bomonike

anthropic-certs.png Deep Dive tips and tricks to get certified: Step-by-step tutorials, videos, practice exams.

US (English)   Norsk (Norwegian)   Español (Spanish)   Français (French)   Deutsch (German)   Italiano   Português   Estonian   اَلْعَرَبِيَّةُ (Egypt Arabic)   Napali   中文 (简体) Chinese (Simplified)   日本語 Japanese   한국어 Korean

Overview

NOTE: Content here are my personal opinions, and not intended to represent any employer (past or present). “PROTIP:” here highlight information I haven’t seen elsewhere on the internet because it is hard-won, little-know but significant facts based on my personal research and experience.

This article was completely hand-crafted (for now).

Anthropic the Company

  1. Visit https://anthropic.com/
    – the corporate marketing landing page.

    Notice “Anthropic is a public benefit corporation dedicated to securing its benefits and mitigating its risks.”

  2. Anthropic’s entry on LinkedIn classifies the company in the “Research Services” industry:

    “Anthropic is an AI safety and research company working to build reliable, interpretable, and steerable AI systems.”

    3M followers. 501-1K employees.

  3. Anthropic was founded in 2021 by seven former employees of OpenAI, including now CEO Dario Amodei was OpenAI’s Vice President of Research.

  4. Click “Read more” at https://www.anthropic.com/research about results from Anthropic’s survey of users.

    1. Click “Posts” tab to view announcements.
    2. Click “Ads” to see videos of Superbowl commercials.

  5. Anthropic’s YouTube channel

Competition

Claude competes with agentic coding tools (aka coding agent IDEs) that read a codebase, edit files, and run commands:

Claude Product Uptime

anthropic-systems-hist.pngREMEMBER: Anthropic doesn’t offer phone or live chat support, only thru chat at support.claude.com. Uptime shows Anthropic’s own production environments:

REMEMBER: Anththropic does not host their own models but use AWS, Azure, GCP, etc. Claude is the only frontier AI model available on all three leading cloud providers: AWS, Google Cloud, and Microsoft. Claude would also be integrated into the Databricks Data Intelligence Platform and Snowflake’s Lakehouse databases.

PROTIP: That enables us to bring costs down by using a downloaded local foundation model while using Claude Code/Work.

References:

Features Glossary

Automation provided by AI agents have gone beyond auto-complete of code.

Productivity: What can you do with Claude?

PROTIP: Improvements in net productivity can be confidently monitized when features are combined to be useful when consistently applied:

CAUTION: Cowork activity is not captured in audit logs or Compliance APIs today, which is why it is not for regulated workloads.

Pricing Subscriptiions

PROTIP: Use merlin.ai’s bulk purchasing costs $5/mo ($60/year) (with code AZ5) to access several LLMs (Claude Sonnet 4.5, OpenAI GPT5, etc.) instead of paying for a Claude AI subscription at https://claude.com/pricing:

Tutorials

Anthropic’s own tutorials are at:

Articles:

YouTube videos with no subscription:

YouTube videos peddling subscriptions:

by Brock Mesarich - AI for Non Techies to pitch $47/mo AI for Non-Technies: “Dispatch” from your phone.

Others when you’re through with the above:

https://www.youtube.com/watch?v=uUGfo8QOsW0&pp=ugUEEgJlbg%3D%3D Claude Mythos 5: Most Powerful Model Ever! AGI, GLM 5.1, Claude Code Update & Codex Plugins! AI NEWS

Use My Claude Code Template

PROTIP: Load my templates repo from GitHub, which contains a curated set from other tutorials.

  1. In your OS Terminal app, create a “bomonike” folder. 
    mkdir -p ~/bomonike
  2. In your OS Terminal app, clone just the main branch: 
    git clone https://github.com/bomonike/claude-templates.git --depth 1
cd claude-templates

    PROTIP: Use this as your base project when you install Claude.

    aliases.sh

    alias cl='claude --dangerously-skip-permissions'
alias clc='cl --continue'   # resume last session with the context/history from the previoius session
# Resume Claude with the context/history from the previoius session but still be able to get back to that point later:
alias clf='claude --resume --fork session'.

Visual Studio Code Install

  1. Install Homebrew (which is based on Ruby).
  2. Install VSCode and start it: 
    brew install --cask visual-studio-code
code
  3. Click the Extensions and enter “Claude Code” in the Marketplace claude-vscode-install.png
  4. Click “Install” to the one from “Anthropic” (marked with a blue star).

