📱 AI Advice Hub

54 TikToks · 6 categories · evaluated & expanded · April 2026

Honest Output
🎯
The Honest Mentor System Prompt
Stop Claude from agreeing with everything — paste into Global Instructions
⭐ High Value▼

By default Claude is trained to be agreeable. This system prompt resets that. Works best in Settings → Global Instructions so it applies to every session.

Act as a rigorous, honest mentor. Do not default to agreement. Identify weaknesses, blind spots, and flawed assumptions. Challenge ideas when needed. Be direct and clear, not harsh. Prioritize helping me improve over being agreeable. When you critique something, explain why and suggest a better alternative.
💡 In Cowork: Settings → Cowork → Edit Global Instructions → paste this.
Power Prompting Techniques
⚡
8 Framing Tricks That Dramatically Improve Output
These work because of how attention mechanisms weight tokens
✅ Solid▼
01
Reference a past conversation
Start a new chat and say Claude explained X yesterday. It avoids contradicting itself and goes deeper.
"You explained React hooks yesterday, but I forgot the part about useEffect"
02
Assign a specific IQ / expertise
Change the number for wildly different quality. 160 gives you principles you've never heard of.
"You're an IQ 145 specialist in marketing. Analyze my campaign."
03
Use "Obviously…" as a trap
State something that might be wrong as obvious. Claude corrects and explains nuances it'd otherwise skip.
"Obviously Python is better than JavaScript for web apps, right?"
04
Pretend there's an audience
Structure changes completely. Adds emphasis, examples, anticipates questions.
"Explain this like you're teaching a packed auditorium"
05
Give a fake constraint
Forces creative thinking. Any domain works — sports, music, cooking.
"Explain this using only kitchen analogies"
06
"Let's bet $100"
Imaginary stakes make Claude scrutinize harder. It hedges, reconsiders, thinks through edge cases.
"Let's bet $100: Is this code efficient?"
07
Introduce disagreement
Forces it to build a strong defense or concede quality. Real analysis either way.
"My colleague says this is wrong. Defend it or admit they're right."
08
Request "Version 2.0"
Treats output like a sequel that needs to innovate, not polish. Much bigger thinking.
"Give me a Version 2.0 of this idea"
Context Management — Read This
🔥
Context Rot: Why Claude Gets Worse Mid-Session
Degradation starts at ~40% context fill — here's why and what to do
⭐ High Value▼

Claude's context window is a spotlight, not storage. More tokens doesn't mean Claude remembers more — the spotlight gets dimmer. This is sometimes called "context rot."

Symptoms of context rot:

Symptom 1: Claude forgets decisions made 20 messages ago and suggests something different.
Symptom 2: Responses go generic — specific details from earlier stop appearing.
Symptom 3: Your system prompt loses influence. It's still there, Claude just attends to it less.
Symptom 4: Contradictions appear within the same session.
Degradation starts at ~40% context fill. NVIDIA's RULER benchmark: effective context is 50–65% of advertised window. A 200K window becomes unreliable around 130K tokens.

Fixes in Cowork:

  1. Start a new session when switching tasks. Reference your key decisions at the top of the new session so Claude has them fresh.
  2. Write a spec file for complex projects. A well-written 1-page brief in your workspace beats re-explaining across 30 messages.
  3. Use your memory files (decisions.md, preferences.md) — Claude reads these fresh every session, bypassing the rot entirely.
  4. Keep Global Instructions tight. Over 300 words = token tax on every request. Trim ruthlessly.

If you use Claude Code (CLI):

The Claude Code CLI has /compact (compress history), /clear (full reset), and /rewind (roll back state). These are developer CLI commands — not available in Cowork or the web interface.
⚠️ Rule of thumb: Start a fresh session when you hit 3+ debug cycles in a row, switch topics entirely, or notice Claude losing context of decisions made earlier.
Setup & Workflow
📁
The Folder Structure That Unlocks Cowork
Set up once, stop re-explaining yourself forever
⭐ High Value▼
  • ABOUT ME/ — about-me.md (background), my-voice.md (tone + phrases you never use), my-rules.md (ask first, show a plan)
  • PROJECTS/ — one subfolder per project with a brief.md inside
  • TEMPLATES/ — reusable prompt templates for your repeated tasks
  • OUTPUTS/ — everything Claude generates, never your originals
  • MEMORY/ — decisions.md, people.md, preferences.md (read every session)

