--- name: linkedin-cover-generator owner: Senthilnathan github: https://github.com/senthilsweb description: Generate polished LinkedIn cover images (article covers, promo cards, profile banners, feed images, carousel pages) from an article, post draft, short idea, local file, or remote URL — editorial left-text / right-visual layout, curated palette system, safe-area rules, and an approval-before-generation workflow. Use whenever Senthil asks for a cover image, banner, promo card, or social image for LinkedIn or an article — even if he just says "create a cover for this post". user-invocable: true --- # linkedin-cover-generator — LinkedIn cover image generator ## Defaults vs rules Hard rules in this file: the **file naming** convention, the **where the files go** convention, the **safe-area** margins, and the **brand and safety** rules. Everything else — sizes, density, palette choices, the topic-to-visual map, chip counts — is a default or a lookup, chosen for the common case. The `topic_visuals` and palette-selection tables below are illustrative, not exhaustive. A topic that is not listed is normal: pick the closest match or compose a new visual, and add the new entry to this file if the topic is likely to recur. Never bend the content to fit a listed topic just because it is listed. ## Purpose Generate polished LinkedIn cover images from an article, post draft, short idea, local file, or remote URL. The goal is to create a professional article-cover or promo-card image that follows a consistent editorial style while adapting the visual theme, palette, wording density, and layout to the topic. This skill is intended for technical thought-leadership posts, open-source project announcements, architecture articles, AI/agent engineering posts, data governance posts, privacy/security posts, and career-positioning content. --- ## Default output standard Unless the user explicitly requests another size, generate the image using: ```text 1279 × 720 px ``` This is the default LinkedIn article/promo cover size used by this workflow. ### Supported size presets ```yaml size_presets: linkedin_article: width: 1279 height: 720 description: Default LinkedIn article and promo cover image. linkedin_profile_cover: width: 1584 height: 396 description: LinkedIn personal profile background image. linkedin_post: width: 1200 height: 627 description: General LinkedIn feed image. linkedin_square: width: 1080 height: 1080 description: Square social image. linkedin_carousel_page: width: 1080 height: 1350 description: Portrait carousel page. wide_banner: width: 1920 height: 1080 description: Wide presentation-style banner. ``` ### Custom size syntax Accept any of these user inputs: ```text size=1279x720 canvas=1279x720 dimension="1279 × 720" width=1279 height=720 ``` Normalize all of them to `{ width, height }`. --- ## File naming (strict — this was missed once) Every generated image file MUST be named: ```text --.png ``` - `` — the post/article title, slugified (lowercase, hyphens only, no spaces or underscores). **Never omit it.** A file called `cover-cream-emerald.png` is wrong: once it leaves the folder nobody knows which post it belongs to, and two posts in the same repo collide. - `` — the palette key from the palette system, hyphenated (`cream_emerald` → `cream-emerald`, `blue_pink_gloss` → `blue-pink-gloss`). - `` — the visual treatment actually applied: `subtle`, `glossy`, `minimal`, `balanced`, or `infographic`. Example: ```text building-an-agent-is-only-half-the-job-blue-pink-gloss-subtle.png ``` When generating a set across several palettes, keep the slug and effect identical across the set so the files sort together and only the palette varies. If a build script produces the images, put the slug and effect in named constants at the top of the script so a rebuild cannot drift from the delivered filenames. ## Where the files go - Cover for a **standalone LinkedIn post** → `linkedin-posts//` - Cover for an **article** → `article-drafts//` Images always live inside the post's or article's own folder — never at the repo root. ## Input sources The skill must accept content from any of the following. ### 1. Direct text input The user may paste the post, article, title, rough notes, or full Markdown body directly in the prompt. ### 2. Local file path The user may provide a local file path such as: ```text input=./articles/my-post.md input=/mnt/data/article.md input=./drafts/post.txt ``` Supported file types: ```yaml supported_input_files: - .md - .markdown - .txt - .html - .htm - .json ``` Optional extended support if available: ```yaml optional_input_files: - .pdf - .docx ``` For `.pdf` and `.docx`, extract readable text first. Do not rely on visual screenshots unless the task specifically requires visual interpretation. ### 3. Remote URL The user may provide: ```text input=https://example.com/article input=https://raw.githubusercontent.com/user/repo/main/article.md input=https://github.com/user/repo/path/to/file.md ``` Behavior: 1. Fetch the URL. 2. Detect whether it is raw Markdown, HTML, or plain text. 3. Extract title, subtitle, description, headings, links, and relevant body text. 4. Ignore unrelated navigation, cookie banners, ads, and page chrome. 