Summary of "It Took Me 847 Hours in Cursor to Learn These 15 Hacks"

Overview

This is a hands‑on tutorial summarizing 15 practical Cursor (AI coding IDE) hacks to speed development and unlock hidden features. Each hack includes what it does and how to enable or use it.

Hacks — quick list with key steps

  1. Show subscription usage

    • Settings → Chat → Usage summary: set to Always to see included usage, monthly spend limit, and billing reset.
  2. Fast mode & model switching

    • Use Shift+Up to toggle modes (Ask, Agent, Plan, Background tasks).
    • Press Cmd// to open the model menu, then type/select a model (e.g., “gpt5”) to switch quickly.
  3. Early access features

    • Settings → Beta → Early access: enable to get new Cursor features earlier.
    • Note: nightly builds exist but can be buggy; use with caution.
  4. Use different models for plan vs build

    • Create a plan with a strong planner model (e.g., GPT‑5 High) and then switch to a faster/cheaper model (e.g., Claude 4.5) for the actual build/execution.
  5. Open multiple project windows

    • File → New Window to work on two projects/repositories side‑by‑side (useful for separate landing page + app projects).
  6. System theme auto‑detection

    • Settings → General → Editor settings → Workbench appearance: pick preferred dark/light themes and enable “window autodetect color scheme” to match your OS.
  7. Split terminals

    • Terminal → New Terminal, then use the split terminal button to run multiple terminal panes simultaneously.
  8. Slash commands & auto‑generate rules

    • Type / to see commands (for example, Generate Cursor Rules).
    • You can feed a docs/blog URL and a fast model (Cheetah) to auto‑create a Cursor rule file (example: Tailwind v4 rule file).
  9. Apply rules selectively

    • In an agent chat, press AddRules to attach specific rule files to a chat/message (useful for applying rules only when relevant, not globally).
  10. Index external docs into Cursor - Settings → Index & Docs → Add doc: paste a docs URL (e.g., Convex docs). Cursor will fetch and index pages so the agent can query full documentation inside the workspace.

  11. (Dangerous) Give Cursor access to local env secrets - Create a file named cursor.ignore and negate default ignores (for example, !.env.local) to allow Cursor to read env files. - Strong warning: do this only in safe/test environments.

  12. Message modes: Send, Queue (Q), Stop+Send - Send: steer the streaming agent without stopping it (good for quick corrections). - Queue (Q): enqueue messages as a to‑do list to be consumed after the current task finishes. - Stop+Send: stops current work then sends a message (can be confusing; use carefully).

  13. Custom modes (beta) - Settings → Chat → Custom modes: create specialized agents/modes with controlled permissions (MCP servers, web search, specific system prompts). - Examples: architect, database analyst, Supabase specialist.

  14. Attach previous chats & show commits (repo history) - Use the @ (add) command to attach prior chat summaries. - @ show commits: enable to give Cursor access to the repository’s commit history and PRs so the agent can reason about past changes and propose reverts/changes.

  15. Duplicate a primed chat (clone context) - In a well‑primed chat, use the three‑dot menu → Duplicate chat to clone the current chat state (context + docs + rules) and branch off multiple tasks from the same starting context.

Notable technologies and models referenced

Practical tips and warnings

Warning: Allowing Cursor to read local environment or secret files can expose sensitive data. Only use that capability in controlled, non‑production environments.

Type of content

This is a tutorial/walkthrough showing settings, workflows, and advanced hidden features — useful as a how‑to guide to get more from a Cursor subscription.

Main speaker / source

The video narrator/creator (unnamed in the subtitles) demonstrates Cursor features and workflow.

Category ?

Technology


Share this summary


Is the summary off?

If you think the summary is inaccurate, you can reprocess it with the latest model.

Video