# Content Pipeline > Use when someone asks to repurpose one idea into a LinkedIn post + cover image + Skool post, run a content distribution pipeline, or create a multi-platform content package from a single input. - Skill: `devenkhatri/content-pipeline` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add devenkhatri/content-pipeline` - Raw SKILL.md: https://api.skillmd.com/api/skills/devenkhatri/content-pipeline/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - Author: devenkhatri (https://skillmd.com/u/devenkhatri) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/devenkhatri/content-pipeline --- # Content Pipeline Transform one idea into a complete multi-platform content package — a high-engagement LinkedIn post, a 1200×627 cover image, a casual Skool community post, and a premium carousel — all from a single input, run as four sequential stages. ## When to Use This Skill Trigger this skill when the user wants to: - Repurpose one idea across LinkedIn and Skool - Run a full content pipeline from a single topic - Create a "content package" for a post - Distribute content across multiple platforms at once - Save time by writing a LinkedIn post, cover, and Skool post in one shot **Trigger phrases:** - "Run the content pipeline on [topic]" - "Create a LinkedIn + Skool package for [idea]" - "Repurpose this for LinkedIn and Skool" - "Make a content package about [topic]" - "Turn this into a full post package" ## Pipeline Overview ``` [Topic / Idea / Key Points] │ ▼ ┌────────────────────────────────────┐ │ Stage 1: LinkedIn Post │ │ • Hook + re-hook │ │ • Body + single takeaway │ │ • Soft CTA │ │ • Repost + Follow lines │ │ • Hashtags │ └────────────────────────────────────┘ │ linkedin-post.md ▼ ┌────────────────────────────────────┐ │ Stage 2: Cover Image │ │ • 1200×627 SVG (1.91:1) │ │ • Central safe zone │ │ • Core insight as hero │ │ • Subtle branding │ │ • + text prompt for AI tools │ └────────────────────────────────────┘ │ cover.svg + cover-prompt.md ▼ ┌────────────────────────────────────┐ │ Stage 3: Skool Post │ │ • 30–40% shorter │ │ • Casual peer tone │ │ • No hashtags │ │ • Tool/tip/insight leads │ │ • Bold closing question │ └────────────────────────────────────┘ │ skool-post.md ▼ ┌────────────────────────────────────┐ │ Stage 4: Carousel │ │ • Delegates to /carousel-creator │ │ • Pre-collected intake (no │ │ re-prompting) │ │ • Premium SVG slides │ │ • Caption + exports │ │ • Outputs land in the SAME │ │ folder as Stages 1–3 │ └────────────────────────────────────┘ │ slides/, caption.md, exports ▼ [Four deliverables, one input, one folder] ``` ## Inputs **Required:** - **Topic or idea** — The core insight, news, tip, or concept to publish **Optional** (gathered during the intake if not provided): - **Key points or outline** — 2–5 bullet points or notes to weave into the post - **Tone preference** — Urgent, educational, or inspirational (auto-inferred if missing) - **Target audience** — Who the post is for (defaults to AI/automation practitioners) - **Tools, prompts, frameworks** — Specific AI tools, prompt templates, or systems the post should mention ## Outputs All saved to a date-wise folder in the current working directory using `YYYYMMDD-HHMM` (real current date/time, computed at intake so all four stages share the same folder): ``` YYYYMMDD-HHMM/ ├── linkedin-post.md # Stage 1: Title + full post (caption + content + CTAs + hashtags) ├── cover.svg # Stage 2: 1200×627 SVG cover image ├── cover-prompt.md # Stage 2: Detailed prompt for Midjourney / DALL-E / Flux ├── skool-post.md # Stage 3: Casual Skool community post ├── slides/ # Stage 4: SVG carousel slides (from /carousel-creator) │ ├── {slug}-slide-1.svg │ └── ... ├── caption.md # Stage 4: Carousel caption (from /carousel-creator) ├── {slug}.pdf # Stage 4: Carousel export ├── {slug}.mp4 # Stage 4: Carousel export └── {slug}-hyperframes.mp4 # Stage 4: Carousel export ``` > **Why a folder:** Each stage produces a distinct deliverable. Bundling them in one folder makes the package easy to share, archive, or re-render. Stage 4 is delegated to `/carousel-creator` and the subagent is responsible for landing all carousel output into the same `YYYYMMDD-HHMM/` folder Stages 1–3 already use. --- ## Workflow ### Step 0: Intake Before generating anything, gather the following. **Compute the timestamp `TS = YYYYMMDD-HHMM` once at the start of intake and use it for all four stages** — this keeps every stage's output in the same folder. #### Content intake (used by Stages 1–3) 1. **Topic or idea** — If not provided in the argument, ask. Never invent a topic. 