$ man content-wiki/content-skills

Agent skills are structured instructions that tell the AI agent in your code editor how to perform a specific task. For content, this means: generate a LinkedIn post from a screenshot, convert a draft to platform-ready text, create a TikTok script from a topic, generate terminal-style images, and push to publishing platforms. Each skill encapsulates the full workflow for one content type — inputs, voice loading, generation rules, output format, and publishing steps. The power: instead of prompting an AI with write me a LinkedIn post and hoping for the best, you invoke a skill that loads the right voice files, follows the right format structure, applies the right anti-slop rules, and outputs exactly what you need. Consistency through codified process, not repeated prompting.

PATTERN

final-copy: takes a markdown draft and converts it to platform-ready plain text. Runs voice normalization, strips AI slop, applies formatting rules. The last step before publishing. play-draft: turns a screenshot of something you built into LinkedIn and X content drafts. Analyzes the screenshot, identifies the play or accomplishment, generates drafts in the voice system's style. skill-play: converts a Cursor skill you use into LinkedIn and X content for the skills I use everyday series. Reads the skill file, extracts the key workflow, generates drafts that explain it. tiktok-script: generates a 16-second TikTok script from a topic or one-sentence description. Follows the hook-demo-result-loop structure from the TikTok playbook. substack-post: drafts Substack newsletter posts with context-aware post numbering and structure selection. final-substack: finalizes a Substack draft and pushes it to Substack via MCP. Runs voice normalization, saves locally, creates the draft on the platform. aios-image and x-tip-image: generate Python Pillow terminal-style images for articles and X tip posts respectively.

CODE

Every content skill follows the same architecture. A SKILL.md file in .cursor/skills/ contains: metadata (name, description, trigger phrases), context loading instructions (which voice files to read first), input requirements (what the skill needs from the user), generation rules (step-by-step instructions for the agent), output format (where to save, what format, file naming conventions), and optional publishing steps (MCP calls, scheduling). The skill file is essentially a detailed recipe. The AI agent reads it and executes each step. Because the recipe is explicit, the output is consistent. Because the recipe lives in the repo, it is versioned and improvable. When you notice a skill producing suboptimal output, you edit the SKILL.md — not your prompting strategy — and every future invocation reflects the improvement.

PATTERN

To create a new content skill: (1) Identify a repeating content workflow that follows the same pattern every time. (2) Document the inputs — what does the skill need from you (a topic, a screenshot, a draft file, a partner name). (3) List the voice files to load — which tier-1, tier-2, and tier-3 files are relevant. (4) Write the generation instructions step by step — be explicit about format, length, structure, and style. (5) Define the output — where does the file get saved, what is the filename convention, what format. (6) Add optional automation — MCP calls for publishing, image generation for visuals. The test for whether something should be a skill: if you have done it manually more than 3 times and the process was the same each time, it should be a skill. The upfront cost of writing the SKILL.md is 30-60 minutes. The ongoing savings are 15-30 minutes per content piece, forever.

PRO TIP

Each new skill reduces the time between idea and published content. The first skill saves 15 minutes per post. The tenth skill means your entire pipeline — from idea capture to multi-platform publishing — runs in minutes instead of hours. The compound effect is not just time savings. It is consistency. Every post goes through the same voice normalization, the same anti-slop scan, the same format checks. The quality floor rises because the process catches errors that manual workflows miss. After 20+ content skills, the content operating system effectively runs itself. You provide the idea and the approval. The system handles everything in between.