Global Instructions (paste this exactly):

Always read my ABOUT ME files first. Never edit my originals — deliver everything to OUTPUTS. Show a plan before you act. Never delete without approval. Ask questions before executing.
📈
The 4 Levels of Claude Mastery
Most people never leave Level 1
✅ Solid▼
Level 1 · Claude Chat
Ask anything, get a response
Claude.ai or the mobile app. Use Opus 4.6 + Extended Thinking for hard problems. Better prompt: "I want to [TASK] for [SUCCESS CRITERIA]"
Level 2 · Cowork (Desktop App)
Claude lives on your computer, reads your actual files
Creates real docx, pptx, xlsx files in your folder. Set Global Instructions once. Connect MCPs for Slack, Gmail, Notion, and more.
Level 3 · Skills + Plugins
Claude builds repeatable workflows for your most-used tasks
"Use the skill-creator to build a skill for [your most repeated task]." Claude interviews you, builds it, tests it. Fires automatically whenever you describe that task.
Level 4 · Automation + Agents
Claude runs parts of your business
Scheduled tasks, artifacts that refresh live data, connected tools. Claude works your task list while you sleep. Some features require MCP connectors or developer setup.
⏱️
Never Hit Rate Limits — 5 Habits
Keep yourself in the fast lane
Tips▼
  • Edit, don't restart. Edit previous messages rather than starting a new chat.
  • New chat every ~15 messages. Shorter sessions = more headroom per session.
  • Batch into one message. Combine follow-ups. One message, one response.
  • Turn off Extended Thinking when you don't need deep reasoning — it burns tokens fast.
  • Pace your day. Don't burn your quota in the first session of the morning.
Model choice: Haiku 4.5 for quick tasks · Sonnet 4.6 for most work · Opus 4.6 for complex reasoning + Extended Thinking.
12 McKinsey-style prompts. Replace [PLACEHOLDERS] with your business details and paste into a fresh Claude chat.
1
Market Sizing & TAM Analysis
McKinsey market analyst
Investor-ready▼
You are a McKinsey-level market analyst. I need a Total Addressable Market (TAM) analysis for [YOUR INDUSTRY/PRODUCT]. Please provide: • Top-down approach: Start from global market → narrow to my segment • Bottom-up approach: Calculate from unit economics × potential customers • TAM, SAM, SOM breakdown with dollar figures • Growth rate projections for the next 5 years (CAGR) • Key assumptions behind each estimate • Comparison to 3 analyst reports or market research firms Format as an investor-ready market sizing slide with clear methodology. Context: My product is [DESCRIBE PRODUCT], targeting [TARGET CUSTOMER] in [GEOGRAPHY].
2
Competitive Landscape Deep Dive
Bain & Company strategy consultant
Comparison tables▼
You are a senior strategy consultant at Bain & Company. I need a complete competitive landscape analysis for [YOUR INDUSTRY]. Please provide: • Direct competitors: Top 10 players ranked by market share, revenue, and funding • Indirect competitors: 5 adjacent companies that could enter this market • For each competitor: pricing model, key features, target audience, strengths, weaknesses, recent strategic moves • Market positioning map (price vs. value matrix) • Competitive moats: What makes each player defensible • White space analysis: Gaps no competitor is filling • Threat assessment: Rate each competitor (low/medium/high threat) Format as a structured competitive intelligence report with comparison tables. My company: [DESCRIBE YOUR BUSINESS AND POSITIONING]
3
Customer Persona & Segmentation
World-class consumer research expert
4 detailed personas▼
You are a world-class consumer research expert. I need deep customer personas for [YOUR PRODUCT/SERVICE]. Please build 4 detailed personas, each with: • Demographics: Age, income, education, location, job title • Psychographics: Values, beliefs, lifestyle, personality traits • Pain points: Top 5 frustrations they experience daily • Goals & aspirations: What success looks like for them • Buying behavior: How they discover, evaluate, and purchase • Media consumption: Where they spend time online and offline • Objections: Top 3 reasons they'd say no to my product • Triggers: What moment makes them actively search for a solution • Willingness to pay: Price sensitivity analysis per segment Also provide: Segment sizing (% of total market) and prioritization matrix. My product: [DESCRIBE PRODUCT] in [INDUSTRY]
4
Industry Trend Analysis
Goldman Sachs Research analyst
Impact ratings 1–10▼
You are a senior analyst at Goldman Sachs Research. I need a comprehensive trend report for the [YOUR INDUSTRY] sector. Please provide: • Macro trends: 5 global forces shaping this industry (economic, regulatory, tech, social, environmental) • Micro trends: 7 emerging patterns from the last 12 months • Technology disruptions: What new tech is changing the game and when it hits mainstream • Regulatory shifts: Upcoming legislation or policy changes to watch • Consumer behavior changes: How buyer preferences are evolving • Investment signals: Where smart money is flowing (VC deals, M&A, IPOs) • Timeline: Map each trend to short-term (0–1yr), mid-term (1–3yr), long-term (3–5yr) • "So what" analysis: What each trend means for a company like mine Format as a trend intelligence brief with impact ratings (1–10) for each trend. My company operates in: [DESCRIBE YOUR BUSINESS AND MARKET]