2. **Key points** — Ask for 2–5 bullet points or supporting material, or extract them from the topic. 3. **Tone** — Default to "educational" unless the topic implies urgency, controversy, or inspiration. 4. **Tools / frameworks** — Ask if the post should mention specific AI tools, prompts, or systems. #### Carousel intake (used by Stage 4) These are passed to the `carousel-creator` subagent so it can skip the carousel's own intake questions. 5. **Platform** — Instagram (1080×1350, 4:5) or LinkedIn (1080×1080, 1:1). Default: **LinkedIn** to match the post platform. 6. **Handle** — The CTA handle. Default: **`@devengoratela`** (matches the LinkedIn Follow CTA). 7. **Audience** — Defaults to the same audience inferred for the LinkedIn post. 8. **Brand colors** — Optional. If provided, pass to the carousel. Otherwise the carousel-creator's premium defaults apply (which already match the cover's dark navy + LinkedIn blue accent). 9. **Slide count** — Optional. If not provided, let the carousel-creator decide based on topic complexity (default range: 6–8). If the user provides minimal input (just a topic), make reasonable inferences for all fields and proceed. Note inferred choices inline in the LinkedIn post output (e.g., "(tone: educational, platform: LinkedIn, handle: @devengoratela — say so if you want any of this changed)"). --- ### Stage 1: LinkedIn Post **Role:** Senior LinkedIn Content Strategist. Your job is a high-engagement, mobile-readable, professional post designed to drive likes, comments, and shares — and funnel interested readers into the Practical AI Skool community. #### Hook (lines 1–2) - **Line 1: under 8 words.** This is the only line visible before "...see more" on mobile. - **Line 2: a punchy re-hook** that earns the click-to-expand. - Both lines must stop the scroll. No "Today I want to talk about…" #### Body - Short paragraphs. White space between them. No walls of text. - Emojis at the end of logical sentences (professional, not decorative). - Adopt a human-to-peer voice. - **Avoid AI slop:** `delve`, `landscape`, `unlock`, `tapestry`, `navigate`, `realm`, `leverage`, `in today's fast-paced world`, `in conclusion`, `game-changer`, `revolutionize`. - One clear takeaway or insight. Don't try to teach five things. #### Soft CTA (before the closing lines) A question or invitation to share experiences. **Tone-matched** to the post: - Urgent / provocative post → punchy, direct question - Educational post → value-driven invitation - Inspirational post → warm, community-focused prompt #### Mandatory closing lines (in this exact order, after the soft CTA, before hashtags) 1. **♻️ Repost** — A line emphasizing the article's value and why the reader's network needs to see it. 2. **➕ Follow** — A line positioning **Deven Goratela** (https://www.linkedin.com/in/devengoratela/) as the go-to authority for staying ahead in AI and automation. The three CTA blocks (soft CTA + Repost + Follow) must all match the article's tone. #### Output format for `linkedin-post.md` ```markdown # [Post Title] [relevant emoji] --- [Hook line 1 — under 8 words] [Re-hook line 2 — punchy] [Body — short paragraphs with emojis at the end of sentences] [Soft CTA — question or invitation] ♻️ [Repost line] ➕ Follow [Follow line positioning Deven Goratela as the AI/automation authority] [Hashtags on their own line, separated by spaces, 3–5 relevant tags] ``` #### Output rules - Title ends with a relevant emoji. - No bold formatting anywhere in the post body. - Hashtags are always included. - The output must be **copy-paste ready**. No further editing. --- ### Stage 2: Cover Image **Role:** Senior visual designer for a high-performing LinkedIn infographic. Visual reinforcement of the post's core insight. #### Specifications - **Dimensions:** 1200 × 627 px (1.91:1 widescreen). - **Aspect ratio:** 1.91:1. - **Safe zone:** Keep all critical content in the **center 1000 × 500 px** (≈100 px margin on all sides). The image is used both as a shared link preview and as an article header — corners may be cropped on different devices. - **No author photo.