Claude Desktop app Install

VIDEO: Fun fact: 90% of code in Claude Code is written by itself, in TypeScript, React, Ink, Yoga, and Bun. The team works at around 5 releases per engineer each day. AI agents are used for code reviews and tests, test-driven development’s (TDD) renaissance, automating incident response, and cautious use of feature flags.

  1. Install pre-requisite utilties NodeJs: 
    brew install node
winget install OpenJS.NodeJS.LTS   # on Windows

    node --version

    v20.18.0
  2. Click “Download desktop app” (claude.dmg to install on macOS) https://claude.ai or https://claude.com/download,
    open your Terminal app and run: 
    curl -fsSL https://claude.ai/install.sh | bash

    PROTIP: We do not recommend “brew install” because it can be out of date, even though it’s more convenient since Homebrew installs to /opt/homebrew/bin for all apps.

  3. Edit your ~/.bashrc and .zshrc file to ensure that the program will be first in the OS $PATH folder by adding this at the bottom of the file: 
    echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc && source ~/.bashrc
  4. Confirm installation success: 
    claude --version

    This should reflect the latest release at https://github.com/anthropics/claude-code/release which was, at time of this writing:

    2.1.86 (Claude Code)

    PROTIP: Notice that Claude is updated daily. So end your day with a backup and start your day with an update.

  5. Confirm installation location: 
    whereis claude

    Claude was not installed if you see: bash: claude:: command not found Otherwise you should see this (where ~ is replaced with /Users/your machine username):

    claude: ~/.local/bin/claude
  6. Open the claude app: 
    $( whereis claude)

    That’s the equivalent of:

    ~/.local/bin/claude

    Alternately, more simply since the path is within $PATH:

    claude

    Alternately: To begin Claude with the context/history from the previoius session:

    claude --resume

    Alternately, to begin Claude with the context/history from the previoius session but still be able to get back to that point later:

    claude --resume --fork session

    Remember that aliases were setup.

    crf

    /login First-time Authentication

  7. The first time that Claude runs:

    claude-code-start.png

    ???

  8. Continue to browser.
  9. Claude Code would like to connect to your Claude chat account
  10. Click “Authorize”.
  11. Press command+W to close the browser window.
  12. Click “Copy Code”. Press command+tab to switch to VSCode.
  13. Click on the entry until the orange border appears.

  14. Press command+V to paste. Click “Authorize”.

    PROTIP: Press shift+command and - or + to make fonts larger or smaller. But that adjusts for all panes. So many prefer to view Claude Code standalone rather than within VSCode.

    PROTIP: Ideally, use three monitor screens: Terminal for Claude Code, Visual Studio (vertical view), Tutorial screen.

  15. Check Authentication status: 
       claude auth status
  16. To disable Authentication: 
    claude auth logout

    REMEMBER: Logout auth before setting up auth for 3rd-party clouds (Amazon, GCP, Microsoft, etc.)

    From Google VertexAI after installing gcloud cli:

    export ANTHROPIC_???_API_KEY="..."
export CLAUDE_CODE_USE_???=1

    From Micrsoft Foundry Project API Key:

    export ANTHROPIC_FOUNDRY_API_KEY="..."
export CLAUDE_CODE_USE_FOUNDRY=1
  17. VIDEO: Run Claude with no token fees (with some privacy invasion). 
    export ANTHROPIC_API_KEY=""
export ANTHROPIC_BASE_URL=http://localhost:11434

    VIDEO: Instead of “Download” at
    https://lmstudio.ai/blog/claudecode

    brew install --cask lm-studio

    Alternately, use Ollama VIDEO:

    brew install ollama
export ANTHROPIC_AUTH_TOKEN=ollama
ollama signin
ollama pull kimi-k2.5:cloud  # on Claude's cloud (AWS)
OLLAMA_CONTEXT_LENGTH=64000 ollama serve
claude --model "kimi-k2.5:cloud"

    WARNING: Kimi (in China) was created (stolen) by distillation of Anthropic’s model.

    Setup auth for free use of moonshot.ai’s Kimi model downloaded for running on Ollama via local relay path. The model features a 1T-parameter Mixture-of-Experts (MoE) Transformer architecture with 32B activated parameters. It supports image, video, PDF, and text inputs up to 256K tokens and excels in benchmarks like MMMU-Pro (78.5), SWE-Bench Verified (76.8), and AIME 2025 (96.1). Trained on approximately 15 trillion mixed visual and text tokens, it enables native multimodality, cross-modal reasoning, and efficient tool use grounded in visual data.

