CLAUDE CODE FULL COURSE 🤯
A hands-on tutorial building an AI marketing officer using Claude Code and the Blotato API. The instructor walks through installing Claude Code, creating custom skills for social media posting, adding brand voice, implementing quality gates via hooks, and using parallel sub-agents to plan and schedule a full week of content across multiple platforms.
Summary
The tutorial opens with an overview of what will be built: an end-to-end AI social media manager that can take a single sentence or URL, research a topic, draft a week's worth of content with visuals, present it for human review, and then schedule posts across multiple social platforms. The instructor emphasizes this is aimed at intermediate content creators who want to batch-produce content efficiently.
The first section covers Claude Code setup fundamentals drawn from the instructor's 500+ hours of experience. This includes installing the official VS Code extension, setting up permissions to avoid constant approval prompts for non-destructive commands, choosing between plan mode (used 90% of the time) versus edit-automatically mode, selecting the most capable model (Opus 4.6 for technical work), and clearing context between distinct features or bugs for better results. Pricing is addressed directly: the instructor recommends at minimum the $100/month plan, as the $20/month plan hits limits quickly with agentic tool calls.
The core of the tutorial involves building two custom Claude Code 'skills' — persistent, reusable command templates. The first skill, /post, accepts a topic, YouTube URL, TikTok link, PDF, or other inputs, extracts content via the Blotato API, generates platform-specific copy, optionally creates visuals (infographics, carousels), and publishes to social media. The instructor demonstrates this live, including debugging an API misinterpretation in real time to illustrate that errors are normal and self-correcting.
Brand voice is then added to the /post skill by feeding it writing samples from Alex Hormozi (for Twitter) and Justin Welsh (for LinkedIn) as stylistic references, plus a 'humanized AI' prompt to strip obvious AI writing patterns. The instructor notes that creators should ideally use their own samples but that modeling known influencers is a reasonable starting point.
Quality gates are implemented using Claude Code hooks — scripts that automatically run before any publish command. These checks enforce rules like character limits per platform, banned words, removal of em dashes, and required elements like a CTA or website URL. The instructor explains these can be extended to LLM-scored virality checks that rewrite posts if they fall below a threshold.
Parallel sub-agents are then introduced to enable simultaneous multi-platform publishing. Instead of posting to one platform at a time, typing 'all' spawns a separate sub-agent per platform, each adapting and publishing the content asynchronously — useful when scheduling 35+ posts in a single batch session.
The second skill, /plan-week, ties everything together. It accepts a topic or URL (demonstrated with a 2-hour Hormozi livestream), extracts content, generates 10 unique angles, drafts posts for each weekday on LinkedIn and Twitter, queues five whiteboard infographics, outputs a full content plan document for human review and editing within VS Code, and then schedules everything to the next open calendar slot via Blotato's slot system. The instructor notes the live demo accidentally published all posts immediately rather than scheduling them, and commits to fixing that before releasing the final prompts.
The session closes with creation of a CLAUDE.md file — a project memory document that records the project's purpose, API locations, brand voice references, conventions, and structural patterns so that each new Claude Code session starts with full context rather than a blank slate. The instructor frames this as essential discipline for any ongoing Claude Code project.
Key Insights
- The instructor spends approximately 90% of coding time in plan mode so Claude cannot accidentally change anything, only switching to 'edit automatically' mode once fully confident in the plan — a discipline he argues prevents costly mistakes in complex codebases.
- The instructor argues that adding the line 'ask me clarifying questions one at a time until you're 95% confident you can complete the task successfully' to any prompt forces Claude to surface blind spots and edge cases the user missed, making lazy or incomplete prompts workable.
- The instructor distinguishes Claude Code from n8n and workflow automation tools, arguing Claude Code is weak at orchestration, pre-built integrations, and scheduled triggers, but that combining Claude Code with n8n via MCP is a strong middle ground that captures benefits of both.
- The instructor argues that hooks (quality gates) can be extended beyond simple rule checks to include LLM-scored virality: posts could be scored against a database of 200 viral posts, and any post scoring below a threshold would be automatically rewritten until it passes — a more sophisticated automated quality loop.
- The instructor warns that full automation of content creation is inappropriate for most people starting out, arguing that if you don't already know how to make good content, setting-and-forgetting an AI pipeline will not teach you — the value is in using the system to handle logistics while the human focuses time on improving the actual drafts.
Topics
Full transcript available for MurmurCast members
Sign Up to Access