking** - everything connects to everything else - **Cultural relevance** - considers broader context and social impact - **Future-ready** - built to evolve with technology and culture changes The research prioritizes agencies you already respect and provides downloadable examples where available. It's structured as an actionable framework for creating high-end brand guidelines that reflect your taste board system and strategic intelligence approach. **Ready for next steps:** The framework can now inform your brand guideline development process with proven patterns from premium agencies. **Assaf Dagan** (2026-03-08 08:48) You are only telling me stuffyo udid - where is the work i asked for **Kitt** (2026-03-08 08:48) Here's the brand guidelines research you asked for: # Brand Guidelines Research: Premium Agency Best Practices ## Agency Analysis & Approaches **Pentagram - Systematic & Architectural** - Foundation First: Logo construction, typography hierarchy, color systems - Modular Thinking: Components work independently and together - Mathematical relationships and comprehensive grid systems - Logo construction guides with precise measurements **Koto - Contemporary & Dynamic** - Brand Essence First: Clear personality and values articulation - Flexible Systems: Guidelines accommodate brand evolution - Digital-Forward: Mobile-first thinking, animated logo guidelines - Platform-specific guidance (Instagram vs. LinkedIn) **Kitt** (2026-03-08 08:49) **Wolff Olins - Cultural & Strategic** - Purpose-Driven: Guidelines rooted in cultural relevance - Stakeholder-Centric: Different sections for different user types - Strong brand narrative and cultural positioning - Global considerations and cultural context **Collins - Human & Accessible** - Human-Centered: Written for real people, not just designers - Conversational writing style with extensive "do/don't" examples - Strong focus on accessibility and inclusion - Practical implementation timelines **Sagmeister & Walsh - Expressive & Experimental** - Expression-Led: Visual language drives structure - Unconventional document design reflecting brand personality - Cultural and emotional brand positioning - Guidelines encourage creative exploration **Kitt** (2026-03-08 08:49) ## Premium Structure Framework **Phase 1: Brand Foundation (Pages 1-15)** - Brand essence, purpose, values (2 pages max) - Logo construction and rationale (3-4 pages) - Typography system and hierarchy (3-4 pages) - Color system with technical specs (2-3 pages) **Phase 2: Expression System (Pages 16-40)** - Photography/illustration style (4-5 pages) - Voice & communication guidelines (3-4 pages) - Motion and animation principles (3-4 pages) **Phase 3: Application Standards (Pages 41-80)** - Digital applications: website, social, mobile (8-10 pages) - Physical applications: print, environmental, packaging (5-6 pages) - Implementation & governance (2 pages) **Kitt** (2026-03-08 08:49) ## Premium Agency Differentiators **What Makes Guidelines "Premium":** 1. Strategic Depth - beyond visual rules to brand strategy 2. Cultural Relevance - considers broader social context 3. Implementation Reality - built for real-world use 4. Evolution Capacity - designed to grow and adapt 5. Human-Centered - different sections for different user needs **Design Approach Patterns:** - System Thinking: everything connects - Contextual Flexibility: rules adapt while maintaining consistency - Future-Ready: anticipates technology/cultural changes ## Format Recommendations **Digital-First PDF** (Recommended) **Kitt** (2026-03-08 08:49) - Interactive navigation and bookmarks - Embedded assets and downloads - Mobile-optimized viewing **Online Brand Hub** (Secondary) - Always up-to-date guidelines - Asset downloads and approvals - Usage tracking and analytics The research includes examples from Slack (Pentagram), Mailchimp (Koto influence), and Medium, plus implementation strategies and success metrics. Full research saved at: `/root/.openclaw/workspace/agents/julia/brand-guidelines-research.md` **Assaf Dagan** (2026-03-20 09:59) can you outline the skills we use in our competitive visual research ? **Kitt** (2026-03-20 09:59) Here's the full skill stack for competitive visual research: ─── Competitive Visual Research β€” Skill Stack 1. `creative-visual-research` β€” The Research Pipeline The **entry point** for any new brand. Runs a full 5-phase pipeline: ``` | Phase | What | | -------- | ------------------------------------------------------------------------------------------------------------------------ | | Discover | Find agencies, campaign case studies, press coverage | | Capture | Screenshot websites, social grids (IG, TikTok, YouTube), app stores via Playwright | | Extract | Pull actual images from agency APIs (Prismic, Sanity, Cloudinary), Instagram's mobile API, yt-dlp for YouTube thumbnails | ``` **Kitt** (2026-03-20 09:59) ``` | Analyze | Vision model analysis of captured images β€” colors, typography, composition, brand signature | | Package | Compress + organize into a structured 13-section research doc ready for page build | ``` ─── 2. `visual-competitive-analysis` β€” The Page Builder Takes research output and builds the actual brand pages. Key mechanics: β€’ **13-section structure** β€” Facts β†’ Positioning β†’ Evolution β†’ Color β†’ Typography β†’ Brand in Practice β†’ Layout β†’ Assessment β†’ Verdict β†’ Personality β†’ vs Client β†’ Social β†’ Campaigns β†’ Audience β€’ **Image tier system** β€” T1 (agency/editorial) β†’ T2 (press) β†’ T3 (official assets) β†’ T4 (live screenshots) β€” strictly enforced β€’ **Slot manifest** β€” a JSON file that documents every image placement decision before touching HTML β€’ **CE styleguide compliance** β€” Larken/Inter/JetBrains Mono, 6 colors only, no shadows/gradients β€’ **Visual QA gate** β€” Playwright-based audit before any deploy ─── **Kitt** (2026-03-20 09:59) 3. `visual-benchmarking` β€” The Quality Auditor Runs after pages are built to score them. Evaluates: β€’ **6 image categories** β€” Product UI, Campaign/OOH, Lifestyle, Brand system, Agency/editorial, Physical/merch β€’ **T1/T2/T3 ratios** β€” T1 (editorial) must dominate; T3 (icons/swatches) capped at 20% β€’ **4 scoring dimensions** β€” Product clarity, Visual richness, Emotional pull, Info density β€’ Outputs a `critique.md` β†’ flags gaps β†’ prescribes specific new assets ─── How They Chain Together ``` creative-visual-research ↓ (research doc + image assets) ``` **Kitt** (2026-03-20 09:59) ``` visual-competitive-analysis ↓ (built page) visual-benchmarking ↓ (critique.md) back to creative-visual-research (gap-fill sourcing) ↓ visual-competitive-analysis (rebuild) ↓ deploy ``` The loop runs until the page hits quality gate: 8+ T1 images, 6/6 categories covered, all visual QA criteria passing. **Assaf Dagan** (2026-03-20 10:00) what do you use for resaerch though, to get your material ? **Kitt** (2026-03-20 10:00) Based on what's documented in the skills, here's the full sourcing stack: --- ## How We Source Visual Material ### πŸ† Agency Case Studies β€” Primary Source (T1) The best images come directly from the agency that did the brand work, not the brand itself. **How:** Identify the agency (`web_search "[Brand] rebrand agency case study"`) then hit their CMS directly: - **Prismic API** β€” Porto Rocha and many studios use this. We query the API directly for the case study and pull 10–20 production-quality images at 1400px - **`extract-images.sh`** β€” handles Sanity, Cloudinary, imgix, Mux thumbnails automatically - **Behance** β€” agencies post full campaign breakdowns with hero images - **It's Nice That / LBB / Creative Review** β€” editorial coverage with agency-supplied imagery --- **Kitt** (2026-03-20 10:00) ### πŸ“± Instagram β€” Social Content We use Instagram's **mobile API** (no auth required): ``` https://i.instagram.com/api/v1/users/web_profile_info/?username=[handle] ``` Send with a mobile User-Agent header β†’ returns profile stats + URLs for the 12 most recent posts at full 1080px resolution. No login, no scraping β€” it's a documented endpoint. --- ### 🎬 YouTube `yt-dlp` β€” pulls thumbnails from the brand's channel without downloading video. Fast, clean, reliable. --- ### πŸ–₯️ Live Site Screenshots β€” Fallback (T4) Only used when T1–T3 sources don't provide enough. The problem: datacenter IPs get Cloudflare-blocked on most fintech sites. So we need either: - **Browser Relay** β€” drive Assaf's actual Chrome browser via the OpenClaw extension **Kitt** (2026-03-20 10:00) - **Residential proxy** β€” route Playwright through a proxy service --- ### πŸ” Web Research - `web_search` β€” campaign discovery, agency identification, press coverage - `web_fetch` β€” brand press rooms, newsroom pages, media kits (often have downloadable hi-res assets) - **Scrapling** (`StealthyFetcher`) β€” anti-bot web scraping when standard fetch gets blocked --- ### 🧠 Vision