    Using a free model means that you can use automatic /loop to iterate through many results, then select the best, like a Monte Carlo simulation.

    But LM Studio using the MLX backend can produce 20 to 30 percent faster generation for the same model on the same hardware. And the Apple M3 Max has more bandwidth than the newer M4 Pro.

    /help for Shortcuts

    claude-code-help.png

    REMEMBER: Just as within Jupyter Notebook, run shell commands prefixed with the ! modifier. For example, ! pwd will run the pwd command and insert the output right into the conversation.

    Settings menu and keyboard shortcuts

  18. claude-settings-menu2.pngClick the Toggle sidebar (squarish) icon to collapse and expand the sidebar menu.

    REMEMBER: The “Usage” Settings menu item does not appear until you have a paid subscription.

  19. PROTIP: From anywhere in Claude, press shift+command+, (comma) for Claude’s Settings at https://claude.ai/settings/general

    But switch off the “AWS Extend Switch Roles” browser extension if that comes up instead.

  20. PROTIP: To chat from any screen, switch to a New Chat prompt by pressing shift+command+O (the letter) and start typing. For the pop-up, press command+K or shift+command+I for incognito (for the prompt to not appear among Recents).

    REMEMBER: When your cursor is within the chat box, use these keyboard shortcuts:

    claude-chat-keys.png

    Projects

    References:

    • “Upload materials, set custom instructions, and organize conversations in one space.”
    • VIDEO

    REMEMBER: Unless you go incognito, every time you run Claude in a directory, a Claude Code Project is created under ~/.claude/projects. So review and remove.

  21. Click the “Project” on the left menu to provide a way for Claude to remember your preferences and customize its responses to your preferences. So you don’t to repeat yourself.

    PROTIP: If you work with different companies or clients, isolate each by creating a different project containing different information.

  22. Click “+ New Project”

    TODO: ???

    Team/Enterprise subscribers can share a Project among themselves.

    Permissions

    shift+tab cycles through the permissions modes, so auto-accept edits is displayed just because currently I’m in the bypass permissions mode. There is one more permission mode plan, in which Claude Code will discuss and plan, but will not make changes to your files.

    Plan mode workflow

    REMEMBER: The revolution to productivity from AI comes from “Plan Mode”, which uses AI to generate a plan rather than “vibe coding” prompts thtat generate results directly. Generating code from plans is more repeatable and enables several people to review and collaborate.

  23. Cycle to plan mode by pressing shift+tab twice (switching): 
    ⏸ plan mode on (shift+tab to cycle)
    1. Write your goals. Build your ability to define objectives clearly.
    2. Let Claude break it into steps.
    3. Review and iterate the plan. Ask Claude improve the plan (in a loop).

    4. Ask Claude to add tests to evaluate whether its solution is complete and valid.

    5. Switch to auto-accept edits mode.
    6. Setup a container during dev.
    7. Have Claude execute the plan end-to-end.
    8. Review output - Refine if needed.

    how-i-structure-claude-code-projects-skills-mcp-v0-ubchqhdo8ujg1.webp

    Prompts

  24. Type your question or command on top of “How can I help you today?”

    REMEMBER, there is a cutoff for when information has been loaded in the model.

    Artifacts

    claude-app-menu.png

  25. Click “+ New artifacts”. Artifiacts are pre-coded small interactive apps such as Productivity Tools.
  26. Click “Artifacts” on the menu and under its “Inspiration” tab, try:
    • click “Flashcards” and provide a CSV file.
    • Click “QR code generator”.
    • Click “Trivia” game.
    • Click “Better than very” to find more expressive words.
    • Click “CSV Data Visualizer”.

  27. To create your own automations, consider the “Cowork” button at the top of the Claude app.

    Cowork and Projects both require a Pro Plan subscription.

    Connectors

  28. Click one Category at a time to see what’s available already: Code, Communication, Data, Design, Development, Financial Services, Health, Life sciences, Productivity, Sales and Marketing.

    REMEMBER: Most services at the end of the connector (such as Zapier) charge money.

  1. PROTIP: Instead of clicking “Download” for “Desktop” within “Claude Code environments”, 
    Setting up Claude Code...
✔ Claude Code successfully installed!        
Version: 2.1.81
Location: ~/.local/bin/claude
Next: Run claude --help to get started
⚠ Setup notes:
• Native installation exists but ~/.local/bin is not in your PATH. Run:
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc && source ~/.bashrc
✅ Installation complete!