** Do not render or include any image of the author. - **No tool branding** like "LinkedIn Cover Image Generator" or any AI tool name on the image. - **Subtle branding:** A small handle (`@devengoratela`) or wordmark in a corner is fine but must not dominate. #### Design system (defaults — override with user brand colors if provided) | Element | Default | Purpose | |---|---|---| | Background | `#0a0e27` → `#1a1f4e` subtle gradient | Professional, eye-catching on light/dark feeds | | Hero text | `#ffffff`, 64–72 px, bold sans-serif | The core insight or hook | | Subtitle | `#a0aec0`, 22–26 px | Supporting context or category tag | | Accent shape | `#4f9eff` (LinkedIn blue variant) | Vertical bar on left edge or geometric shape | | Footer | `#718096`, 16–18 px | `@devengoratela` and post topic | | Font stack | `system-ui, -apple-system, "Segoe UI", Roboto, sans-serif` | Clean, native, no font loading required | #### SVG layout template ``` ┌─────────────────────────────────────────────────────────────┐ │ ▌ │ │ ▌ [CATEGORY / TOPIC TAG] │ │ ▌ │ │ [CORE INSIGHT — HERO TEXT] │ │ [wrapped to 2–3 lines max] │ │ │ │ [supporting visual — icon, stat, or shape] │ │ │ │ @devengoratela [year/date] │ └─────────────────────────────────────────────────────────────┘ ``` #### Process for `cover.svg` 1. Read the LinkedIn post from Stage 1. 2. Extract the **core insight** (the single biggest takeaway, usually 6–12 words). If the hook itself works, use it. 3. Extract the **category / topic** (e.g., "AI Tools", "Automation", "Career", "Prompting"). 4. Build the SVG with the layout above. Use `text-anchor="middle"` for centered hero text. Wrap hero text into 2–3 lines manually (use `` with `x` and `dy` for line breaks). 5. Validate the SVG: ```bash python3 -c "import xml.etree.ElementTree as ET; ET.parse('cover.svg'); print('Valid')" ``` Fix any errors and re-save until it parses cleanly. #### Process for `cover-prompt.md` In parallel with the SVG, write a detailed image generation prompt suitable for Midjourney, DALL-E 3, or Flux. Users can use this to upgrade to a higher-fidelity raster render. **`cover-prompt.md` template:** ```markdown # LinkedIn Cover Image Prompt Use this prompt with Midjourney, DALL-E 3, Flux, or any AI image generator. --- ## Prompt \``` [Full detailed prompt — subject, style, composition, lighting, palette, typography] \``` ## Specifications - **Dimensions:** 1200 × 627 px (1.91:1) - **Aspect ratio:** 1.91:1 - **Safe zone:** Keep all text in the center 1000 × 500 px area (corners may be cropped on mobile link previews) - **Style:** [e.g., "Modern minimal infographic, dark mode aesthetic"] - **Color palette:** [hex codes] - **Typography:** [e.g., "Bold sans-serif, large hero text"] ## Do NOT include - No photo of the author - No text that says "LinkedIn Cover Image Generator" or any tool name - No watermarks - No logo of any AI tool or platform ``` --- ### Stage 3: Skool Post **Role:** Casual content conversion assistant for the Practical AI Skool community. The LinkedIn post is the source material; the Skool post is a peer-to-peer reframe. #### Persona - You sound like a fellow AI enthusiast — casual, curious, genuinely helpful. - You are NOT a corporate content writer. No polished, formal, or salesy language. - You understand the Practical AI community values **real-world application over hype**. #### Conversion rules 1. **Tone:** Casual and practical — peer-to-peer, not guru-to-student. 2. **Strip LinkedIn-isms:** Remove professional posturing, humble-brags, and "thought leadership" phrases. 