5
SWOT + Porter's Five Forces
Harvard Business School strategy professor
Full framework▼
You are a Harvard Business School strategy professor. I need a combined SWOT and Porter's Five Forces analysis for [YOUR COMPANY/PRODUCT]. For SWOT, provide: • Strengths: 7 internal advantages with evidence • Weaknesses: 7 internal limitations with honest assessment • Opportunities: 7 external factors we can exploit • Threats: 7 external factors that could harm us • Cross-analysis: Match strengths to opportunities (SO strategy) and identify threat-weakness combos (WT risks) For Porter's Five Forces: • Supplier power: Who are our key suppliers and how much leverage they have • Buyer power: How much negotiating power our customers have • Competitive rivalry: How intense competition is and what drives it • Threat of substitution: What alternatives exist beyond direct competitors • Threat of new entry: How easy is it for new players to enter Rate each force (1–10) and provide overall industry attractiveness score. My business: [DESCRIBE COMPANY, PRODUCT, INDUSTRY, STAGE]
6
Pricing Strategy Analysis
Fortune 500 pricing consultant
With $ recommendations▼
You are a pricing strategy consultant who has worked with Fortune 500 companies. I need a comprehensive pricing analysis for [YOUR PRODUCT/SERVICE]. Please provide: • Competitive pricing audit: Map all competitor prices, tiers, and packaging • Value-based pricing model: Calculate price based on customer value delivered • Cost-plus analysis: Determine floor price from cost structure • Price elasticity estimate: How sensitive is demand to price changes • Psychological pricing strategies: Anchoring, charm pricing, decoy strategies • Tiering recommendation: Design 3 pricing tiers with feature allocation • Discount strategy: When to discount, how much, and for whom • Revenue projection: Model 3 pricing scenarios (aggressive, moderate, conservative) • Monetization opportunities: Upsells, cross-sells, usage-based pricing Format as a pricing strategy deck with specific dollar recommendations. My product: [DESCRIBE PRODUCT, CURRENT PRICE, TARGET CUSTOMER, COST STRUCTURE]
7
Go-To-Market Strategy
Chief Strategy Officer, 20+ product launches
GTM Playbook▼
You are a Chief Strategy Officer who has launched 20+ products across B2B and B2C markets. I need a complete go-to-market plan for [YOUR PRODUCT]. Please provide: • Launch phasing: Pre-launch (60 days), Launch (week 1), Post-launch (90 days) • Channel strategy: Rank the top 7 acquisition channels by expected ROI • Messaging framework: Core value proposition, 3 supporting messages, proof points • Content strategy: What content to create for each stage of the funnel • Partnership opportunities: 5 strategic partners that could accelerate growth • Budget allocation: How to split a [BUDGET] marketing budget across channels • KPI framework: 10 metrics to track with target benchmarks • Risk mitigation: Top 5 launch risks and contingency plans • Quick wins: 3 tactics that can generate traction within the first 14 days Format as an actionable GTM playbook with timelines and owners. My product: [DESCRIBE PRODUCT, MARKET, BUDGET, TIMELINE]
8
Customer Journey Mapping
Top-tier customer experience strategist
Full lifecycle▼
You are a customer experience strategist at a top consulting firm. I need a complete customer journey map for [YOUR PRODUCT/SERVICE]. Please map every stage of the customer lifecycle: • Awareness: How do they first discover us? What triggers the search? • Consideration: What do they compare? What information do they need? • Decision: What makes them convert? What almost stops them? • Onboarding: First 7 days experience that builds or kills retention • Engagement: What keeps them coming back? Key activation moments? • Loyalty: What turns users into advocates? Referral triggers? • Churn: Why do they leave? Early warning signals? For each stage provide: • Customer actions, thoughts, and emotions • Touchpoints (digital and physical) • Pain points and friction moments • Opportunities to delight + key metrics to track Format as a detailed journey map with emotional curve visualization described in text. My business: [DESCRIBE PRODUCT, CUSTOMER TYPE, CURRENT CONVERSION RATE]
9
Financial Modeling & Unit Economics
VP Finance at high-growth startup
With benchmarks▼