    WARNING: Installing using curl would require adding to the $PATH in your ~/.zshrc or ~/.bashrc file this line:

    export PATH="$HOME/.claude/bin:$PATH"

    RECOMMENDED: In a Terminal, install Claude Code</a>

    brew info claude-code
brew install claude-code

    Terminal-based AI coding assistant
install: 170,173 (30 days), 390,990 (90 days), 585,358 (365 days)
==> Moving App 'Claude.app' to '/Users/johndoe/Applications/Claude.app'
  2. Confirm app folder location: 
    tree ~/.claude

    folders:

    backups
cache
downloads

  3. REMEMBER: The free Claude.ai plan does not include Claude Code access. Upgrade to a Claude Pro, Max, Teams, Enterprise, or Console account.\

    Select the subscription level:

    Claude Code can be used with your Claude subscription or billed based on API usage through your Console account.
                                                            
Select login method:            

❯ 1. Claude account with subscription · Pro, Max, Team, or Enterprise
                  
   2. Anthropic Console account · API usage billing
                                       
   3. 3rd-party platform · Amazon Bedrock, Microsoft Foundry, or Vertex AI

    Documentation:
    · Amazon Bedrock: https://code.claude.com/docs/en/amazon-bedrock · Microsoft Foundry: https://code.claude.com/docs/en/microsoft-foundry
    · Vertex AI: https://code.claude.com/docs/en/google-vertex-ai

    /terminal-setup slash commands

    ???

    slash commands

  4. Type just the / slash character for a menu:

    /batch
/claude-api
/compact    # summarize the conversation and replaces the current context with the summary. 
/context    # token usage by each system component
/cost       # tokens spent
/debug
/extra-usage
/heapdump
/loop
/pr-comments
/release-notes
/review
/security-review
/simplify
/update-config
/schedule

    Others:

    /help       # menu below:
/connect    # establish connection
/start      # Begin a new session

/memory     # 
/statusline # below the prompt defined in customizable ~/.claude/statusline.sh
/settings   # menu

/clear      # (aka /reset) is faster than exiting and starting Claude Code again.

/search     # through the database
/upload     # files
exit        # from Claude UI/CLI program

/status     # overview of your current Claude Code setup
/config     # configuration
/loop
/doctor

    Additional slash commands:

    /insights   # file://$HOME/.claude/usage-data/report.html
/effort   # Effort Level Controls https://www.youtube.com/watch?v=brLhhkUqcn4&t=18618s">max for Opus only. high, medium, low, auto.
/remote-control   # 
/batch   # Batch Tasks & PRs 
/simplify   # Code Review
/loop   # Schedule Prompts
/btw   # side question

    REMEMBER: Each session is a 5-hour rolling window (at time of this writing). ???

    Models reset ???

    /model default   # to switch to the sonnet model
/model haiku     # to switch to using the latest Haiku model.
/model Sonnet (1M context)  # to switch to using the latest Opus model.
/model Opus (1M context)    # to switch to using the latest Opus model.
/model mythos               # new Capybarra March 28, 2026 to Cyber Defenders.
/fast                       # to speed up Opus model execution.

    Hooks & Automation Rules

  5. Install utility a program ccusage to analyze session logs:
    https://github.com/ryoppippi/ccusage/

    See ccusage.com/guide/session/reports

### /status

Example:

   ❯ /status
   Version: 2.1.3
   Session name: /rename to add a name
   Session ID: 4eb36de6-c9f2-4c22-8ad3-a8232ea6c078
   cwd: /Users/gigi
   Auth token: none
   API key: /login managed key
   Organization: Perplexity AI
   Email: gigi.sayfan@perplexity.ai

   Model: opus (claude-opus-4-5-20251101)
   MCP servers: notion ✔, linear ✔, datadog ✔
   Memory: user (.claude/CLAUDE.md)
   Setting sources: User settings, Shared project settings, Project local settings

### /config

Configuration choices are stored by Claude in its file .claude/settings.json

   ❯ /config
      Auto-compact                              true
      Show tips                                 true
      Thinking mode                             true
      Prompt suggestions                        true
      Rewind code (checkpoints)                 true
      Verbose output                            false
      Terminal progress bar                     true
      Default permission mode                   Accept edits
      Respect .gitignore in file picker         true
      Auto-update channel                       latest
      Theme                                     Dark mode
      Notifications                             Auto
      Output style                              default
      Language                                  Default (English)
      Editor mode                               normal
      Model                                     opus

/doctor or CLI claude doctor

VIDEO

Token /context usage

Claude’s context window is 200K, meaning it can ingest 200K+ tokens (about 500 pages of text or more) when using a paid Claude plan. The Claude API can ingest 1M tokens when using Claude Opus 4.6 or Sonnet 4.6.