Analysis Once images are captured, the `image` vision tool (Gemini) analyzes them for colors (with hex values), typography, composition patterns, and brand signature β€” turns raw captures into structured research data. --- ### The Priority Order **Kitt** (2026-03-20 10:00) ``` T1: Agency case study (Prismic/Sanity/Behance) T2: Press / editorial / OOH campaign photography T3: Official brand assets (App Store, brand guidelines pages) T4: Live screenshots via Browser Relay / residential proxy ❌: Never use β€” WAF blocks, CAPTCHA pages, error screens ``` The rule: **never go lower than necessary**. A Prismic pull from Porto Rocha gives you 15 art-directed images in one API call. That beats 50 Playwright screenshots every time. **Dan Peguine** (2026-03-20 10:01) Can you verbatim output the skill dos for visual research here? **Kitt** (2026-03-20 10:01) ## SKILL: `creative-visual-research` β€” Part 1/3 ``` --- name: creative-visual-research description: "Run a full creative visual research pipeline on a brand: discover agency work + campaign imagery, capture social media content, screenshot digital presence, analyze visual identity, and package everything for a competition page build." --- # Creative Visual Research Full-pipeline skill for researching a brand's visual identity, capturing imagery from every available source, and packaging it for a 13-section competition page. ## Input A brand config (provided inline or as JSON): ``` **Kitt** (2026-03-20 10:01) ``` { "brand": "Cash App", "website": "https://cash.app", "social": { "instagram": "cashapp", "tiktok": "cashapp", "youtube": "CashApp", "x": "CashApp", "linkedin": "cash-app" }, "known_agencies": ["Anomaly", "pgLang"], "known_campaigns": ["Cash In", "That's Money", "Heirloom"], "client": "eToro" } ``` **Kitt** (2026-03-20 10:01) ``` ## Pipeline Overview DISCOVER β†’ CAPTURE β†’ EXTRACT β†’ ANALYZE β†’ PACKAGE | Phase | What | Tools | |-------|------|-------| | 1. Discover | Find agency case studies, press coverage, campaign articles | web_search, web_fetch | | 2. Capture | Screenshot websites, social grids, app stores | Playwright, Pinchtab | | 3. Extract | Pull actual image URLs from APIs and CDNs | Instagram API, Prismic/Sanity APIs, extract-images.sh | | 4. Analyze | Vision analysis of captured images | image vision model | | 5. Package | Compress, organize, build research doc | compress-images.js, markdown output | ``` **Kitt** (2026-03-20 10:01) ## SKILL: `creative-visual-research` β€” Part 2/3 ``` ## Phase 1 β€” Discover Sources ### 1a. Agency identification web_search "[Brand] rebrand agency case study" web_search "[Brand] brand identity agency" web_search "[Brand] advertising agency 2024 2025" Check these agency portfolio sites: Anomaly, Collins, Pentagram, Wolff Olins, Porto Rocha, Koto, DixonBaxi, Character, R/GA, Droga5, W+K, TBWA, pgLang ### 1b. Campaign discovery web_search "[Brand] campaign [Year] case study" web_search "[Brand] [Agency] behance" ``` **Kitt** (2026-03-20 10:01) ``` web_search "[Brand] campaign ads of the world" web_search "[Brand] campaign little black book LBB" Priority sources: 1. Behance β€” agencies post full campaign breakdowns with hero images 2. It's Nice That / Creative Review β€” editorial with agency-supplied imagery 3. LBB Online β€” campaign launches with stills 4. Ads of the World β€” archived campaign creative 5. The Drum / Campaign / AdAge β€” trade press with imagery 6. Agency websites directly β€” case study pages ### 1c. Brand press room web_fetch "https://[brand-domain]/press" web_fetch "https://[brand-domain]/newsroom" web_fetch "https://[brand-domain]/media" ``` **Kitt** (2026-03-20 10:01) ``` ### 1d. Social presence discovery web_search "[Brand] instagram tiktok social media presence" ## Phase 2 β€” Capture Screenshots All captures use Playwright (headless Chromium). Output to work/[project]/[brand]/captures/ ### 2a. Brand website playwright screenshot --full-page "https://[brand-website]" captures/website-homepage.png ### 2b. Social profile grids playwright screenshot "https://www.instagram.com/[handle]/" captures/instagram-grid.png playwright screenshot "https://www.tiktok.com/@[handle]" captures/tiktok-grid.png playwright screenshot "https://www.youtube.com/@[handle]" captures/youtube-channel.png ``` **Kitt** (2026-03-20 10:01) ``` ### 2c. App store screenshots playwright screenshot "https://apps.apple.com/app/[brand-slug]" captures/appstore.png playwright screenshot "https://play.google.com/store/apps/details?id=[package]" captures/playstore.png ``` **Kitt** (2026-03-20 10:02) ## SKILL: `creative-visual-research` β€” Part 3/3 ``` ## Phase 3 β€” Extract Images ### 3a. Instagram API (no auth required) curl -s "https://i.instagram.com/api/v1/users/web_profile_info/?username=[handle]" \ -H "User-Agent: Instagram 275.0.0.27.98 Android" \ -H "X-IG-App-ID: 936619743392459" Returns: Profile stats + up to 12 post images at full 1080px with engagement data. Rate limits: ~200 requests/hour without auth. ### 3b. Agency case study images (Prismic API) # 1. Get API ref MASTER_REF=$(curl -sL "https://[studio].cdn.prismic.io/api/v2" | \ ``` **Kitt** (2026-03-20 10:02) ``` python3 -c "import json,sys; d=json.load(sys.stdin); print([r['ref'] for r in d['refs'] if r.get('isMasterRef')][0])") # 2. Find brand case study + extract all image URLs # Add &w=1400 to any Prismic image URL for high-res download ### 3c. Generic image extraction (handles Sanity, Cloudinary, Mux, standard HTML) bash scripts/images/extract-images.sh "https://agency.com/case-study/brand" ./agency-images 15 ### 3d. YouTube thumbnails yt-dlp --write-thumbnail --skip-download --playlist-items 1-5 \ "https://www.youtube.com/@[handle]/videos" -o "youtube/%(title)s.%(ext)s" ## Phase 4 β€” Analyze Run vision analysis on best 8-12 captured images: Prompt: "Analyze this brand image. Extract: (1) dominant colors with hex values, (2) typography style and weight, (3) composition/layout pattern, (4) mood/energy, (5) what makes this distinctive vs competitors." ``` **Kitt** (2026-03-20 10:02) ``` ## Phase 5 β€” Package node scripts/images/compress-images.js work/[project]/[brand]/ ## Tool Reference | Tool | What | |------|------| | Playwright | Browser screenshots | | Pinchtab | Lightweight screenshots + text | | Instagram API | Post images + profile data (no auth) | | extract-images.sh | Agency page image extraction | | yt-dlp | YouTube/TikTok thumbnails | | Scrapling | Anti-bot web scraping | | compress-images.js | Image compression | | Gemini Vision | Image analysis | ``` **Kitt** (2026-03-20 10:02) ``` | Nano Banana Pro | Image generation (fill gaps) | ## Quality Checklist - [ ] Agency identified and case study images extracted (Phase 3b/3c) - [ ] Instagram posts downloaded (Phase 3a) β€” at least 6 images - [ ] Website homepage captured (Phase 2a) - [ ] Social grids captured (Phase 2b) β€” minimum Instagram + TikTok - [ ] 8+ images analyzed with vision model (Phase 4) - [ ] All 13 sections of research doc populated with sourced data - [ ] Images compressed and organized in assets folder - [ ] Financial data includes dates (not stale numbers) - [ ] Campaign data includes agency attribution ``` **Kitt** (2026-03-20 10:02) ## SKILL: `visual-competitive-analysis` β€” Part 1/4 ``` --- name: visual-competitive-analysis description: Create visual competitive analysis pages that break down competitor brands by color palette, typography, layout patterns, positioning, and strategic threat level. Produces CE-styleguide-compliant HTML pages with structured sections. --- # Visual Competitive Analysis ## Page Structure β€” Single Brand | # | Section | Component | Content | |---|---------|-----------|---------| | A | Hero Image | image-hero | One single large image β€” full container width (1100px), 480px height, object-fit cover. Must be reflective of brand identity. T1–T3 only. | | TL;DR | Strategy Essence | tldr-block | Mandatory. Two-column: left = strategy essence (display heading + body + Threat/Vector/Market), right = one representative brand image (portrait 4:5 crop, 280px wide). | ``` **Kitt** (2026-03-20 10:02) ``` | 1 | Quick Facts | facts-grid (4-col) | Revenue, margin, users, AUM, avg balance, growth, founded | | 2 | Positioning | Narrative + callout + tags | Market position, mission, recent pivots, key exec quotes | | 3 | Brand Evolution | timeline | Chronological brand moments | | 4 | Color Palette | palette-grid (5-col) + image-alongside | 3-6 swatches: hex, role name, extracted from live CSS | | 5 | Typography | type-specimen + type-details + image-alongside | Font family, weights, character | | B | Brand in Practice | image-grid (3-col) | 6 agency images in a 3-col grid. No captions. After Typography, before Layout. | | 6 | Layout & UX | two-col (2 rows) | Density, theme posture, grid structure, interaction patterns | | 7 | Assessment | table (accordion) | What works / what doesn't. Open by default. | | 8 | Strategic