The Pro Moves Pack.

Your first prompt is mid. Your fiftieth is a skill. The difference isn't talent — it's a loop you run on purpose. This is the feedback engine that turns prompts into skills, plus the map of where to go when you want to cross from operator into builder.

The 3-Use Test — Before You Promote Anything

Does it work at all?

First pass. Just checking it functions.

Does it work consistently?

Or did Use 1 just get lucky?

What do I edit every single time?

That same edit belongs in the prompt. The edit you keep making by hand belongs in the file, not in your hands.

The Self-Critique Add-On — Paste At The End

The single addition that turns a 7/10 into an 8.5. The formal, end-of-prompt version of the "score it" move — bolt it onto any prompt you care about.

After you produce your answer: 1. Score it 1-10: does it match my spec? Is it in my voice? Is anything weak, generic, or wrong? 2. If any score is below 8 - tell me why, and give me a v2 that fixes it. 3. End with one question I should ask myself before I ship this.

The Weekly Iteration Ritual — 15 Min, Every Friday

Ask	The move
Which prompts did I use most?	Those are your high-leverage candidates.
Which did I edit the same way every time?	Fold those edits into v2.
Any prompt used 5+ times that's stable?	Promote it to a skill.
Any skill untouched in 30 days?	Archive it — it wasn't solving a real problem.

The rewrite ruleWhen you iterate, rewrite — never append. A tight 250 words beats a bloated 800. A prompt gets worse when you bolt corrections onto the end. Strip it back and rewrite clean.

Two Fast Skill Wins

Skill Creator

Spent an hour teaching Claude a process? Type /skill-creator, hand it the whole chat, it builds the skill. Be specific or it won't fire — "weekly reports, headline metric first, 3 sections max," not "I write reports."

Clone-to-Customize

Plugin skills are read-only — but you're not stuck. Find one close, copy its instructions into a NEW skill, swap their tone for yours. Find it. Clone it. Tune it.

The Brutal Instagram Audit — Run It In Chrome

The reverse-engineer pointed at you. Claude reads your live profile as a first-time visitor and roasts it. Works because you're on the page, present — not a bot scraping in the background. Paste the whole thing with Claude in Chrome connected.

Act as a senior Instagram strategist and give me a brutally honest audit of my account. Use the Chrome connection to look at my live profile. Look at: 1. My bio, profile photo, name field, and link in bio 2. My grid - scroll the last 12 posts as a first-time visitor would 3. Click into 3 recent posts; read the full captions, hooks, CTAs 4. My Highlights covers and titles Then write the audit with these exact sections: BIO AUDIT - Is it clear? Does a stranger instantly get who I help, what I sell, and why to follow? What exact words would you change? GRID AUDIT - Looking at the last 12 together, what's the overall message? Cohesive? Does it make someone want to follow? Right vibe for a course creator? HOOK AUDIT - Read my first 1-2 caption lines. Are my hooks stopping the scroll? Quote them back and tell me what's weak. CONTENT MIX AUDIT - What % is lifestyle vs education vs personal vs selling? What should I post more of? Less of? THE ROAST - Top 5 things to fix THIS WEEK to grow faster and convert more followers into buyers. Be blunt. Don't be nice. I'd rather be embarrassed now than broke later.

The Creator Scrub — 5 Ideas Every Morning

The reverse-engineer on a timer. A scheduled task that scans your niche and hands you ~5 content ideas before coffee. Set it as a daily scheduled task. (Analytics pull = Meta connector; reading live posts = Claude in Chrome on an open page.)

Every morning, act as my content strategist and bring me 5 content ideas for today, pulled from what's actually working in my niche right now. For each idea give me: - The hook (first 1-2 lines, in my voice) - The angle / through-line - The format (reel, carousel, story) - WHY it'll land for MY audience specifically Rules: - Reverse-engineer the MECHANISM behind what's working, don't copy anyone's post - Match my voice and my offers, not generic advice - Rank the 5 by which I should post first and why End with one question that sharpens tomorrow's batch.

Want it fully automated?The "mess goes in a side window" cheat code has a full automated version called sub-agents — a second Claude that runs a sub-task in its own clean window and reports back just the answer. That's builder territory (Claude Code), and Anthropic Academy teaches it. You don't need it to run the scrub — just know it's the next rung.