PROTIP: Take action when token usage is above 50%. See Rewind Mode (Escape x2)

claude-token-usage.webp

The first line in the example above:” 51k tokens (26%)” is what is currently used. Users on Claude Code with a Max, Team, or Enterprise plan, Claude Opus 4.6 have a 1M token context window.

REMEMBER: The Autocompact Buffer: 45k tokens (22.5%) is reserved for autocompaction. When your conversation approaches the context window limit, Claude summarizes earlier messages to make room for new content. Claude Code does this automatically when the context window fills up, but here’s the thing - automatic compaction might keep less important stuff and throw away useful insights. But that takes time and require work space. The context window limit applies to input + output combined. When autocompaction triggers, the model needs room to generate the summary. Without reserved space, a full context would leave no room for output. So right off the bat, you only have about half the context window for your actual conversation.

System Overhead: The system prompt and tools reserve almost 20k tokens (~10%).

The more MCP servers are used, the more “MCP Tools” tokens are used. Each tool within an MCP server consumes token before it even starts. Each of several tools are usually a part of each MCP server. For example, Notion has a tool for * create-pages * create-comment * update-page * update-database

/cost tokens spent

   ❯ /cost
  ⎿  Total cost:            $2.69
      Total duration (API):  5m 12s
      Total duration (wall): 9h 39m 12s
      Total code changes:    10 lines added, 1 line removed
      Usage by model:
             claude-haiku:  42.1k input, 790 output, 0 cache read, 11.9k cache write ($0.0609)
          claude-opus-4-5:  3.4k input, 10.7k output, 1.7m cache read, 235.3k cache write, 1 web search ($2.63).  x          ```


   <a name="loop"></a>

   ## /loop

   The <tt>/loop</tt> command parses natural language specifications into three parameters of a <strong>CronCreate call </strong>, which is not just a repetitiive <a target="_blank" href="https://ghuntley.com/loop/">"Ralph loop"</a>. It can also <strong>schedule a task</strong> that fires based on a timer, in the current Claude Code session. Close the terminal, exit Claude, or lose your connection, and all scheduled tasks vanish. 

   <pre>
   {
   "cron": "*/10 * * * *",
   "prompt": "Check the CI status on PR #42 and summarize any failures",
   "recurring": true
   }
   </pre>




   <a name="CLAUDE.md"></a>
   
   ### CLAUDE.md file

   REMEMBER: Claude Code has no memory. On every new single session, it wakes up with <strong>zero context</strong> about your project.
   So history and preferences must be added added as context.

1. At the <strong>Claude CLI</strong>, 
1. Copy in files from ???

   * CLAUDE.md referenced by
   * state.md — current state of the project
   * architecture.md — how everything fits together
   * terraform-CLAUDE.md
   * python-CLAUDE.md
   * MEMORY.md 
   <br /><br />

1. Integrate from those who shared theirs:
   * https://github.com/anthropics/courses/blob/master/tool_use/README.md
   * https://github.com/citypaul/.dotfiles/blob/main/claude/.claude/CLAUDE.md
   * https://github.com/jarrodwatts/claude-code-config

   * https://github.com/centminmod/my-claude-code-setup?tab=readme-ov-file#alternate-read-me-guides
   * Git Worktrees (for <a target="_blank" href="https://code.claude.com/docs/en/desktop#work-in-parallel-with-sessions">Parallel Sessions in Claude Code</a> via Claude Desktop apps
   * https://github.com/Piebald-AI/claude-code-system-prompts?tab=readme-ov-file
   * etc. ???
   <br /><br />

   * https://github.com/Piebald-AI/claude-code-system-prompts?tab=readme-ov-file#system-reminders