Verdict | verdict-grid (4-col) | Borrow / Reject / Watch / Threat Level | | 9 | Brand Personality | three-col (accordion) | Archetype, voice, target audience signal. Collapsed by default. | | 10 | vs Client | table + callout (accordion) | Head-to-head comparison. Collapsed by default. | | C | More from Agency | image-strip (horizontal scroll) | 4-6 remaining agency images. Source credit: "Via [Agency Name]" | | 11 | Social Strategy | platform-grid + content-pillars | Per-platform presence, content pillars %, top formats. Accordion open. | | 12 | Content & Campaign | campaign-card + timeline-row | Hero campaigns, agency, year, media mix. Accordion collapsed. | | 13 | Audience & Community | three-col + callout | Demographics, sentiment, UGC, community management. Accordion collapsed. | ``` **Kitt** (2026-03-20 10:03) ## SKILL: `visual-competitive-analysis` β€” Part 2/4 ``` ## Image Source Hierarchy β€” READ BEFORE SOURCING ANYTHING | Tier | Source | Quality | Notes | |------|--------|---------|-------| | T1 | Agency case study (Porto Rocha, Collins, Koto, etc.) | ⭐⭐⭐⭐⭐ | Art directed, professionally composited. Always prefer. | | T2 | Press / editorial photography | ⭐⭐⭐⭐ | Brand in real world β€” events, OOH, product launches. Community forums (Reddit, Discourse) valid T2. | | T3 | Official brand assets | ⭐⭐⭐ | App store screenshots, official marketing decks, investor materials. | | T4 | Live site screenshots (via residential IP or relay) | ⭐⭐ | Only when T1–T3 don't exist. Requires clean capture. | | ❌ | Error pages / WAF blocks / CAPTCHA screens | β€” | NEVER use. Discard immediately. | Hero rule: Hero image MUST be T1–T3. Never a live screenshot, never an error page. Minimum image count: 8 images per page. ``` **Kitt** (2026-03-20 10:03) ``` ## Image Audit β€” MANDATORY before placing any asset Before any image is placed, check: - Is this a real brand/product/campaign image, or a WAF block / CAPTCHA / error page? - Red flags: filename is generic scrape output, file under 50KB, dimensions 1440Γ—900 with white BG and centered text - Quarantine folder: assets/_junk/ β€” move bad assets here, never delete ## Slot Manifest β€” WRITE BEFORE TOUCHING HTML slot-manifest.json defines every image placement before HTML is touched: { "slots": { "A-hero": { "file": "...", "tier": "T1", "why": "..." }, "tldr-image": { "file": "...", "tier": "T2", "why": "..." }, "color-palette-alongside": { "file": "...", "tier": "T2", "why": "..." }, ``` **Kitt** (2026-03-20 10:03) ``` "typography-alongside": { "file": "...", "tier": "T3", "why": "..." }, "B-grid-0" through "B-grid-5": { "file": "...", "tier": "...", "why": "...", "section_label": "..." }, "C-strip": [ { "file": "...", "tier": "T2", "caption": "..." } ] } } Rules: - Every image in HTML must have a corresponding slot entry - Change images by editing manifest first, then HTML - "why" field is mandatory β€” explains decision for future editors - Manifests live at assets/../slot-manifest.json alongside the page ## Agency Source Extraction β€” ALWAYS first Before touching the live site, identify who did the brand work: ``` **Kitt** (2026-03-20 10:03) ``` web_search "[Brand Name] rebrand agency case study" | Agency CMS | Extraction method | |---|---| | Prismic (Porto Rocha) | Hit /api/v2 β†’ get master ref β†’ query case_study type by uid | | Sanity CDN (Koto, Wolff Olins) | extract-images.sh handles automatically | | Mux video (Collins, W+K) | extract-images.sh handles automatically | | Cloudinary / imgix | extract-images.sh handles automatically | | Standard HTML | extract-images.sh "https://agency.com/case-study/brand" ./output 15 | Download at 1400px: add &w=1400 to any Prismic URL. Output: 10–20 production-quality images saved to ./agency-images/[brand]/ ``` **Kitt** (2026-03-20 10:03) ## SKILL: `visual-competitive-analysis` β€” Part 3/4 ``` ## CE Style Guide Compliance β€” Non-Negotiable - Background: white (#fff) β€” never dark - Fonts: Larken (display/headings), Inter (body), JetBrains Mono (data/code) - Larken: Load via Typekit https://use.typekit.net/ffj8sbd.css - Colors: Only 6 values: --red: #cc0000 --black: #1a1a1a --grey: #666 --light: #999 --border: #eee --bg: #fff - Container: max-width: 1100px ``` **Kitt** (2026-03-20 10:03) ``` - Section labels: 10px, uppercase, 0.1em tracking, --red - Headings: Larken, 400 weight, -0.02em letter-spacing - Max text width: 680px for body paragraphs - Borders: 1px solid #eee only - NO: gradients, box-shadows, font-weight 