You are a VP of Finance at a high-growth startup. I need a complete unit economics and financial model for [YOUR BUSINESS]. Unit economics breakdown: • Customer Acquisition Cost (CAC) by channel • Lifetime Value (LTV) calculation with assumptions • LTV:CAC ratio and payback period • Gross margin per unit/customer • Contribution margin analysis 3-year financial projection: • Revenue model (monthly for year 1, quarterly for years 2–3) • Cost structure breakdown (fixed vs. variable) • Break-even analysis: when and at what volume • Cash flow forecast with burn rate • Sensitivity analysis: best case, base case, worst case • Key assumptions with justification for each • Benchmark comparison: How do my metrics compare to industry standards • Red flags: What numbers should worry me and trigger action Format as a financial model summary with clear tables and formulas. My business: [DESCRIBE BUSINESS MODEL, CURRENT REVENUE, COSTS, GROWTH RATE]
10
Risk Assessment & Scenario Planning
Deloitte risk management partner
15 risks + 4 scenarios▼
You are a risk management partner at Deloitte. I need a comprehensive risk analysis and scenario plan for [YOUR BUSINESS/PROJECT]. Risk identification — List 15 risks across: • Market risks (demand shifts, competition, pricing pressure) • Operational risks (supply chain, talent, technology failures) • Financial risks (cashflow, currency, funding gaps) • Regulatory risks (compliance, policy changes, legal exposure) • Reputational risks (PR crises, customer backlash, data breaches) For each risk provide: • Probability rating (1–5) + Impact severity rating (1–5) • Risk score (probability × impact) + Early warning indicators • Mitigation strategy + Contingency plan if risk materializes Scenario planning: • Best case: What goes right and what it looks like • Base case: Most likely outcome • Worst case: What could go wrong simultaneously • Black swan: The unlikely event that changes everything • For each: Revenue impact, timeline, strategic response Format as an executive risk report with a prioritized risk matrix. My business context: [DESCRIBE BUSINESS, STAGE, KEY DEPENDENCIES]
11
Market Entry & Expansion Strategy
Global expansion strategist, 30+ markets
12-month roadmap▼
You are a global expansion strategist who has helped companies enter 30+ new markets. I need a market entry analysis for expanding [YOUR BUSINESS] into [TARGET MARKET/GEOGRAPHY]. Market attractiveness scoring (1–10 each, weighted total): • Market size and growth rate • Competitive intensity • Regulatory environment • Customer accessibility + Infrastructure readiness Entry mode analysis — Evaluate and recommend between: • Direct entry (build from scratch) • Partnership/joint venture • Acquisition • Licensing/franchise • Digital-first entry • For each: Pros, cons, cost, and timeline Localization requirements: • Product/service adaptations needed • Pricing adjustments for local purchasing power • Cultural considerations for marketing • Legal, compliance, talent, and operational needs 12-month entry roadmap: Month-by-month action plan with milestones Investment requirement: Budget estimate with resource allocation Success metrics: KPIs for first 6 and 12 months My business: [DESCRIBE CURRENT BUSINESS, TARGET MARKET, AVAILABLE RESOURCES]
12
Executive Strategy Synthesis ⭐ The Master Prompt
McKinsey senior partner to CEO — use this first
Use this first▼
💡 Give this full context about your business for best results. This is the one to start with.
You are the senior partner at McKinsey & Company presenting to a CEO. I need you to synthesize everything about [YOUR BUSINESS] into one strategic recommendation. Please provide: • Executive summary: 3-paragraph overview a CEO can read in 2 minutes • Current state assessment: Where the business is today (be brutally honest) • Strategic options — 3 distinct paths forward: - Option A: Conservative/low-risk approach - Option B: Balanced growth approach - Option C: Aggressive/high-risk approach For each option: Expected outcome, investment required, timeline, key risks • Recommended strategy: Your top pick with clear reasoning • Priority initiatives: The 5 highest-impact actions in the next 90 days, ranked • Resource requirements: People, money, and tools needed • Decision framework: A simple matrix for the next 10 strategic decisions • "If I only had 1 hour" brief: The single most important insight and action Format as a McKinsey-style strategy deck summary with clear recommendations. My business: [PROVIDE FULL CONTEXT — product, market, stage, team size, revenue, goals, biggest challenge]
These are the built-in superpowers of Cowork — no developer setup needed. Just describe your task and the right skill triggers automatically.
Built-in Cowork Skills (Say the task, Claude triggers the skill)
📄
pdf
Create, extract, merge, or fill PDF forms. "Turn this into a PDF" is enough.
📝
docx
Professional Word documents with tables, headings, page numbers, letterhead.
📊
xlsx
Excel spreadsheets with formulas, charts, data analysis, budgets.
📋
pptx
PowerPoint decks, pitch presentations, slide content from any source.
⏰
schedule
Create a scheduled task that runs automatically on an interval.
🛠️
skill-creator
Build custom skills for your most-repeated tasks. Claude interviews you and builds it.
💡
How to Trigger Skills
Just describe the task — Claude recognizes which skill to use
Start Here▼

