SECTION 1 - The Foundations (Concepts 1-8)
SECTION 2 - Making Claude Personal (Concepts 9-14)
SECTION 3 - Power Features (Concepts 15-23)
SECTION 4 - Using Claude Autonomously (Concepts 24-27)

00:00 - Every Claude Code Concept Explained
01:00 - What is Claude Code?
01:40 - The Terminal
02:30 - Prompts
03:09 - Permissions
05:29 - Tools (Read / Write / Bash)
06:22 - Context Window
07:20 - Conversation History / Resume
08:00 - Token Usage / Cost Tracking
08:52 - CLAUDE.md
09:25 - Memory Usage
10:30 - Compact Context
11:26 - Models (Sonnet, Opus, Haiku)
12:05 - Denying Access to Files
12:56 - Flags
14:13 - Extended Thinking
15:03 - Slash Commands
15:50 - Skills
16:46 - Hooks
17:23 - MCP Servers
18:00 - Sub-Agents
19:30 - Agent Teams
20:48 - Image / Screenshot Support
21:38 - Checkpoints / Undo
22:19 - Git Integration
23:09 - Headless Mode
24:23 - Claude Max vs API (How You Pay)
25:02 - Worktrees
26:00 - What next?

Create a detailed outline of this video, broken down by sections and subjects that are given in the video description.

Here is a detailed outline of the video, organized by the four sections listed in the description and the named chapters within each section.[youtube]​

Section 1 – The Foundations (Concepts 1–8)

1. Every Claude Code Concept Explained (Intro) [00:00][youtube]​

   • Problem setup: non‑technical users are overwhelmed by terms like context window, sub‑agents, MCP servers.[youtube]​

   • Goal: build a clear mental model of Claude Code via 27 short concepts, from basics to advanced.[youtube]​

2. Concept 1 – What is Claude Code? [01:00][youtube]​

   • Contrast with chatbots (Claude, ChatGPT, Gemini) that “only talk” vs Claude Code which can act on your computer.[youtube]​

   • Capabilities: create files, build websites, set up databases, install packages, automate repetitive tasks via plain‑English instructions.[youtube]​

3. Concept 2 – The Terminal [01:40][youtube]​

   • Claude Code runs in the terminal (not the browser) as a text interface to your computer.[youtube]​

   • Minimal commands you need: open Claude (claude), close (Ctrl+C twice), clear memory (/clear later), while Claude handles most other terminal details.[youtube]​

4. Concept 3 – Prompts [02:30][youtube]​

   • Prompts are natural‑language instructions like “build me a landing page” or “fix this error.”[youtube]​

   • Emphasis on specificity: describe structure, styling, and requirements (e.g., single page, green color scheme, contact form, three service cards) for better results.[youtube]​

5. Concept 4 – Permissions [03:09][youtube]​

   • Claude Code can create, edit, and delete files and run commands, so it requests approval by default for significant actions.[youtube]​

   • Strategy: start with frequent approvals to learn what it’s doing, then pre‑approve safe actions via settings.json or /permissions while gating risky ones.[youtube]​

   • Safe to pre‑approve: reading files/folders, running dev server, running tests, git operations.[youtube]​

   • Keep gated: installing packages, deleting files/folders, network/API calls or anything touching the internet.[youtube]​

   • Avoid using --dangerously-skip-permissions; instead, maintain an allow‑list in settings.json that can be shared with your team.[youtube]​

6. Concept 5 – Tools (Read / Write / Bash) [05:29][youtube]​

   • Core tools: read (inspect files/data), write (create or modify files), bash (run shell commands like ls, installing packages, running tests, moving files).[youtube]​

   • You describe the goal; Claude chooses appropriate tools automatically (e.g., using bash ls to list files).[youtube]​

7. Concept 6 – Context Window [06:22][youtube]​

   • Context window = Claude’s short‑term memory: all messages, files read, and responses in the current session.[youtube]​

   • When it fills up, “context rot” occurs: earlier information is lost and outputs degrade.[youtube]​

   • Mitigation: keep conversations focused, start new sessions for new tasks, use /clear or /compact to manage context; track usage via the green bar.[youtube]​

8. Concept 7 – Conversation History / Resume [07:20][youtube]​

   • Claude Code automatically saves session history.[youtube]​

   • Use claude-resume to reopen the most recent conversation and continue with the same context.[youtube]​

   • You can view a list of previous sessions, see what each was about, and resume long‑running projects over multiple days.[youtube]​

9. Concept 8 – Token Usage / Cost Tracking [08:00][youtube]​

   • Costs are based on tokens (≈ three‑quarters of a word); both inputs, outputs, and files read consume tokens.[youtube]​

