A ten-minute tactical loop through the keyboard tricks, hidden modes, and headless workflows that make Claude Code feel less like a terminal and more like a coding partner.
Claude Code's terminal interface hides ten critical productivity features—from planning mode and image analysis to headless scripting—that transform it from a basic coding tool into a full-featured development environment.
Claude Code becomes dramatically more useful once you treat it as a coding partner with hidden modes rather than a plain terminal. The video walks through ten mechanics: shift+tab toggles planning mode so Claude researches and proposes architecture before touching code, Ctrl+V (even on Mac) pastes screenshots for UI analysis, /init writes a CLAUDE.md that auto-loads as context in every future session, IDE integration plus a popped-out window enables two-monitor flow, custom slash commands live as markdown files in .claude/commands, the keyword ultrathink unlocks the largest reasoning budget, double-escape forks the conversation to an earlier prompt, dropping URLs triggers ad-hoc web fetches, and claude -p runs headlessly so you can pipe shell output like npm audit straight into scripted workflows.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →
Claude 4 is the best AI coder, but it lives in a terminal — here are 10 tips to tame it.
Shift+Tab twice → planning mode. No code edits, just a written plan with web searches and a project-structure report. Shift+Tab again to leave.
On Mac, Cmd+V doesn't paste screenshots into Claude Code — Ctrl+V does. Once pasted, it shows up as 'image 1' and Claude reads UIs surprisingly well.
Run /init to scan the project and generate a CLAUDE.md. Auto-loaded into every new session as context — basically Cursor rules, except free and persistent.
Claude Code auto-installs an IDE extension when it detects one (Cursor/VS Code). Two-way comms: highlighting code in the editor scopes Claude's edits; visual diffs appear in the editor.
Right-click the Claude Code terminal → 'move into new window' → drag to second monitor. Editor on one screen, Claude on the other.
Create .claude/commands/<name>.md with arguments and a prompt template. Invoke with /project:<name> <args>. Reusable, project-specific workflows.
Drop 'think', 'think hard', or 'ultrathink' into a prompt to escalate the thinking budget. Gray thinking tokens appear before the answer.
Esc once = interrupt. Esc twice = jump to any previous message, fork the conversation, and clear history forward. Caveat: auto-accepted edits don't get rolled back.
Outside of planning mode Claude rarely searches the web on its own — but if you paste a URL into the prompt, it'll fetch and use the contents (great for pulling in fresh docs).
claude -p '<prompt>' runs Claude in the terminal with no UI, just stdout. Pipe inputs in, pipe outputs to files — e.g. npm audit --json | claude -p '...' > vulnerabilities.md.
Hope you learned a thing or two — more AI software-development videos coming, subscribe.
Pick a single tool you use every day, find ten quirks nobody talks about, and shoot them as one numbered list with hard cuts and title cards.
“If you ever done software development before, you're gonna know that a little bit of planning goes a long way. And Cloud is a secret way to do this.”
“You'd think on a Mac would just be your normal command V. Click on that, nothing happens. The trick is you actually have to do a control V, which is counterintuitive.”
“Every time you do a new chat or start a new session with Claude code, it puts this into the context. So if you put this project in the side for six months, came back to Cloud Code later, it would actually bring in all this and remember all this as context.”
“All the way up to ultra think, which if it sees that word, it's really gonna grind away and try to give you the very best answer.”
“If you hit the escape key twice, it's gonna give you all the previous messages you had. And then you can go back to the one they interrupted, say. And it's gonna clear the history and go back to that prompt, which is a nice clearing way basically to fork the conversation, get it back on track.”
“You can do the Claude dash p again for the headless mode, and then ask it to prioritize these security vulnerabilities and describe them what's critical fixes needed, and then pipe all that to vulnerabilities dot m d.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
The whole video is a single, well-disguised promise: "Claude Code is the best coding model out there, but it's hiding ten of its best tricks behind quirks and keyboard shortcuts no one told you about." AI Unleashed opens with the Claude Code splash screen on one side and his face on the other — terminal-shy viewer, meet patient teacher — and immediately launches into tip #1 before anyone can scroll away.
Shift+Tab once = auto-edits on; twice = planning mode on; third time = back to normal. One key, three workflow modes.
/init scans the codebase and writes a markdown file (architecture, conventions, components). Auto-loaded into every new session. You can hand-edit it to inject permanent context.
Magic words inside a prompt that escalate Claude's reasoning budget. The gray text streamed before the answer is the thinking trace.
Drop a markdown file at .claude/commands/<name>.md with $ARGUMENTS placeholders → invoke as /project:<name> <args>. No code, no plugin system, just files.
claude -p turns Claude Code into a stdin → stdout filter. Composes with pipes, redirects, and any other CLI tool.
“Make sure you subscribe to the channel. And I hope you have an amazing day. I'll talk to you in the next one.”
Soft outro CTA, no urgency. Mid-roll newsletter pitch at 6:30 is more interesting — wedged between two tips so it doesn't feel like a sponsor break.
00:00
00:1000:17
00:25
00:32
00:40
00:47
00:54
01:02
01:09
01:16
01:2401:32
01:38
01:47
01:53
02:00
02:07
02:15
02:22
02:29
02:3702:45
02:51
02:59
03:06
03:13
03:21
03:28
03:35
03:43
03:5003:57
04:04
04:12
04:1904:29
04:31
04:41
04:5104:55
05:03
05:10
05:18
05:25
05:32
05:40
05:47
05:5406:01
06:09
06:16
06:2306:33
06:39
06:45
06:53
07:00
07:07
07:15
07:2207:31
07:37
07:44
07:51
07:58
08:0608:14
08:20
08:28
08:35
08:42
08:50
08:57
09:04
09:12
09:19
09:26
09:3309:41
29:19A systems playbook for Claude Code and Codex: the five-step loop, the instruction files, and the guardrails that separate trustworthy AI output from expensive rework.
July 16th
17:17A 17-minute screen-recorded walkthrough of installing, configuring, and running the mattpocock/skills repo on a real codebase — from a vague idea to a reviewed, committed change.
July 16th
27:47A reluctant 28-minute tour of the Claude Code features every competing harness should steal.
June 17th
16:16A viral Boris Cherny tweet mapped five stages of Claude Code adoption — this breakdown turns the jargon into a pizzeria anyone can picture.
August 6th
16:17A breakdown of why maxing out effort settings on Claude, GPT, Grok, and Gemini rarely makes the output better — and the framework for picking the right level every time.
July 13th
1:14:40A 75-minute live stream where one vague feature idea is talked into a decision-complete spec, one parallel grilling session at a time.
July 13th