Skills work automatically based on what you ask for. You don't need to name them.

  • "Write a report on…" → Claude uses the docx skill → saves a real .docx to your folder
  • "Make a presentation about…" → pptx skill → saves a real .pptx file
  • "Build me a spreadsheet for…" → xlsx skill → saves a real .xlsx with formulas
  • "Save this as a PDF" → pdf skill → creates a formatted PDF
  • "Remind me every Monday to…" → schedule skill → sets up a recurring task
  • "Build a skill for my most-repeated task" → skill-creator → interviews you, builds it
💡 Build your own custom skills with skill-creator. Say: "Use skill-creator to build a skill for [your most repeated task]." Claude will interview you about your process, then create a reusable skill that fires automatically whenever you describe that task in the future.
MCP Connectors (Connect Claude to your tools)
💬 Slack
📧 Gmail
📁 Google Drive
📓 Notion
✅ Asana
🐙 GitHub
☁️ Cloudflare
📅 Google Calendar
🗂️ Airtable
🔗 Linear
💾 Postgres / MySQL
➕ 100+ more
🔌
What MCPs Do & How to Install Them
Connect Claude to the tools you already use
High Impact▼

MCP (Model Context Protocol) connectors let Claude read and act on data from your real tools — Slack messages, Gmail threads, Notion pages, GitHub issues, databases, and more.

How to install:

  1. In Cowork, click Settings → Plugins or the plugin icon in the sidebar.
  2. Browse the marketplace or search for the tool you want to connect.
  3. Click Install. Authenticate with OAuth or paste an API key.
  4. Done. Claude can now read from and act on that service.