   • Different models (e.g., Sonnet vs Opus) have different pricing; Claude Code tracks token usage so you can monitor spend.[youtube]​

   • Commands like /cost and /stats show tokens used in the current session.[youtube]​

Section 2 – Making Claude Personal (Concepts 9–14)

10. Concept 9 – CLAUDE.md [08:52][youtube]​

• A project‑local claude.md file stores your preferences, rules, and project structure.[youtube]​

• Claude reads this file first on every instance, letting you define language choices, style rules (e.g., no em dashes), brand voice, and other persistent instructions.[youtube]​

• Functions as an instruction manual you refine over time, and it applies across all conversations for that project.[youtube]​

11. Concept 10 – Memory Usage [09:25][youtube]​

• Distinguishes between project‑specific claude.md and Claude’s automatic, cross‑session memory.[youtube]​

• Claude learns recurring preferences (e.g., JavaScript usage, coding style, project conventions) and stores them in an internal memory store, not directly visible in the project directory.[youtube]​

• You can ask Claude to view, add, or remove remembered items to tune personalization.[youtube]​

12. Concept 11 – Compact Context [10:30][youtube]​

• Automatic compaction: when context usage hits roughly 85–95%, Claude summarizes key info and drops noise, carrying forward the summary.[youtube]​

• Manual compaction: use /compact with arguments (e.g., “keep info about API calls”) to control what remains.[youtube]​

• Goal: keep Claude “sharp” during long sessions by preserving essentials while freeing context.[youtube]​

13. Concept 12 – Models (Sonnet, Opus, Haiku) [11:26][youtube]​

• Haiku: fastest and cheapest, suited for simple or quick tasks.[youtube]​

• Sonnet: balanced all‑rounder for most work.[youtube]​

• Opus: most capable and expensive, ideal for complex reasoning‑heavy tasks.[youtube]​

• You can switch models mid‑conversation with /model; match model choice to task importance and complexity.[youtube]​

14. Concept 13 – Denying Access to Files [12:05][youtube]​

• Not all files should be visible to Claude: some are huge, sensitive (API keys, passwords), or irrelevant.[youtube]​

• Use a deny‑list in settings.json to mark files and folders (e.g., .env, secrets/, credentials) as off‑limits.[youtube]​

• Denied files are not discoverable, not included in search, and not readable even if explicitly requested.[youtube]​

15. Concept 14 – Flags [12:56][youtube]​

• Flags configure Claude Code at launch for a specific session.[youtube]​

• Examples: -model to pick the model, -allowed-tools to constrain tool access, -verbose for detailed output, --dangerously-skip-permissions as the unsafe “nuclear” option.[youtube]​

• These flags let you pre‑set behavior before the agent starts working.[youtube]​

Section 3 – Power Features (Concepts 15–23)

16. Concept 15 – Extended Thinking [14:13][youtube]​

• Some tasks require deep multi‑step reasoning rather than quick replies.[youtube]​

• Extended thinking gives Claude an internal “reasoning budget” of tokens to think before acting; it is enabled by default and no longer needs explicit “think hard” prompts.[youtube]​

17. Concept 16 – Slash Commands [15:03][youtube]​

• Slash commands are typed as /command and trigger predefined actions.[youtube]​

• Examples: /init to bootstrap a project and create config files, /compact to shrink context, /clear to reset a conversation.[youtube]​

• /help lists all available slash commands; you can turn your own repetitive tasks into commands by placing them in the commands folder.[youtube]​

18. Concept 17 – Skills [15:50][youtube]​

• Skills are specialized instruction bundles that teach Claude to excel at specific task types (e.g., copywriting, slide design, front‑end design).[youtube]​

• When your request matches a skill’s description, Claude loads that file (and supporting files) into context, improving quality at the cost of more context usage.[youtube]​

• You can install community skills or write your own expert playbooks.[youtube]​

19. Concept 18 – Hooks [16:46][youtube]​

• Hooks are non‑AI scripts that run automatically on certain events, like after a file save or when a command runs.[youtube]​

• Uses: auto‑formatting files on save, logging every command Claude runs, enforcing guardrails without spending AI tokens.[youtube]​

• Require some technical setup, which you can ask Claude to help with.[youtube]​

20. Concept 19 – MCP Servers [17:23][youtube]​

• MCP (Model Context Protocol) connects Claude Code to external tools and SaaS platforms (Airtable, Notion, Asana, etc.).[youtube]​