3. **Length:** Cut the LinkedIn post by 30–40%. Punchy and direct. 4. **Hashtags:** Remove all hashtags. 5. **Lead with the tool / tip / insight** — not the backstory or hook. 6. **Keep intact:** Any tools, prompts, tips, or frameworks. Reframe them as "here's what worked for me." 7. **"We / us" language:** Use occasionally — e.g., "for those of us using AI daily…" 8. **No hype / buzzwords:** Keep everything grounded and actionable. 9. **Tools in bold:** Use **bold** for tool names (e.g., **Cursor**, **Claude Code**, **n8n**). 10. **Closing question:** End with **one** strong, open-ended question relevant to AI learners, **bolded**, to spark discussion. #### Output format for `skool-post.md` ```markdown # [Hook — 1 line, casual, curiosity-driven] [Body — short paragraphs, 1–2 lines each] [Optional: a tool, prompt, or framework block — preserved from the LinkedIn post, reframed as personal experience] **[Strong open-ended question to spark discussion among AI learners]** ``` > **No "Practical AI Skool" branding in the post body itself.** The community is the destination, not a tagline in every post. --- ### Stage 4: Carousel (delegated to `/carousel-creator`) Stage 4 invokes the **`carousel-creator`** skill to produce a premium Instagram or LinkedIn carousel. Reusing the carousel-creator means we don't duplicate its design system, content strategy, and export pipeline — the subagent simply runs the skill with pre-collected intake info and lands the output in the parent pipeline's folder. #### Delegation pattern Spawn a subagent using the **Task tool** with `subagent_type: "general-purpose"`. The subagent's task prompt is self-contained and includes: 1. **Pre-collected intake** — topic, key points, tone, platform, handle, audience, brand colors, slide count (from Step 0). 2. **Parent folder path** — `{cwd}/{TS}/` — the same folder Stages 1–3 saved to. 3. **Existing files** — the subagent must know that `linkedin-post.md`, `cover.svg`, `cover-prompt.md`, `skool-post.md` already exist in the parent folder. 4. **Output directive** — "Save all carousel output (slides/, caption.md, PDF, MP4, Hyperframes MP4) into the parent folder `{TS}/`. Do not create a separate timestamped folder." #### Subagent task prompt template ``` You are running Stage 4 of the content-pipeline skill. Your job is to invoke the `carousel-creator` skill with the pre-collected intake below, then ensure all carousel output lands in the parent pipeline's folder. PARENT FOLDER: {cwd}/{TS}/ This folder already contains: - linkedin-post.md - cover.svg - cover-prompt.md - skool-post.md PRE-COLLECTED INTAKE (all fields are provided — do NOT re-ask the user): - Topic: {topic} - Key points: {key_points} - Tone: {tone} - Platform: {platform} - Handle: {handle} - Target audience: {audience} - Brand colors: {brand_colors_or_default} - Slide count: {slide_count_or_default} STEPS: 1. Invoke the `carousel-creator` skill (via the Skill tool). The skill's intake will be skipped because all fields are provided above. 2. The carousel-creator will create its own timestamped folder based on "actual current date and time". In most cases this matches {TS}, so the folder is shared naturally. If a minute boundary was crossed during the parent pipeline's run, the folder name will differ by one minute (e.g., 20260616-1649 vs 20260616-1650). Detect this with `ls -la {cwd}/` and compare timestamps. 3. If the carousel-creator's folder is a DIFFERENT folder than {TS}, move all its contents into {TS}/: - mv {TS_carousel}/* {TS}/ - rmdir {TS_carousel} 4. Run the carousel-creator's export scripts (PDF, MP4, Hyperframes MP4) per its own SKILL.md instructions. Confirm all four files end up in {TS}/. 