What you can do once connected:

  • Slack: "Summarize what happened in #general this week." "Draft a message to the team about the launch."
  • Gmail: "Find all emails about the Johnson account and summarize them." "Draft a reply to the last email from Sarah."
  • Notion: "Pull the project brief from Notion and turn it into a presentation."
  • Asana/Linear: "What tasks are assigned to me? What's overdue?"
  • GitHub: "Summarize open PRs and flag anything that's been stale for over a week."
MCPs are read-write where you grant permission. Claude will always ask before sending messages or modifying data.
🪄
Artifacts — Pages You Can Reopen Anytime
Turn a one-time answer into a live dashboard you check repeatedly
⭐ High Value▼

When Claude answers a question that you'll want to check again — open tasks, weekly metrics, project status — ask it to turn the answer into an artifact. This creates a persistent HTML page that lives in your Cowork sidebar and can pull fresh data each time you open it.

When to request an artifact:

  • Any status page you'll check more than once (project tracker, hiring pipeline, support queue)
  • Recurring reports — weekly metrics, team digest, budget summary
  • Interactive data explorers — filter tasks by status, search a table, drill into records
  • Any answer Claude just gave as a markdown list that you'll want refreshed tomorrow

How to request one:

Turn this into a live artifact I can reopen anytime. Pull fresh data from [TOOL] each time it loads.
💡 This hub (AI Advice Hub) is itself an artifact — built once, reopened forever, always current.
⏰
Scheduled Tasks — Set It and Forget It
Native Cowork feature — no developer setup required
No Code Needed▼

Scheduled tasks let Claude run a job automatically on a schedule — daily summaries, weekly reports, recurring reminders — without any developer setup or cron jobs.

What to say:

Every Monday morning, pull my open Asana tasks and send me a summary Slack message with what's due this week.
Every evening at 6pm, check my calendar for tomorrow and draft a briefing note in my OUTPUTS folder.

Good use cases:

  • Weekly digest of Slack / email highlights
  • Daily agenda pulled from Google Calendar
  • Monday morning task review from Asana or Linear
  • Monthly metrics report from a connected database
  • Recurring reminders to review decisions.md entries
Requires at least one MCP connector (Slack, Calendar, etc.) for the task to pull live data. File operations (read/write to your workspace) work without any connector.
📎
File Uploads & Multimodal
Attach images, PDFs, spreadsheets — Claude analyzes them directly
Underused▼

Claude can read and analyze files you attach — screenshots, photos, PDFs, spreadsheets, CSV files, and more. This is one of the most underused features.