600+, border-radius > 4px, emojis, stock photos ## Visual QA Gate β€” MANDATORY No page deploys without passing visual QA. ### Process 1. Deploy page to staging 2. Run: python3 scripts/visual-qa/section-audit.py - Playwright scrolls full page (triggers lazy-load), captures screenshot - ImageMagick slices into 900px sections ``` **Kitt** (2026-03-20 10:03) ``` - Vision model audits each section against 7 criteria 3. All FAILs must be fixed before reporting complete 4. WARNs flagged but don't block deploy ### 7 Criteria (locked) 1. Image fit β€” no stretching, no overflow, no cropping that loses subject 2. Image visibility β€” actually loads, no broken refs, no alt-text showing 3. Aspect ratio β€” natural for subject, consistent within grids 4. Content relevance β€” right image in right section context 5. Spacing & alignment β€” no dead space, balanced columns, no mismatched heights 6. Text readability β€” sufficient contrast, no clipping, no orphan lines 7. Visual hierarchy β€” clear section flow, hero prominence, proper label/heading/body order ## Quality Checklist ``` **Kitt** (2026-03-20 10:03) ``` - [ ] Typekit loaded (ffj8sbd) - [ ] Only Larken / Inter / JetBrains Mono - [ ] Only 6 CE colors used - [ ] grep -i "gradient" β†’ 0 results - [ ] grep "box-shadow" β†’ 0 results - [ ] grep "font-weight.*[6-9]00" β†’ 0 results - [ ] Container max-width correct - [ ] All 13 sections present - [ ] Responsive at 600px and 900px breakpoints - [ ] Data sourced and current - [ ] Sticky section nav present and scrollspy working - [ ] Assessment accordion open; Personality + vs Client closed - [ ] Image audit run β€” zero error pages / WAF blocks in assets folder - [ ] All images T1–T3 minimum - [ ] Minimum 8 images on page ``` **Kitt** (2026-03-20 10:03) ``` - [ ] Agency source images present - [ ] Hero image placed (Section A) β€” T1–T3 only - [ ] Brand in Practice grid present (Section B) with agency credit - [ ] Horizontal image strip present (Section C) - [ ] Image assets copied to assets/[brand]/ folder (not hotlinked from tmp) ``` **Kitt** (2026-03-20 10:03) ## SKILL: `visual-benchmarking` β€” Part 1/2 ``` --- name: visual-benchmarking description: "Benchmark, compare, and improve competitive analysis pages by auditing visual richness across 6 categories, scoring against gold-standard references, and prescribing specific asset placements." --- # Visual Benchmarking & Improvement ## 6-Category Visual Richness Benchmark Every brand page must cover ALL 6 categories with 3+ assets each to score 4/5: | # | Category | What qualifies | Reliable sources | |---|----------|---------------|-----------------| ``` **Kitt** (2026-03-20 10:03) ``` | 1 | Product UI | Real app screens showing features in use | App Store, Google Play, Mobbin | | 2 | Campaign/OOH | Paid media, billboards, digital ads | Brand social accounts, agency sites | | 3 | Lifestyle | Real people using product in context | Brand Instagram, press kits, partnership imagery | | 4 | Brand system | Design tokens, typography, color in application | Live site CSS extraction, brand guidelines pages | | 5 | Agency/editorial | Case studies, press, design community coverage | Behance, Dribbble, Brand New, agency portfolios | | 6 | Physical/merch | Cards, packaging, physical brand objects | Product photography, unboxing content | ## T1/T2/T3 Image Quality Tiers | Tier | Name | Definition | Rule | |------|------|-----------|------| | T1 | Editorial | Campaign photography, lifestyle, OOH, agency creative, people-in-context | Min 8 per page. Hero must always be T1. | | T2 | Product | Card renders, hero UI in context, feature screenshots, app mockups | Max 40% of total image count | | T3 | Reference | Icon grids, color swatches, typography specimens, small UI crops | Max 20% of total. Never use as filler. | ``` **Kitt** (2026-03-20 10:03) ``` KEY INSIGHT: Image count is the wrong metric. Adding T3 icons pads the count while lowering visual quality. v3 with 49 images scored WORSE than v2 with 20, because T3% was 33% and T1% was only 8%. ## Scoring Dimensions (1–5 each) | Dimension | What to evaluate | |---|---| | Product clarity | Can you understand the product from visuals alone? | | Visual richness | Depth beyond basic screenshots β€” variety of section types and image categories | | Emotional pull | Does it make you want to use the product? | | Information density | Right amount of visual content per section | ``` **Kitt** (2026-03-20 10:04) ## SKILL: `visual-benchmarking` β€” Part 2/2 ``` ## Audit Process ### Step 1: Count and classify - Count total tags in HTML - List all asset files in brand's folder - Classify each as: used/unused, category (1–6), T1/T2/T3 tier - Flag junk: 404 pages, CAPTCHA captures, blank/corrupt files, duplicates ### Step 2: Score current state - Score each of 4 dimensions (1–5) - Count categories covered (need 3+ assets per category) - Record: X/6 categories, T1/T2/T3 percentages ``` **Kitt** (2026-03-20 10:04) ``` ### Step 3: Identify gaps - Which categories have <3 assets? - Which unused assets should be on the page? - What new assets need sourcing? - Is T1% above 20%? Is T3% below 20%? ### Step 4: Write critique.md # [Brand] β€” Visual Audit & Critique ## Scores | Dimension | Score | Notes | ## Category Coverage: X/6 | # | Category | Assets | Status (βœ…/⚠️/❌) | ## T1/T2/T3 Breakdown T1: X (X%) | T2: X (X%) | T3: X (X%) ## High-value unused assets ``` **Kitt** (2026-03-20 10:04) ``` | Filename | Category | Tier | Recommended placement | ## Junk files (delete) ## New assets needed | Category | Description | Source | ## Quality Gate (before build starts) - [ ] T1 images β‰₯ 8 - [ ] T3 images ≀ 20% of total - [ ] All 6 categories have 3+ assets - [ ] Hero is T1 editorial - [ ] No file over 200KB (JPG) / 300KB (PNG) - [ ] No duplicate files - [ ] 30-second scroll test: every scroll window contains β‰₯1 T1 or strong T2 image ## Reliable Enrichment Sources (by success rate) ``` **Kitt** (2026-03-20 10:04) ``` 1. App Store / Google Play β€” always public, never blocked 2. Mobbin β€” real UI screenshots, unblocked 3. Vimeo β€” campaign videos, embeddable 4. Behance β€” agency case studies (search by brand + agency name) 5. Brand guidelines pages β€” many fintechs publish publicly Sources that frequently block automation: - Brand websites (Cloudflare/CAPTCHA) β€” flag as "needs manual collection" - Instagram (login wall) β€” needs human to save and upload - Medium (Cloudflare) - Most editorial sites (cookie consent walls) ## Benchmark References - CashApp β€” gold standard for visual richness (71 images, culture + lifestyle sections), score 72 - Robinhood β€” gold standard for brand depth (53 images, personality + layout sections) ``` **Kitt** (2026-03-20 10:04) ``` Both are read-only reference β€” do not modify. ## Comms Protocol - Auditor posts RESEARCH_COMPLETE β†’ builder (Kitt) picks up β†’ QA β†’ Assaf sign-off - Kittoro does NOT commit to git or deploy β€” Kitt only - Versioning: never overwrite originals. Next version = [brand]-vN+1/ ``` --- That's all three skill docs verbatim. Full files also at: - `skills/creative-visual-research/SKILL.md` - `skills/visual-competitive-analysis/SKILL.md` - `skills/visual-benchmarking/SKILL.md` **Assaf Dagan** (2026-03-20 10:06) are you using last30days at all in your work ? **Kitt** (2026-03-20 10:06) Short answer: **no, not currently.** The `creative-visual-research` skill doesn't reference or call `last30days` anywhere in its pipeline. Where it *could* be useful: **Phase 1 – Discover** is the gap. Right now we use raw `web_search` + `web_fetch` to find campaigns and agency work. `last30days` would be strictly better for that because it: - Searches Reddit, X, YouTube, and web in one shot - Has persistent history β€” so if we've researched a brand before, it can surface accumulated knowledge instead of re-searching from scratch - Has a watchlist mode β€” we could monitor competitors on a schedule and get briefings when something changes **Practical use case:** `last30 Robinhood brand campaign` would pull recent Reddit discussion, X posts, YouTube coverage in one call β€” currently we'd run 3–4 separate `web_search` queries to cover the same ground. Want me to integrate it into the `creative-visual-research` skill? Specifically in Phase 1 β€” replace or augment the manual `web_search` calls with `last30days` for campaign discovery and social listening. **Assaf Dagan** (2026-03-20 10:22) switch to opus 4.6 please **Kitt** (2026-03-20 10:22) Switched to Opus 4.6. **Dan Peguine** (2026-03-20 10:30) We want to package these skills and open source them. I want you to make a plan for how to do this. 1. Create a new project folder for this project. 