Summary of "Secret codes for Claude 😳"

Overview

The video claims to share seven “secret codes” (special commands/skills) for Claude that can increase productivity (up to “10X”) by:

changing reasoning depth,
reducing token usage,
automating recurring tasks,
improving workflow management.

The presenter says these commands work for both new and experienced Claude users, and the video includes live walkthroughs plus a “free swipe file” (not included in the subtitles).

The Seven “Secret Codes”

1) Ultra Think (keyword)

Purpose: Make Claude spend the maximum time on reasoning for the specific prompt.
Behavior: It’s said to override the user’s default reasoning effort (e.g., switch from low/medium to high/max effort just for that request).
Suggested benefit: Save tokens by keeping normal tasks at lower reasoning effort, then using Ultra Think only for complex/technical prompts.
Where it applies: Mentioned in contexts like Claude Code / Claude cowork (subtitles imply it’s available across modes).

2) /caveman (open-source skill)

Claimed benefit: Significantly reduce output tokens.
Installation: “Ask Claude” to install the open-source skill.
Usage: Run /caveman in prompts.
Suggested use case: Especially useful on lower tiers (e.g., around a ~$20/month plan) to help manage usage limits and token costs.

3) /insights (usage analyzer)

Usage: Type /insights.
Behavior: Claude is said to analyze all sessions from the past 30 days.
Output: Generates a detailed HTML report including:
- what’s going well,
- what’s holding you back / anti-patterns,
- improvements and specific, tailored recommendations based on your actual usage (not generic advice).
Recommendation: Run about once a month as a “reality check.”

4) /loop (recurring tasks on your computer)

Usage: /loop with parameters such as an interval and the skill to run (example given: writing a draft post).
Scheduling behavior:
- repeats at a short interval (minimum stated as 1 minute),
- tasks expire after 3 days,
- tied to your current session,
- requires your computer to be on to run.
Example: “in 5-minute intervals” it writes something.
Downsides: Local dependency and the 3-day expiry.

5) /schedule (recurring tasks in the cloud; experimental)

Usage: /schedule.
Key difference vs /loop:
- runs independently in the cloud,
- does not require your computer to remain on.
Behavior: It clones your repo/project and runs scheduled tasks without needing local file access.
Constraints mentioned:
- minimum interval is longer than /loop (stated as 1 hour minimum),
- no local file access,
- no 3-day expiry mentioned,
- caps on runs per day, such as:
  - Pro: 5 runs/day
  - Next tier: 15 runs/day
  - Team/Enterprise: 25 runs/day
Suggested use cases: daily standups, recurring automations, and scheduled recurring reports where local access isn’t needed.

6) /btw (“by the way” parallel follow-up)

Usage: /btw while Claude is already busy.
Claimed behavior: Lets you ask another question or run another prompt without interrupting Claude’s main long-running task.
Workflow benefit:
- avoids starting a new session (which would lose conversation context),
- enables parallelization: main work continues while the “btw” question is answered.

7) /clear (reset session context)

Usage: /clear (subtitles also show “/cle” as a likely typo).
Claimed behavior: Deletes all previous context in the current session.
Reasons given:
- accuracy can drop when irrelevant context accumulates,
- can save tokens because Claude won’t need to process old context.
Practical recommendation: Use it when switching tasks (example: switching from a YouTube script to a TikTok script).

Main Speaker / Source

Single presenter/host (name not provided in the subtitles).
The speaker appears to be the channel owner who:
- claims to teach AI “to millions of people,”
- reports “30 million views last month,”
- says they’ve tested thousands of Claude prompts.