  • Screenshot of an error → "What's wrong here and how do I fix it?"
  • Photo of a whiteboard → "Turn this into a structured action list."
  • PDF contract → "Summarize the key obligations and any red flags."
  • Excel file → "Analyze this data and identify trends."
  • Screenshot of a UI → "What UX issues do you see and how would you fix them?"
  • Photo of handwritten notes → "Transcribe and organize these."
💡 In Cowork, Claude also has direct access to your connected workspace folder — no need to upload files that are already in your folder. Just say "look at [filename]" and Claude reads it.
⚠️ The cron job and Gmail API prompts require Claude Code (CLI) or developer setup. The memory system and decision logger work great in regular Cowork.
Memory & Persistence
🧠
Never Forget Again — Persistent Memory System
Claude remembers every decision, preference, and person across sessions
⭐ High Value▼
Create a persistent memory system for yourself. Make a /memory directory with files organized by category: decisions.md, people.md, preferences.md, and user.md. Write a CLAUDE.md that instructs you to read these files at the start of every session and update them at the end.
In Cowork: Create a MEMORY/ folder in your workspace with those 4 files, then add "Always read my MEMORY/ files at the start of our session and update them before we end" to your Global Instructions.
📋
Hold Yourself Accountable — Decision Logger
Every decision logged with reasoning + a 30-day review date
✅ Solid▼
Create a decision logging system. Every time I describe a decision I'm making, log it to decisions.csv with: date, decision, reasoning, expected outcome, and a 30-day review date. Set up a cron job that checks daily if any decisions have hit their review date and appends a "REVIEW DUE" flag. Build a review.sh script that surfaces only those flagged items.
Simpler Cowork version: Ask Claude to maintain a decisions.md in your workspace. Each decision gets logged with date + reasoning + 30-day review date. Add to Global Instructions: "When I mention making a decision, log it to decisions.md."
Inbox & Task Management
📬
Clean Up Your Inbox — AI Email Triage
Hourly scan, auto-labels, draft replies. Never auto-sends.
🔧 Requires Gmail API▼
Build an hourly cron job that scans my Gmail inbox, triages new emails into URGENT / NEEDS REPLY / FYI / JUNK, auto-labels using my existing labels, and saves drafted replies for URGENT and NEEDS REPLY emails — tone-matched to my last 20 sent emails. Never auto-send. Store state in emails_processed.json. Log to inbox_manager.log.
Simpler Cowork version: Connect the Gmail MCP, then ask: "Check my inbox, triage the last 20 emails into Urgent / Needs Reply / FYI, and draft replies for the urgent ones."
✅
Build a Task Agent — Claude Works Your To-Do List
Autonomous priority-ordered task execution with full logging
🔧 Technical Setup▼
Build a to-do dashboard with a local web UI. I can add, edit, and delete tasks with priority levels. Set up an hourly cron job — if the task list is not empty, work through tasks autonomously starting with highest priority, log what was done, and mark complete. Store tasks in tasks.json. Log all activity to tasks.log.
For Developers (Claude Code CLI)
⚙️
Claude Code Workflow Orchestration Rules
Senior-engineer-level behavior for any coding project
🔧 Claude Code CLI▼
  1. Plan Mode Default: Use for any non-trivial task (3+ steps). Stop and re-plan immediately if something goes wrong.
  2. Subagent Strategy: Use subagents to keep main context clean. Offload research and parallel analysis.
  3. Self-Improvement Loop: After every session, update tasks/lessons. Iterate on failing tests until they pass.
  4. Verify Before Done: Diff changes, run tests, check logs. Demonstrate correctness before claiming complete.
  5. Demand Elegance: Ask "Is there a more elegant way?" Don't push unseen solutions.
  6. Zero Handholding: Fix bugs at root cause. Zero context switching required from user.
  7. Task Management: Write a checkable plan first. Explain changes. Capture lessons.
🔒
7-Step Production Hardening Framework
"Vibe coding is the starting line, not the finish line"
🔧 Developers▼
  1. Extract Architecture — Claude reverse engineers its own code into a master markdown file.
  2. Boundary Testing — Force failing unit tests for all auth and payment paths.
  3. Establish Telemetry — Integrate error logging (Sentry etc.) to catch silent failures.
  4. Refactor for Modularity — Break monolithic files into single-responsibility modules.
  5. Sanitize Dependencies — Audit every imported library for vulnerabilities.
  6. Document Deployment — Create a containerized build script.
  7. Threat Model — Run an adversarial AI agent to find injection flaws before hackers do.
All repos require Claude Code (CLI) or a developer environment. Links below are verified as of April 2026 — search GitHub by name if a URL changes.
Browser Automation
🌐 Browser-use ⭐⭐
Claude controls a real browser — clicks, fills forms, screenshots, parallel sessions. 84K+ stars. The most battle-tested browser automation for Claude.
browser-use.com
🔄 n8n-MCP
Connects Claude Code to n8n automations. Claude tests and validates workflows. Great bridge between Claude and n8n's 400+ integrations.
github.com/czlonkowski/n8n-mcp
Multi-Agent Orchestration
🤖 oh-my-claudecode
Multi-agent autopilot. One agent plans, one codes, one reviews. Automatically picks the cheapest model for each role.
github.com/Yeachan-Heo/oh-my-claudecode
👥 Agency Agents
Expert personas you can switch between with one command — frontend dev, sales engineer, UX designer, and more. Instant role-switching.
github.com/msitarzewski/agency-agents
Design Quality
✨ Impeccable
Teaches Claude real design intelligence — typography, spacing, color theory. Eliminates generic school-project-looking UIs. Essential for front-end work.
github.com/pbakaus/impeccable
Memory
🧠 claude-mem
Persistent memory across sessions. Captures and compresses coding sessions, then injects relevant context into future sessions automatically.
github.com/thedotmack/claude-mem
⚠️ Always check the last commit date and open issues before installing anything. Repos go stale. Best starting points with strong community support: Browser-use and Impeccable.
These techniques weren't in the TikTok collection — but they're from Anthropic's own prompting documentation and are among the highest-ROI things you can do. See docs.claude.com for more.
1
Ask Claude to Question YOU First
Probably the single highest-ROI technique nobody talks about
Impact