2. Add all these as context to learn how to make great skills. Read them all and synthesize the advice into a thorough document Here's everything I found on skill creation advice from your reading list: The big ones: πŸ”₯ Thariq (@trq212) β€” "Lessons from Building Claude Code: How We Use Skills" https://x.com/trq212/status/2033949937936085378 πŸ”₯ Ole Lehmann (@itsolelehmann) β€” "How to 10x your Claude Skills using Karpathy's autoresearch method" https://x.com/itsolelehmann/status/2033919415771713715 πŸ”₯ Aakash Gupta (@aakashgupta) β€” Follow-up on Ole's autoresearch method applied to skills https://x.com/aakashgupta/status/2034439536927740287 Practical how-tos: πŸ›  Blum (@Blum_OG) β€” "How to create an OpenClaw skill in 3 min" https://x.com/Blum_OG/status/2031479745263841650 πŸ›  @Shpigford β€” Built an open source macOS skills organizer + editor app https://x.com/Shpigford/status/2034365122638954924 Conceptual / advanced: 🧠 @tricalt (Vasilije/Cognee) β€” Self-improving skills for agents: observeβ†’inspectβ†’amendβ†’evaluate loop β€” couldn't recover the exact tweet link, but their Cognee plugin is at https://x.com/tricalt/status/2033193383733145751 🧠 Thariq β€” "Seeing like an Agent" (companion piece, action space design) https://x.com/trq212/status/2027463795355095314 3. make a plan for how to turn Thes skill into a shippable GitHub repo. 4. Look into how Corey launched his marketing skills. How did he do it? How did it become so viral? We want to build a plan that makes this skill launch viral: Here’s the link to the tweet: https://x.com/coreyhainesco/status/2013272998191812906 5. Create a a task list doc in the project folder to track this [embed: β€” Lessons from Building Claude Code: How We Use Skills] [embed: β€” How to 10x your Claude Skills \(using Karpathy's autoresearch method\)] [embed: β€” You need to steal this autoresearch application to skills right NOW] [embed: β€” how to сreate OpenClaw skill in 3 min skills are a key part of a quality AI agent u can't do without them and custom skills are often essential saf] [embed: β€” ugh\. fine\. i'll drop TWO macOS apps\. https://t.co/n103RxY0E0 \- simple markdown file viewer/editor \(with quick look\!\) https://t.co/GikMTTbGYu \] **Kitt** (2026-03-20 10:36) Done. Here's what I built: **Project folder:** `work/pitches/open-source-skills/visual-research-skills/` **4 documents created:** ### 1. `research/skill-creation-synthesis.md` Synthesized all 8 sources into a comprehensive guide β€” the "10 Commandments of Great Skills": - Thariq (Anthropic): Skills are directories, not files. Give memory. Don't teach what the model knows. Design action spaces like equipping a "human in a box." - Ole Lehmann: Autoresearch loop β€” generate β†’ score against binary evals β†’ mutate β†’ iterate overnight - Aakash Gupta: Apply autoresearch to any domain skill. Define binary yes/no criteria, let the agent self-optimize. - Anthropic Official Guide: Progressive disclosure (metadata β†’ body β†’ references). Context window is a public good. Description field = the only trigger mechanism. - Cognee/tricalt: observe β†’ inspect β†’ amend β†’ evaluate loop for continuous improvement ### 2. `research/corey-launch-analysis.md` Corey's `marketingskills` repo: **14.9k stars, 2.2k forks**. What worked: - 30+ skills covering the entire marketing stack (massive scope, all free) **Kitt** (2026-03-20 10:36) - ASCII dependency diagram in README (signals engineering depth) - Foundation skill pattern (`product-marketing-context` read by all others) - Platform-agnostic ("Works with Claude Code, OpenAI Codex, Cursor, Windsurf") - Free skills β†’ paid agency/newsletter/AI agent (value ladder) - Perfect timing β€” launched while skills were trending on Twitter ### 3. `PLAN.md` Full 5-phase launch plan: 1. **Prepare** β€” Strip internal refs, generalize styles, bundle scripts, create foundation skill, add autoresearch evals 2. **Repo structure** β€” Clean directory layout with 4 skills (foundation + 3 core), bundled scripts, examples, evals 3. **README design** β€” Pipeline diagram, live examples, one-command install, compatibility section 4. **Launch strategy** β€” Main tweet + 7-tweet thread showing real output, DM 10-15 people, Reddit/HN/ClawHub same day 5. **Value ladder** β€” Skills free, CE consulting behind them ### 4. `TASKS.md` Full task list with checkboxes, organized by phase. Includes 6 decisions needed from you two. **Kitt** (2026-03-20 10:36) **Key decisions to make:** 1. GitHub org name? 2. Ship CE styleguide as default or fully generic? 3. Include real brand pages as examples? 4. Ship fast (this week) or polish more? 5. Link to CE services from README? 6. Who owns maintenance?