1. Customize System prompts using https://github.com/Piebald-AI/tweakcc


   <a name="init"></a>

1. generate a starter CLAUDE.md as a starting point:
   ```bash
   /init
  1. List folders and files: 
    tree

    TODO:

     ├── api
 ├── web
 ├── .editorconfig
 ├── .env.example
 ├── .gitignore
 ├── CLAUDE.md
 ├── README.md
 └── docker-compose.yml

  2. Edit file CLAUDE.md, the long-term memory file.

    The file guides Claude Code (claude.ai/code) when working with code in this repository.

    REMEMBER: At the start of each agent session, Claude looks for a CLAUDE.MD file in each GitHub repository root, in parent directories for monorepo setups, or in your home folder for universal application across all projects. So the file must be named with uppercase “CLAUDE”, lowercase “.md” (like GitHub looks for “README.md”). Providing this context up front helps agents avoid running incorrect commands or introducing architectural or stylistic inconsistencies when implementing new features.

    Each CLAUDE.md file holds markdown-formatted project-specific context that should be repeated in every prompt: Project context (basic rules), About this project, Key directories, Standards, structure, conventions, workflows, style, domain-specific terminology. Example:

  3. PROTIP: Keep CLAUDE.md files to a maximum of 100–200 lines. Long files are a code smell and take up precious context. CLAUDE.md should be a routing file, not a knowledge dump.

    Point to .claude/rules/*.md for detailed specs and docs/ for architecture. Otherwise it gets so long that Claude skims it and misses the important stuff.

    Delete what you don’t need — deleting is easier than creating from scratch.

    • “Creating the Perfect CLAUDE.md for Claude Code” by Ivan Kahl January 15, 2026
    • https://medium.com/@CodeCoup/i-wasted-8-minutes-per-change-in-claudes-code-heres-what-fixed-it-4baeeef1c07f
    • https://github.com/ArthurClune/claude-md-examples which is based on:
    • https://github.com/modelcontextprotocol/python-sdk/blob/main/CLAUDE.md
    • https://github.com/p33m5t3r/vibecoding/blob/main/conway/CLAUDE.md
    • https://github.com/saaspegasus/pegasus-docs/blob/main/CLAUDE.md
    • https://github.com/centminmod/my-claude-code-setup

  4. Explore Claude Plugin Marketplace of Curated plugins, agent skills, and MCP servers for Claude Code: https://claudemarketplaces.com/learn

    https://claudemarketplaces.com

  5. Install HeadsUpDisplay (HUD) plugin to add up to 4 lines below your input prompt to know if it’s still making progress or is stuck. 
    /plugin marketplace add <a target="_blank" href="https://github.com/jarrodwatts/claude-hu">jarrodwatts/claude-hud</a>
/plugin install claude-hud
/reload-plugins   # to activate
/claude-hud:setup     # to ~/.claude/settings.json 
/restart Claude Code
code ~/.claude/plugins/claude-hud/config.json

    Updates every ~300ms. $80/yr Masterclass

  6. Consider https://github.com/BayramAnnakov/claude-reflect on Claude Plugin Marketplace - a self-learning system (/reflect-skills) for Claude Code that captures corrections, positive feedback, and preferences — then syncs them to CLAUDE.md and AGENTS.md.

Create an iPhone app

VIDEO: Chris Raroque runs Claude Code Opus inside a Warp client referencing a [paid] mobbin.com design template. Voice dictates changes. Breaks down generation section by section. No hand edits.

Anthropic provides free tutorials at https://anthropic.skilljar.com/

Claude Partner Network

https://claude.com/partners

“Anthropic invests $100 million into the Claude Partner Network” (announced Mar 12, 2026) mentions “technical” Claude Certified Architect (CCA) Foundations certification.

#CAExamPrep

“A significant proportion of our $100 million investment will go directly to our partners as direct support for training and sales enablement, and for market development (including work to make customer deployments successful) and co-marketing for joint campaigns and events. “

The Partner Portal at https://partnerportal.anthropic.com/s/login/ provides Academy training materials, sales playbooks used by our own go-to-market team, and other co-marketing documentation.

At the Services Partner Directory, enterprise buyers can find firms with Claude implementation experience.

Partners get priority access to new certifications as they roll out.

Additional certifications for sellers, architects, and developers.

Certifications

  1. Use your personal email to sign up for their newsletter.

  2. Use your personal email to sign In to https://anthropic.skilljar.com

Claude Certified Architect (CCA), Foundations

Exam Domains from Anthropic’s Exam Guide.pdf:

  1. 27% Agentic Architecture & Orchestration - how agents loop, coordinate with subagents, and enforce rules with hooks vs prompts. STARTER
    • The Agentic Loop
    • Hub-and-Spoke Architecture
    • Prompts vs. Hooks
    • Anti-patterns: natural language parsing to determine loop termination; arbitrary iteration caps as the primary stopping mechanism; checking for assistant text as a completion indicator.
  2. 18% Tool Design & MCP Integration - how Claude connects to external systems and how tool descriptions determine routing.
    • Tool Descriptions
    • MCP Scoping
    • Tool overload
  3. 20% Claude Code Configuration & Workflows - skills, commands, plan mode, and CI/CD.
    • Configuration Hierarchy
    • When to Use What
    • CI/CD Integration
  4. 20% Prompt Engineering & Structured Output - structured output with JSON schemas, and validation loops.
    • Few-Shot Advantage
    • Guaranteed structured output with JSON schemas
    • Validation Loop
  5. 15% Context Management & Reliability



The community confirms is the exam’s focus areas: fallback loop design, Batch API cost optimization, JSON schema structuring to prevent hallucinations, and MCP tool orchestration.

IBM AI Engineering (Coursera) ML/DL concepts and model deployment Conceptual + hands-on Cloud-agnostic

Anthropic Academy is at https://www.anthropic.com/learn

https://anthropic.skilljar.com/claude-certified-architect-foundations-access-request

References:

Models

Claude Model Family:

  Claude Opus Claude Sonnet Claude Haiku Mythos
Description Highest level of intelligence Balance of quality, speed, cost Most cost-efficient and latency-optimized model  
capabilities
(Best used for)		 advanced reasoning Common coding tasks Quick code completions and suggestions  
Cost: Highest Medium Lowest  
Input/Output $/MTok $5/$25 $3/$15 $1/$5  
Prompt caching Read/Write $/MTok $0.50/$6.25 $0.30/$3.75 $0.10/$1.25  
max_input_tokens (Context window) 1M tokens 1M tokens 200k tokens  
max_tokens (Max output) 128k tokens 64k tokens 64k tokens  
Tokens/min Input & Output 30K/8K 30K/8K 50K/10K  
Comparative Latency: Moderate Fast Fastest  
Supports Reasoning
& Adaptive Thinking		 Yes Yes No!  

REMEMBER: Each model used has a different ID and version on each cloud: See DOCS: API codes for each Claude Model version list or GET https://api.anthropic.com/v1/models

On AWS, the full model_id = “us.anthropic.claude-3-7-sonnet-20250219-v1:0”

Feature Claude Opus 4.6 Claude Sonnet 4.6 Claude Haiku 4.5
Claude API ID claude-opus-4-6 claude-sonnet-4-6 claude-haiku-4-5-20251001
Claude API alias used by API calls claude-opus-4-6 claude-sonnet-4-6 claude-haiku-4-5
GCP Vertex AI ID claude-opus-4-6 claude-sonnet-4-6 claude-haiku-4-5@20251001
AWS Bedrock ID anthropic.claude-opus-4-6-v1 anthropic.claude-sonnet-4-6 anthropic.claude-haiku-4-5-20251001-v1:0
Reliable knowledge cutoff: - - February 2025
Training data cutoff: - - July 2025

TODO: Microsoft Foundry?

REMEMBER: The Reliable knowledge cutoff is the date through which knowledge is most extensive and reliable.

Training Data Cutoff is the broader range of data used.

Advanced reasoning:

Common coding tasks:

Quick code completions and suggestions:

Chat API call using Claude Opus

  1. Get your API key from the Claude Console API keys page.
  2. Save the value in your Password Manager.
  3. In a Terminal app, set environment variable: 
    export ANTHROPIC_API_KEY='sk...your-api-key-here'
  4. For API usage, buy $5 of credits from https://platform.claude.com/settings/billing

    https://platform.claude.com/settings/keys

  5. Run the simple-msg.py from https://github.com/bomonike/claude-templates… ‘'’python import anthropic client = anthropic.Anthropic() message = client.messages.create( model=”claude-sonnet-4-6”, max_tokens=1024, messages=[{ “role”: “user”, “content”: “Hello, Claude” }] ) print(message.content[0].text) ```

  6. Run the curl-model-info.sh from https://github.com/bomonike/claude-templates…

    curl https://api.anthropic.com/v1/messages \
-H "Content-Type: application/json" \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-d '{
   "model": "claude-opus-4-6",
   "max_tokens": 1000,
   "messages": [
      {
      "role": "user",
      "content": "What are the capabilities of Claude Opus 4.5 and its Reliable knowledge cutoff date and Training data cutoff dates?"
      }
   ]
}'

    An example of the response: ???

    {
"id": "msg_01HCDu5LRGeP2o7s2xGmxyx8",
"type": "message",
"role": "assistant",
"content": [
   {
      "type": "text",