5. Preserve source links only if the user asks to include them on the image. ### 4. Reference image The user may provide a visual reference and ask to follow its layout, size, or style. Behavior: - Reuse the style direction, not exact copyrighted design elements. - Preserve the user’s established layout patterns if they refer to earlier successful covers. - If a reference image is used only for size, match the aspect ratio and composition safety rules. --- ## Content extraction Extract a concise creative brief from the input. ### Required extraction fields ```json { "title": "", "subtitle": "", "short_support_line": "", "topic": "", "category": "", "audience": "", "core_message": "", "keywords": [], "must_include_text": [], "must_avoid_text": [], "links_to_include": [], "brand_or_company_names": [], "personal_links": [], "tone": "", "visual_metaphor": "", "sensitivity": "normal" } ``` ### Extraction rules 1. Prefer the article title when it is strong and concise. 2. If the title is too long, create a shorter cover title. 3. Keep subtitles short and readable at LinkedIn preview size. 4. Avoid turning the cover into a full summary. 5. Extract only the strongest 3–5 concepts for visual chips. 6. Preserve user-specified title text unless it is too long or visually unsuitable. 7. Ask for approval before using employer names, product names, or customer names unless the user explicitly requests them. 8. If the user says “safe play,” “personal post,” “my view,” or “thought leadership,” avoid company/product names and use generic conceptual wording. 9. If the user asks for GitHub, LinkedIn, demo, or repo links, include them as small footer text only. --- ## Default layout pattern Use the established LinkedIn cover style: ```yaml layout: structure: editorial_left_visual_right left: - category badge - large title - short subtitle - optional quote or support line - 3 to 5 feature chips right: - abstract visual metaphor OR clean architecture illustration - minimal text - icons/cards only when useful footer: - optional small link strip ``` ### Safe-area rules For `1279 × 720`: ```yaml safe_area: margin_left: 56 margin_right: 56 margin_top: 48 margin_bottom: 52 footer_height_max: 72 ``` Rules: 1. Keep main title away from all edges. 2. Do not place important text at extreme bottom. 3. Avoid very small text. 4. If a footer is required, keep it inside a clean footer strip. 5. Do not put important content near corners. 6. Avoid overcrowded right-side diagrams. 7. Prefer fewer words over more words. --- ## Visual density levels Accept: ```yaml density: minimal: description: Executive, abstract, thought-leadership style. Least text. balanced: description: Title + subtitle + few chips + visual metaphor. Default. infographic: description: More labeled cards and architecture elements. Use only when requested. ``` Default: ```text balanced ``` Use `minimal` for personal opinion, thought leadership, career, leadership, or conceptual articles. Use `infographic` for architecture, workflow, agent pipeline, product walkthrough, or implementation-heavy articles. --- ## Color palette system Choose a palette based on the article theme and recent visual history. Avoid repeating the same palette in consecutive images unless the user asks for consistency. ### Palette options ```yaml palettes: cream_emerald: mood: calm, premium, modern, trustworthy best_for: - data workflows - job-scout agents - multi-agent architecture - homelab / hands-on engineering colors: background: "#F8F2E6" primary: "#062E2B" secondary: "#0B6B55" accent: "#D8C98A" soft_accent: "#BFD8C8" deep_navy_cyan_violet: mood: technical, futuristic, engineering-heavy best_for: - AI agents - architecture - testing - observability - distributed systems colors: background: "#061326" primary: "#F8FAFC" secondary: "#38DDF5" accent: "#8B5CF6" soft_accent: "#1E40AF" blue_pink_gloss: mood: modern, energetic, polished best_for: - AI era - elegant engineering - career positioning - forward-deployed engineering colors: background: "#F8F5EE" primary: "#071C3A" secondary: "#2563EB" accent: "#EC4899" soft_accent: "#C7D2FE" dark_neon_pink_blue: mood: bold, futuristic, high-impact best_for: - AI engineering - agent systems - technical essays - modern software roles colors: background: "#050B1E" primary: "#FFFFFF" secondary: "#38BDF8" accent: "#F472B6" soft_accent: "#7C3AED" privacy_orange_coral: mood: alert but polished, sensitive data, privacy best_for: - privacy - DLP - PII - data protection - security workflows colors: background: "#FFF7ED" primary: "#111827" secondary: "#F97316" accent: "#EF4444" soft_accent: "#FDBA74" light_blue_purple: mood: clean, intelligent, data/AI best_for: - document fingerprinting - MinHash - LSH - data intelligence - search/indexing colors: background: "#EAF6FF" primary: "#071C3A" secondary: "#2563EB" accent: "#7C3AED" soft_accent: "#67E8F9" charcoal_gold_teal: mood: premium, serious, production-grade best_for: - production systems - durable agents - frameworks - orchestration colors: background: "#07110F" primary: "#F8FAFC" secondary: "#14B8A6" accent: "#FBBF24" soft_accent: "#2DD4BF" soft_slate_copper: mood: elegant, mature, editorial best_for: - leadership - engineering craft - career reflection - executive writing colors: background: "#F6F1EA" primary: "#1E293B" secondary: "#475569" accent: "#B77945" soft_accent: "#CBD5E1" ``` ### Color selection rules 1. For privacy/DLP/security: prefer `privacy_orange_coral`. 