The Delegation Kickoff — Stop Babysitting The Build

The fix for a 20-message correction marathon: make the decisions up front, where they're cheap, then let Claude run. Developers call this "plan mode." Your version: tell Claude not to build until it's sure, make it interview you, approve the plan, then release. Paste this as your first message on anything non-trivial — in a Project or in Cowork. Fill in the one bracket.

I want to delegate this to you properly, not babysit it. Here's what I want: [describe it, even roughly - e.g. "a simple landing page for my dog grooming business with a booking button, my hours, and a gallery of before/afters"] Before you build anything: 1. Don't create or change any files yet. Plan first. 2. Ask me questions until you're 95% sure exactly what I want. Don't assume. Ask about the parts where you'd otherwise guess: look and feel, the must-have pieces, anything I didn't mention, and what "done and good" looks like to me. A few at a time, then wait for my answers. 3. Once you have my answers, show me a clear step-by-step plan, including anything you're unsure about. Wait for me to approve. 4. After I approve, execute the whole plan. If you hit a real fork you didn't plan for, pause and ask. Otherwise run to the finish. 5. If I tell you you're off track, stop immediately and re-align. Don't finish the wrong version first. Start by asking me your questions.

The real winDelegating isn't losing control — it's the opposite. You make the calls up front where they're fast, then let Claude carry them out. Align, approve, release. If you run a certain task all the time, save this as a skill so you never retype it.

The Quarterly Memory Audit — Run It Every 90 Days

The monthly cleanup keeps Memory tidy. This one goes deeper — it makes Claude show you everything it believes about you, interview you on what's gone stale, and rewrite the picture. Drift between you-now and the-you-Claude-remembers is real, and it makes every output a little more off. Paste this into a normal chat with Memory on. About 10 minutes.

Run a memory audit on yourself, with me. The goal: clean up everything you have stored about who I am and what I do, so your future answers reflect the current me, not an outdated version. Run it in three rounds. ROUND 1 - AUDIT WHAT YOU REMEMBER Write out everything you currently believe about me, under these headings, one line each: - ROLE & WORK (what I do day to day) - TOOLS & STACK (the tools I use) - PRIORITIES & GOALS (next 90 days) - WORKFLOWS & HABITS (how I work, with whom) - COMMUNICATION PREFERENCES (how I like you to talk to me) - EXPLICIT INSTRUCTIONS (anything I've told you to always/never do) For each item, mark confidence: HIGH (I stated it), MEDIUM (inferred from a few chats), LOW (one-off, might be stale). ROUND 2 - INTERVIEW ME ON THE WEAK ONES Go through every LOW and MEDIUM item and ask me about it, one at a time, short and specific. Examples: - "I have you down as a [thing]. Still accurate? (Yes/No/Partly)" - "I have us using [tool] for [purpose]. Still true, or did you switch?" Wait for my answer before the next. After each, tell me in one line what changed. ROUND 3 - UPDATE AND REPORT Update your memory with the corrected picture. Then give me: - THE UPDATED PROFILE: a clean current version - WHAT CHANGED: the items that were stale or wrong - THREE RECOMMENDATIONS: the highest-leverage workflow changes I should make in the next 30 days, based on the new picture End by asking if I want to save this as a recurring quarterly habit.

Make it stickCalendar it — first Saturday of every quarter, 10 minutes. If you want it one-click, save the prompt as a skill (/skill-creator, hand it this prompt) and trigger it with "run my memory audit." Outputs get noticeably sharper the week after.

Where To Go Deeper — The Honest Map

This bootcamp makes you dangerous as an operator. If you want to cross into builder — the API, MCP servers, the dev cert — Anthropic teaches all of it free. Here's exactly where to go.

Anthropic Academy · free

anthropic.skilljar.com. ~19 self-paced courses, a certificate each, just an email to enroll. Start with Claude 101 and AI Fluency for small businesses.

Claude Certified Architect

Their first real credential. Proctored, 60 questions, closed-book, no AI assistance. Deeply technical — agentic architecture, MCP servers in Python, RAG. A developer cert, not a business one.

Want to learn…	Go to (Anthropic Academy)
Building MCP servers from scratch	Intro to MCP → MCP: Advanced Topics
The Claude API — building apps ON Claude	Building with the Claude API (~8 hrs)
Claude Code — the developer mode	Claude Code 101 → Claude Code in Action
Sub-agents (the full "side window" cheat code)	Introduction to Subagents
The proctored résumé credential	Claude Certified Architect (CCA)

Knowing the edge of what this bootcamp teaches is the operator skill — same as knowing the edge of what AI should do for you.