Learning Timeline
Key Insights
Benefits of Isolated Context Windows
Each 'Agent' you create has a fresh 200k token context window. This means heavy tasks won't 'clutter' or consume the token limit of your main conversation (Main Context).
Output Style as a Persona
You can create custom 'Output Styles' (like Script Writing or Brutal Critic) that act as a permanent system prompt every time you access that project.
Prompts
Research Agent Builder Prompt
Target:
Claude Code
You are a research expert. You will use Gemini in headless mode to research. Use it like this: [insert command examples]. Reference existing documents in the directory using @ symbol.
Step by Step
How to Install and Set Up Claude Code
- Open the terminal on your computer.
- Run the command 'npm i -g @anthropic-ai/claude-code' for a global installation.
- Navigate to your project directory using the 'cd' command.
- Launch Claude by typing 'claude' in the terminal.
- Follow the on-screen instructions to log into your Anthropic account.
- Grant Permission for Claude to access your project folder when prompted.
How to Build and Use AI Agents in the Terminal
- Start a Claude session in your terminal.
- Type the '/agents' command to open the terminal menu for agent management.
- Select 'Create a new agent' and give it a name (e.g., 'Research Guru').
- Define the agent scope as either 'Project-specific' or 'Global'.
- Write a persona or specific task for the agent (System Prompt).
- Press 'Enter' to save and 'Esc' to return to the main terminal.
- Use the '@' symbol followed by a document or agent name to mention them in the conversation.
- Press 'Ctrl+O' to monitor agent activity running in the background.
Running Claude Without Restrictions (Danger Mode)
- Stop the current Claude session by pressing 'Ctrl+C' twice.
- Relaunch Claude using the additional argument: 'claude --dangerously-skip-permissions'.
- Use the 'claude -r' command if you want to resume a previous conversation.
- The agent will now perform web searches and file edits automatically without asking for manual confirmation every time.
Switching Working Modes (Thinking & Planning)
- Press 'Tab' while in the Claude terminal to activate 'Thinking Mode' for complex problem-solving.
- Press 'Shift + Tab' to switch to 'Planning Mode'.
- In 'Planning Mode', Claude will first outline an action plan.
- Review the plan and type 'Yes' or 'Approve' to let the AI execute the tasks automatically.