2. For data governance/adaptive operating model: prefer `deep_navy_cyan_violet` or minimalist abstract variant. 3. For job-search/data pipeline agents: prefer `cream_emerald`. 4. For AI engineering / elegant engineering: prefer `blue_pink_gloss` or `dark_neon_pink_blue`. 5. For document fingerprinting / MinHash / LSH: prefer `light_blue_purple`. 6. For multi-agent orchestration/frameworks: prefer `cream_emerald`, `charcoal_gold_teal`, or `deep_navy_cyan_violet`. 7. For homelab and hands-on engineering: prefer soft cream, teal, coral, and light data-catalog style. 8. If user asks “different color than previous two,” avoid the last two used palettes. 9. If user asks “subtle,” reduce saturation and use a soft background. 10. If user asks “glossy,” use glass cards, soft shadows, glow, and polished highlights. --- ## Text generation rules ### Title The title must be short, high-impact, readable at thumbnail size, no more than 2–3 lines, and ideally under 55 characters. If the article title is long, create a shorter cover title and preserve the article title in the post, not the image. ### Subtitle The subtitle should explain the value in one line or two short lines. Examples: ```text A Testing Architecture for AI-Augmented Quality Engineering Near-Duplicate Detection Without AI Embeddings Orchestrator + Sub-Agents Pattern Data governance needs an adaptive operating model How My Homelab Keeps Me Hands-On ``` ### Support line Optional. Use sparingly. Examples: ```text AI did not invent elegant engineering. It made it essential. Build the intelligence once. Make every surface a thin adapter. Structured. Scalable. Observable. Data model first • Verify sources • Never fabricate ``` ### Chips Use 3–5 short chips. Examples: ```text DuckDB Deterministic Referral-First Agentic Toggle ``` ```text PFC Tree Pairwise Observability Triage Lint Gates ``` ```text CLI REST API Python MCP Chat ``` ```text Business-first Testing Reliability Security Reuse ``` --- ## Footer link rules Only include footer links if the user explicitly asks. Supported footer types: ```yaml footer_links: github: icon: GitHub mark or simple code/repo icon text_format: "github.com//" linkedin: icon: LinkedIn or profile icon text_format: "linkedin.com/in/" live_demo: icon: globe or launch icon text_format: "" docs: icon: document icon text_format: "" ``` Rules: 1. Keep footer link text small but readable. 2. Prefer one clean footer strip. 3. If both LinkedIn and GitHub are requested, place LinkedIn on the left and GitHub on the right. 4. Avoid raw `https://` unless the user explicitly wants the full visible URL. 5. For LinkedIn posts, use non-shortened visible links only if the user requests them. 6. Do not invent links. --- ## Brand and safety rules 1. Do not add employer names unless explicitly requested. 2. Do not add product names unless explicitly requested. 3. If a user says “safe play,” avoid company/product/customer names. 4. Do not use official logos unless the user provides them or explicitly requests them. 5. Generic icons are preferred over brand marks. 6. Do not claim certifications, employment status, partnerships, or endorsements. 7. Avoid making the cover look like an official company advertisement unless requested. 8. For personal thought leadership, use “My View” style only if the user asks. --- ## Topic-to-visual mapping ```yaml topic_visuals: ai_augmented_quality_engineering: visual: - capability folder - PFC tree - AI agent cards - CI lint gate chips: - PFC Tree - Pairwise - Observability - Triage - Lint Gates diagram_agents: visual: - central orchestrator - renderer variations - reporter metrics chips: - Orchestrator Driven - Specialized Sub-Agents - Every Run Tracked - Framework When It Matters job_scout_agent: visual: - job postings into database - extract and structure - fit score - referral path - verify and trust chips: - DuckDB - Deterministic - Referral-First - Agentic Toggle adaptive_datagovops: visual: - abstract adaptive flowing paths - minimalist governance motion chips: - Algorithms - Business Rules - Custom APIs - Optional AI/ML document_fingerprinting: visual: - documents into fingerprint - MinHash - LSH index - distributed cluster chips: - MinHash - LSH - Jaccard - Dagster Pipeline privacy_shield: visual: - shield - PII cards - detect / anonymize / reverse anonymize chips: - Detect PII - Anonymize - Reverse Anonymize - Local-first homelab: visual: - Mac mini - Raspberry Pi - cloud platforms - mesh/tunnel lines - containers and services chips: - Mac mini - Raspberry Pi - Cloud - Docker - Tailscale elegant_engineers_ai_era: visual: - elegant engineer core - AI engineer ready - forward-deployed ready - checklists chips: - Business-first - Testing - Reliability - Security - Reuse ``` --- ## Prompt template for image generation Use this structure when calling an image generation model: ```text Create a professional LinkedIn cover image in exactly {width} × {height} px. Style: - {palette_name} palette: {palette_description} - modern editorial technology cover - polished, premium, clean, readable - subtle gradients, glass cards, soft shadows, and balanced whitespace - no clutter, no tiny unreadable text Layout: - left side: category badge, large title, subtitle, optional support line, 3–5 chips - right side: {visual_metaphor} - footer: {footer_description} - keep all important text inside safe margins - do not crop title or footer - avoid placing important content at the extreme edges Text to include: - Badge: "{badge}" - Title: "{title}" - Subtitle: "{subtitle}" - Support line: "{support_line}" - Chips: {chips} - Footer links: {footer_links} Do not include: - {must_avoid_text} - extra company names - invented product names - long paragraphs - unreadable microtext - distorted logos - fake UI brand marks Rendering requirements: - high-resolution - sharp typography - readable at LinkedIn preview size - clean technical visual metaphor - final output must be exactly {width} × {height} px ``` --- ## Approval workflow If the user asks to approve the theme before generation, provide a proposal first. ### Proposal format ```markdown ## Proposed cover direction **Size:** 1279 × 720 **Palette:** cream_emerald **Theme:** Modern data-agent workflow **Title:** Job Hunting Is a Data Problem **Subtitle:** I Treated It Like One **Visual:** Job postings flow into a structured data model, then out to fit score, referral path, verification, and optional agent layer. **Footer:** github.com/senthilsweb/ai-agents Please confirm, and I will generate the cover. ``` Do not generate the image until the user confirms. If the user directly says “create now,” “generate,” or “approved,” proceed. --- ## Revision workflow When revising an image: 1. Preserve what the user liked. 2. Change only what they requested. 3. If they say “previous one was better,” return to that palette/layout and apply only the requested change. 4. If they ask for links to be added, preserve the design and add a footer strip. 5. If they complain about cut-off, reduce text size, increase safe margins, and use the proven `1279 × 720` composition. 6. If they ask for different colors, generate 2 variations using clearly different palettes. 7. Avoid creating a totally new concept unless requested. --- ## Quality checks Before finalizing, check: ```yaml quality_checks: dimensions: required: true expected_width: user_or_default_width expected_height: user_or_default_height readability: title_readable: true subtitle_readable: true footer_readable_if_present: true layout: title_not_cut_off: true footer_not_cut_off: true no_overcrowding: true safe_margins_respected: true content: required_title_present: true requested_links_present: true forbidden_names_absent: true no_invented_brand_claims: true style: palette_matches_request: true professional_linkedin_cover: true ``` If any hard check fails, revise once. Do not enter an unbounded revision loop. --- ## Output response After generation, do not over-explain. Provide either: ```text Done. ``` or, if files are created outside the image generation UI: ```markdown Done — here is the generated cover image: [Download cover image](sandbox:/mnt/data/.png) ``` For non-image generation implementations, include output path, size, palette, and any warnings. --- ## Examples ### Example 1 User: ```text Create a LinkedIn cover image for my article: "Job Hunting Is a Data Problem. I Treated It Like One." Use subtle green and cream. Add GitHub link at the footer. ``` Extracted cover spec: ```json { "size": "1279x720", "palette": "cream_emerald", "badge": "AI Agents / Job-Scout Agent", "title": "Job Hunting Is a Data Problem", "subtitle": "I Treated It Like One", "support_line": "Deterministic job search with DuckDB, scoring, referrals, and an optional agentic layer", "chips": ["DuckDB", "Deterministic", "Referral-First", "Agentic Toggle"], "footer_links": ["github.com/senthilsweb/ai-agents"], "visual_metaphor": "job postings flow into a structured data model and out to fit score, referral path, verification, and optional agent layer" } ``` ### Example 2 User: ```text Create a cover for Adaptive DataGovOps, minimalist, no company names. ``` Extracted cover spec: ```json { "size": "1279x720", "palette": "deep_navy_cyan_violet", "badge": "Data Governance / Adaptive Operating Model", "title": "Adaptive DataGovOps", "subtitle": "Data governance needs an adaptive operating model", "support_line": "Not one-size-fits-all", "chips": [], "footer_links": [], "visual_metaphor": "abstract adaptive flowing data paths, no labeled boxes" } ``` ### Example 3 User: ```text Use pink and blue variation. Add LinkedIn and GitHub footer links. ``` Extracted cover spec: ```json { "size": "1279x720", "palette": "blue_pink_gloss", "footer_links": [ "linkedin.com/in/senthilsweb", "github.com/senthilsweb" ] } ```