• Once connected, Claude can pull data from these services, push updates back, and orchestrate workflows across your whole stack from one conversation.[youtube]​

21. Concept 20 – Sub‑Agents [18:00][youtube]​

• Sub‑agents are specialist agents with their own context windows, system prompts, and tool access, still governed by the main claude.md.[youtube]​

• The main agent delegates tasks to them; they work independently and return results.[youtube]​

• Benefits: higher quality (clean, focused context) and speed (multiple sub‑agents can run in parallel).[youtube]​

• Rule of thumb: use a skill to improve performance inside the same conversation, use a sub‑agent when the task is self‑contained (e.g., “analyze this report and come back”).[youtube]​

• Claude can automatically create sub‑agents when it deems them useful.[youtube]​

22. Concept 21 – Agent Teams [19:30][youtube]​

• Limitation of sub‑agents: they only report to the main agent; they cannot see each other’s work directly.[youtube]​

• Agent teams allow multiple agents (e.g., researcher, writer, reviewer) to communicate with each other and share a task list, reducing the main agent bottleneck.[youtube]​

• Each teammate runs in its own Claude instance with a partial view of the conversation; useful for complex builds like SaaS apps needing coordinated work.[youtube]​

• Usage guideline: single agent for one feature, sub‑agents for specialized but separate tasks, agent teams for complex multi‑role projects.[youtube]​

23. Concept 22 – Image / Screenshot Support [20:48][youtube]​

• You can paste screenshots directly into Claude Code to show UI bugs or layouts instead of describing them in text.[youtube]​

• Claude can also interpret reference designs and generate approximate implementations (often 60–70% correct on first pass).[youtube]​

• Images consume context, but they can greatly speed diagnosis and design work.[youtube]​

24. Concept 23 – Checkpoints / Undo [21:38][youtube]​

• Claude Code automatically creates checkpoints (snapshots) before file edits.[youtube]​

• Use /re to open a list of prompts and associated checkpoints, then choose a restore point.[youtube]​

• You can restore only code, only conversation, or both, enabling safe rollback when the agent takes an undesirable path.[youtube]​

Section 4 – Using Claude Autonomously (Concepts 24–27)

25. Concept 24 – Git Integration [22:19][youtube]​

• Git/GitHub track every change Claude makes: what changed, when, and why.[youtube]​

• Connecting GitHub powers checkpointing, diff review, selective rollback, and safe team collaboration without overwrites.[youtube]​

• You don’t need deep Git expertise; Claude can manage many Git operations in the background.[youtube]​

26. Concept 25 – Headless Mode (CLI Mode) [23:09][youtube]​

• Formerly “headless,” now called CLI mode; lets Claude run without interactive back‑and‑forth.[youtube]​

• Add the -p flag to run a full agentic loop that executes the task end‑to‑end with no approval prompts, then returns results.[youtube]​

• Combine with flags like -allowed-tools to pre‑approve or restrict what Claude can do when running unattended.[youtube]​

• Mentions the “Ralph loop” plugin, which repeatedly re‑invokes Claude with updated files to iteratively refine until the task is complete.[youtube]​

27. Concept 26 – Claude Max vs API (How You Pay) [24:23][youtube]​

• Two main payment models: Claude Max subscriptions (e.g., Pro around $20/month, Max at higher tiers) vs pay‑as‑you‑go API billing per token.[youtube]​

• Subscriptions offer predictable costs and generous usage, suitable for heavier or mid‑build usage; API can be cheaper for light workloads.[youtube]​

• Recommendation: for serious building, prefer Pro/Max to avoid constantly worrying about token spend mid‑project.[youtube]​

28. Concept 27 – Worktrees [25:02][youtube]​

• Worktrees create multiple isolated working directories/branches within the same Git repo, each with its own files and Claude context.[youtube]​

• Use the -worktree or -w flag to start Claude in a named worktree; the value after -w becomes directory and branch name.[youtube]​

• Example usage:

  • claude -worktree feature-auth for a new feature.[youtube]​

  • claude -worktree bugfix-login for bug fixes.[youtube]​

  • claude -worktree refactor-api for refactors, all running in parallel without interference.[youtube]​

• After exiting, Claude handles cleanup depending on whether changes were made; sub‑agents can also be assigned their own worktrees for parallel isolated work.[youtube]​

29. Closing – What Next? [26:00][youtube]​

• Encouragement not to memorize all 27 concepts but to bookmark the video and revisit specific topics when needed.[youtube]​

• Invitation to join the AI Automation Academy / Claude Code track for hands‑on projects and resources, plus a call to comment on which concept “clicked.”[youtube]​