Claude Code: 15 Features From the Creator's Repo

@keshavsuki

7 from the video. 8 more from the guide.

01

The 7 From the Video

01

Ctrl+G Edit the Plan Before Execution

In Plan Mode, after Claude builds the plan, press Ctrl+G. The plan opens in your external text editor. Edit it directly: remove steps, add constraints, reorder the approach. Claude executes exactly what you save and close.

Confirmed in official docs and changelog

02

AskUserQuestion Let Claude Interview You

In Plan Mode, Claude uses AskUserQuestion to ask structured multiple-choice questions before proposing a plan. You can also trigger it directly by prompting Claude to interview you about a feature before writing the spec.

Available since v2.0.21. 60-second timeout per question. Main agent only, not subagents.

Interview me using the AskUserQuestion tool.

Ask about technical implementation, UI/UX, edge cases,

concerns, and tradeoffs. Don't ask obvious questions.

Dig into the hard parts I might not have considered.

When finished, write a complete spec to SPEC.md.

Then tell me to start a fresh session for clean execution.

03

/btw Zero Context Cost Questions

Type /btw before any question mid-session. The answer appears in a dismissible overlay and never enters conversation history. Context stays clean.

Use for: quick syntax checks, API lookups, confirming variable names - anything you would normally Google mid-session

04

/powerup Interactive Lessons in the Terminal

Interactive animated demos teaching Claude Code features, live inside your terminal. No docs, no tutorials, no tab-switching. Type it in any session to start.

Added April 1, 2026 (v2.1.90). First official in-terminal learning system for Claude Code.

05

effort: Frontmatter in Skills

Add an effort level directly to any SKILL.md. Forces that thinking depth every time the skill runs, regardless of session-level effort setting.

Use on code review, security audit, and architecture skills

# .claude/skills/security-review/SKILL.md

---

name: security-review

description: Review code for vulnerabilities.

effort: high

---

Review this code for security vulnerabilities...

06

CLAUDE_CODE_NO_FLICKER=1 Flicker-Free Rendering

Opts into an experimental alt-screen renderer with virtualized scrollback. Eliminates terminal flicker, enables mouse support, anchors the prompt at the bottom, and keeps memory flat in long sessions.

Added v2.1.88. Opt-in, experimental. Set in ~/.zshrc or settings.json.

# Add to ~/.zshrc or ~/.bashrc

export CLAUDE_CODE_NO_FLICKER=1

# Then reload

source ~/.zshrc

# Or add directly to settings.json

# "CLAUDE_CODE_NO_FLICKER": "1"

07

@agents Named Subagents in @ Mentions

Create named subagents in .claude/agents/ with YAML frontmatter. Type @ in any session and your subagents appear in the typeahead alongside files and MCP resources. Invoke instantly by name.

Subagents added to @ typeahead confirmed in changelog

# .claude/agents/security-reviewer.md

---

name: security-reviewer

description: Reviews code for vulnerabilities.

tools: Read, Grep, Glob, Bash

model: opus

---

You are a senior security engineer...

# Invoke in any session

@security-reviewer

02

8 More From the Guide

08

Shift+Tab Plan Mode

Switches Claude into read-only mode. Reads files, answers questions, uses AskUserQuestion, makes zero changes. Enter with Shift+Tab twice from Normal Mode (first press enters Auto-Accept, second enters Plan Mode).

Use before every major task

# Full plan mode workflow

Shift+Tab (twice)    # Enter Plan Mode

                     # Claude reads codebase, asks questions

                     # Claude proposes plan

Ctrl+G               # Edit plan in your text editor

Shift+Tab (twice)    # Return to Normal Mode

                     # Execute

09

Task Subagent Investigation

Spawn a subagent to investigate a specific part of your codebase. The subagent reads files in its own context window and reports back a summary. Your main context stays clean.

Use for: codebase research, dependency analysis, security reviews, anything that requires reading lots of files

Use a subagent to investigate how our auth system

handles token refresh. Read every relevant file.

Report back: what it does, what it misses, what could break.

10

Sessions Writer/Reviewer Pattern

Session A builds the feature. Session B (fresh context, no bias) reviews it for edge cases and race conditions. Session A addresses the findings. Fresh context catches bugs that the original session normalizes.

Use for every critical feature before shipping

# Session A

Build a rate limiter for our API endpoints.

# Session B (fresh context)

Review the rate limiter in @src/middleware/rateLimiter.ts.

Look for edge cases, race conditions, and consistency

with existing patterns.

# Session A

Here's the review feedback: [paste]. Address these issues.

11

claude -p Headless Automation

Runs a single Claude task non-interactively and exits. Use in CI pipelines, pre-commit hooks, scheduled scripts, and loops over multiple files.

Since v2.x. Use --allowedTools to restrict permissions

# Single task

claude -p "Explain what this project does"

# Loop through files

for file in $(cat files.txt); do

  claude -p "Migrate $file from React to Vue. Return OK or FAIL." \

    --allowedTools "Edit,Bash(git commit *)"

done

# Structured output

claude -p "List all API endpoints" --output-format json

12

/compact Targeted Compression

Not just /compact - pass focus instructions to control exactly what survives. Claude keeps what you specify and compresses everything else.

Run before context gets large, not after it overflows

# Targeted compression

/compact Focus on the API changes and preserve the list of modified files

# Preserve multiple things

/compact Preserve modified files, test commands, and active error states

13

@imports Modular CLAUDE.md

Reference other files from CLAUDE.md using @filename syntax. Keeps each file focused, makes team sharing clean, and lets personal overrides stay out of the project repo.

Files load at session start alongside CLAUDE.md

# CLAUDE.md

See @README.md for project overview

Git workflow: @docs/git-instructions.md

Personal overrides: @~/.claude/my-instructions.md

14

Audit CLAUDE.md Pruning Rule

For every line in CLAUDE.md ask: would removing this cause Claude to make a mistake? If no, delete it. Bloated CLAUDE.md means Claude ignores your critical rules. Target: under 50 lines.

Run monthly. Move task-specific instructions into skills instead.

15

/rewind Checkpoints

Every action Claude takes creates a checkpoint automatically. Press Esc+Esc or type /rewind to open the restore menu. Choose to restore code only, conversation only, or both. Checkpoints persist across sessions.

/rewind restores to a previous state. It is not a compression tool. Use /compact for compression.

# Open the restore menu

/rewind

# Or press Esc twice

# Options:

#   Restore code and conversation  -- full rollback

#   Rewind code only               -- keep conversation, undo file changes

# Checkpoints persist across sessions

# Close your terminal, come back tomorrow, still rewindable

/rewind restores. /compact compresses. Different operations.

/rewind: undo changes by restoring to a saved checkpoint

/compact: summarize conversation history to free context window space

Command / Variable

What It Does

Shift+Tab (x2)

Enter Plan Mode from Normal Mode (first press goes to Auto-Accept)

Ctrl+G

Open plan in external text editor for direct editing

/btw <question>

Side question, zero context cost, dismissible overlay

/powerup

Interactive animated lessons for Claude Code features

/rewind

Open restore menu. Choose checkpoint to restore code, conversation, or both

/compact <focus>

Targeted context compression with preservation instructions

CLAUDE_CODE_NO_FLICKER=1

Flicker-free alt-screen rendering with mouse support

claude -p "prompt"

Non-interactive single task, use in scripts and CI