5. Report back with the final list of files in {TS}/. Do NOT ask the user any intake questions — every field is pre-collected. ``` #### Why a subagent (not inline) - **Context cleanliness** — The carousel-creator's intake, design system, and export logic would otherwise clutter the main content-pipeline context. - **Pre-collected intake** — The subagent prompt has all the fields, so the carousel-creator's intake step is a no-op. The user is never asked the same question twice. - **Folder reconciliation** — The subagent can detect timestamp drift between the parent pipeline's `TS` and the carousel-creator's auto-generated folder, and move files to reconcile. #### Why "actual current date and time" is fine The carousel-creator's instruction to use "actual current date and time" is a default, not a hard constraint. Because: - content-pipeline computes `TS` at intake (Step 0) - Stage 4 runs within seconds to minutes of intake - The carousel-creator's invocation typically lands in the same minute - If a minute boundary is crossed, the subagent reconciles with a `mv` (Step 3 above) In 99% of runs, the timestamps match and no `mv` is needed. #### Output from Stage 4 After Stage 4, the parent folder contains everything from Stages 1–3 plus: - `slides/{slug}-slide-{n}.svg` — one SVG per slide - `caption.md` — carousel caption (also usable as a bonus LinkedIn post variant) - `{slug}.pdf` — multi-page PDF export - `{slug}.mp4` — MP4 slideshow (5 s/slide) - `{slug}-hyperframes.mp4` — Ken Burns + cross-dissolve cinematic MP4 --- ## Stage Handoffs Each stage must consume the previous stage's output: | Stage | Reads | Writes | |---|---|---| | 1. LinkedIn | `topic`, `key points`, `tone` | `linkedin-post.md` | | 2. Cover | `linkedin-post.md` (core insight + category) | `cover.svg`, `cover-prompt.md` | | 3. Skool | `linkedin-post.md` (full post) | `skool-post.md` | | 4. Carousel | pre-collected intake (platform, handle, audience, brand colors, slide count) + parent folder `{TS}/` | `slides/`, `caption.md`, `{slug}.pdf`, `{slug}.mp4`, `{slug}-hyperframes.mp4` (all in `{TS}/`) | If a stage needs clarification that the user did not provide, make a reasonable inference and note it. Do not stop the pipeline to ask. --- ## Quality Checklist Before reporting done, verify: **LinkedIn post:** - [ ] Hook is under 8 words - [ ] Re-hook is punchy and earns the expand - [ ] Single clear takeaway - [ ] No AI slop words (delve, landscape, unlock, etc.) - [ ] Soft CTA is a question or invitation - [ ] Repost + Follow lines present in correct order - [ ] Tone of all CTAs matches the article - [ ] 3–5 hashtags, no bold in the body **Cover image:** - [ ] SVG is 1200×627 with `viewBox="0 0 1200 627"` - [ ] Critical content sits inside the center 1000×500 safe zone - [ ] No author photo, no tool-name branding on the image - [ ] Subtle `@devengoratela` handle visible somewhere - [ ] SVG validates with `xml.etree.ElementTree.parse` - [ ] Companion `cover-prompt.md` exists with full prompt + do-not-include list **Skool post:** - [ ] 30–40% shorter than the LinkedIn post - [ ] No hashtags - [ ] No LinkedIn-isms (humble-brags, "thought leader" framing) - [ ] Tool names in **bold** where they appear - [ ] Ends with one bolded open-ended question - [ ] Body reads like a peer talking, not a brand broadcasting **Carousel (delegated to /carousel-creator):** - [ ] All five files (`slides/`, `caption.md`, `.pdf`, `.mp4`, `-hyperframes.mp4`) sit inside `{TS}/` - [ ] At least 5 SVG slides in `slides/` - [ ] CTA slide includes `@devengoratela` - [ ] No separate timestamped folder exists alongside `{TS}/` (timestamp drift reconciled via `mv`) - [ ] User was NOT re-prompted for carousel intake (all fields pre-collected) --- ## Final Behavior After all four stages complete: 1. Confirm every file is saved to the `YYYYMMDD-HHMM/` folder (Stages 1–3 by the main agent, Stage 4 by the carousel-creator subagent). 2. List the saved files with one-line descriptions. 3. Remind the user: - The LinkedIn post is copy-paste ready. - The SVG cover can be opened in any browser or imported into Figma. - The `cover-prompt.md` can be pasted into Midjourney / DALL-E for a higher-fidelity render. - The Skool post is ready to drop into the community. - The carousel `slides/`, `caption.md`, PDF, MP4, and Hyperframes MP4 are all in the same folder — ready to upload to Instagram / LinkedIn. **Do NOT ask for permission between stages.** Run all four in sequence from a single input.