Before Claude starts, have it identify what it actually needs to know. This eliminates the biggest source of bad output: Claude making assumptions instead of asking.

❌ Without It
"Write me a marketing email." → Generic, wrong audience, wrong tone, wrong CTA.
✅ With It
Claude asks: Who's the audience? What action? Warm or formal? Then writes something that actually works.

The prompt:

Before you start, ask me the 5 most important questions you need answered to do this well. Don't begin until I've responded.

Power variation:

Before writing anything, identify your assumptions about what I want. List them. Ask me to confirm or correct each one.
2
Few-Shot Examples — Show, Don't Just Tell
The most underused technique. Works better than any prompt description.
Impact

Instead of describing the format you want, show Claude 2–3 examples of the exact output. Claude's pattern matching is exceptional — it will nail your format, tone, and structure on the first try.

❌ Without It
"Write a tweet in my style." → Sounds nothing like you.
✅ With It
Paste 3 of your best tweets, say "Write 5 more like these." → Sounds exactly like you.

Template:

Here are examples of the output I want: Example 1: [PASTE YOUR EXAMPLE] Example 2: [PASTE YOUR EXAMPLE] Example 3: [PASTE YOUR EXAMPLE] Now create [NUMBER] more in exactly this style for [YOUR TOPIC/TASK].
3
XML Tags for Complex Prompts
Anthropic's own recommended technique — dramatically improves accuracy on long prompts
Impact

When your prompt has multiple parts, wrap each section in XML tags. Claude was trained on structured data and natively understands XML hierarchy — it attends to each section more accurately.

❌ Without Tags
One long paragraph mixing context, task, and format. Claude loses track of what's what.
✅ With Tags
Each section clearly delineated. Claude attends to context separately from task. Much cleaner output.

Template:

<context> [Background information Claude needs to know] </context> <task> [Exactly what you want Claude to do] </task> <examples> [Any examples of good output] </examples> <format> [How you want the response structured] </format>
4
The Self-Critique Loop
Claude catching its own mistakes — before you have to
Impact

After Claude gives you an answer, ask it to critique its own response. It finds gaps and missing angles it didn't catch the first time. The second pass is almost always noticeably better.

Basic (one follow-up):

Now critique your own response. What did you miss, get wrong, or oversimplify? What would a skeptic push back on?

Full loop (build into the original prompt):

Answer my question, then immediately critique your answer — identify weaknesses, gaps, and anything you're uncertain about. Then give me a revised, improved version based on that critique.

For important decisions:

Play devil's advocate against your own recommendation. What's the strongest case that you're wrong?
5
Negative Prompting — Tell Claude What NOT to Do
Often more powerful than describing what you want
Impact

Claude has strong default behaviors — bullet points, hedging, summaries, corporate-speak. Explicitly prohibiting these breaks habits faster than trying to describe the style you want.

❌ Vague positive
"Write in a casual, direct style." → Claude tries but still defaults to some habits.
✅ Clear negative
"No bullet points. No hedging. No 'In conclusion.'" → Claude breaks its defaults completely.

Universal don'ts (add to any prompt):

Do not use bullet points or numbered lists. Do not hedge with phrases like "it's important to note" or "it's worth mentioning." Do not add a summary or conclusion paragraph. Do not start sentences with "Certainly" or "Absolutely." Do not use corporate jargon.

For writing specifically:

Do not write how an AI would write. Avoid: starting sentences with "I", using em-dashes excessively, perfect parallel structure, the word "delve", sounding like a LinkedIn post.
💡 Build your personal "don't" list over time. Every time Claude does something you hate, add it to a running list in your Global Instructions.