      "text": "Here are some effective search strategies to find the latest renewable energy developments:\n\n## Search Terms to Use:\n- \"renewable energy news 2024\"\n- \"clean energy breakthrough\"\n- \"solar/wind/battery technology advances\"\n- \"green energy innovations\"\n- \"climate tech developments\"\n- \"energy storage solutions\"\n\n## Best Sources to Check:\n\n**News & Industry Sites:**\n- Renewable Energy World\n- GreenTech Media (now Wood Mackenzie)\n- Energy Storage News\n- CleanTechnica\n- PV Magazine (for solar)\n- WindPower Engineering & Development..."
   }
],
"model": "claude-opus-4-6",
"stop_reason": "end_turn",
"usage": {
   "input_tokens": 21,
   "output_tokens": 305
}
}
  7. Review Claude token usage at https://platform.claude.com/usage

Multi-Provider

The “AI-6 framework” at February 2026 Packt BOOK: “Design Multi-Agent AI Systems Using MCP and A2A” (on OReilly.com) by Gigi Sayfan referencing his book GitHub repo https://github.com/Sayfan-AI/ai-six.

AI Fluency Class

https://www.anthropic.com/learn/claude-for-you

AI Fluency 11-video playlist on YouTube

01Introduction to AI Fluency

02The AI Fluency Framework

03Deep Dive 1: What is Generative AI?

04Delegation

05Applying Delegation

06Description

07Deep Dive 2: Effective Prompting Techniques

08Discernment

09The Description-Discernment Loop

010Diligence

Text Chat using Claude API

https://platform.claude.com/docs/en/get-started

curl https://api.anthropic.com/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-opus-4-6",
    "max_tokens": 1000,
    "messages": [
      {
        "role": "user",
        "content": "What should I search for to find the latest developments in renewable energy?"
      }
    ]
  }'

Making a request

Multi-Turn conversations work by you maintaining your own chat history.

Chatbot

PROTIP: By default, Chat returns message with code between backticks so its explanation text can be added. To retrieve just the code returned with “stop sequences”:

import json

# Parse as JSON to validate and format
parsed_data = json.loads(text.strip())

# Or just strip whitespace for other data types
clean_text = text.strip()

messages = []
add_user_message(messages, "Generate a very short event bridge rule as json")
add_assistant_message(messages, "```json")
text = chat(messages, stop_sequences=["```"])

System prompts

Temperature

Streaming

Controlling model output

Structured data

MCP

PROTIP: This is a common misconception. MCP Servers and tool use are complementary but different concepts. Tool use is about Claude calling functions to accomplish tasks. MCP is about who provides those functions - instead of you writing them, someone else has already implemented them in an MCP Server.

The key insight is that MCP Servers provide tool schemas and functions already defined for you, while direct tool use requires you to author everything yourself. Both involve Claude using tools, but MCP dramatically reduces the development work required on your end.

Run in Containers

Instead of sitting around monitoring every prompt like a hall monitor just in case a rogue rm -rf slips by.

So consider a Code Container to mount every project into an isolated container where I can let my harness run loose with full permissions while the actual machine stays untouched.

  1. Install container in NodeJs: 
    npm install -g code-container

    However, although actions within a container can’t affect your real system, it breaks when it needs network access, host filesystem access, or anything that crosses the sandbox boundary. Most real workflows need at least one of those.

    So auto mode lets an AI classifier decide what’s safe, block what isn’t, and ask you only when it’s genuinely unsure. Auto mode runs two separate security systems. One watches what goes into the agent’s context.A server-side detector scans content for prompt injection attempts.

    The second line of defense evaluates what the agent wants to do before it does it. Before the agent executes any action with real consequences, the “transcript classifier” built on Claude Sonnet 4.6 classifier evaluates the action against a set of decision criteria using full chain-of-thought reasoning.

    This BLOG by Marco Kotrotsos reports a 17% false negative which allowed dangerous actions, including 5.7% data exfiltration attack success rate. But that’s still better that letting everything through when using the time-saving:

    –dangerously-skip-permissions

    Auto mode is not a replacement for judgment on high-stakes operations.

  2. To see the complete default configuration for customization: 
    claude auto-mode defaults

    References:

    • https://www.youtube.com/watch?v=OBQtXEUe3Ik&pp=0gcJCdkKAYcqIYzv

300ms startup time!

References:

https://www.youtube.com/watch?v=IjiaCOt7bP8&pp=ugUHEgVlbi1VUw%3D%3D Agent Skills: Code Beats Markdown (Here’s Why) Sam Witteveen

26-03-28 v019 doc: HUD :anthropic-certs.md created 2026-03-19