How I Built a Facebook Post Generator Skill in Claude Code (Full Walkthrough)

Build a reusable Claude Code skill that turns your brand book into Facebook posts on demand — no re-prompting required. Every marketer who's used Claude or ChatGPT has hit the same wall: every new conversation forgets your brand, your tone, your guardrails. You end up pasting the same 800-word brand brief into the prompt 30 times a week. This video shows you the way out. In this walkthrough I rebuild a Facebook post generator skill from scratch using Claude Code, plan mode, brainstorming, and MCP integrations. You'll see exactly how I feed Claude my brand identity, set guardrails, plan before I commit, dispatch parallel agents, and finally audit the output so the skill self-corrects. This is the workflow I now use across every Markitecht client engagement. Once the skill is built, generating a new post is a one-line command — no re-explaining who the brand is. ⏱ Time Stamp 0:00 Intro & bypass permissions 1:08 Feeding Claude your brand & memory 2:58 The 5-step content flow 4:13 Plan mode & brainstorming 5:55 Plan mode in action (Q&A) 8:30 Image/text generation & ENV credentials 9:33 Connecting MCPs & external apps 10:46 Risks, edge cases & plan approval 11:29 Agent swarm: parallel work 12:00 Running the skill end-to-end 12:54 Modular skills & speed vs. control 14:47 Auditing outputs & the QA loop 16:55 Wrap-up 🛠 What you'll learn • How to scaffold a Claude Code skill that reads from a brand book • When to use plan mode vs. just prompting • How to wire MCP servers + APIs (Gemini, image generation) via an ENV file • The difference between sequential vs. parallel agent dispatch • How to bake in a QA / audit layer so the skill catches its own mistakes • Modular skill architecture — letting one skill call another ❓ Common questions answered • What's the difference between a prompt and a skill? → Skills are saved, reusable, parameterized prompts with logic. Prompts are one-off. • Do I need to know how to code? → No. Claude Code writes the skill files for you in plan mode. • Is bypass permission safe? → Only when you scope Claude to specific folders. I cover this at 0:30. • Does this work for Instagram / LinkedIn / X? → Yes. Change the output format prompt and you're done. 🔗 Resources • Claude Code official: claude.com/claude-code • Model Context Protocol: modelcontextprotocol.io • Gemini API: ai.google.dev • My Markitecht agency: markitecht.co (case studies + the prompts I use) 🔔 Subscribe for more I publish a new Claude Code / AI marketing workflow every week. If you build with AI for a living (or want to), the bell is worth it. 💬 Tell me in the comments What skill should I build next? I'm taking requests. #ClaudeCode #AI #PromptEngineering #MCP #Automation #ContentCreation #MarketingAutomation #AIForMarketers #BrandVoice #ContentAI ```