12 Hidden Settings to 10x Your Claude Code thumbnail

12 Hidden Settings to 10x Your Claude Code

Simon Scrapes · ~14 min
12 Hidden Settings to 10x Your Claude Code
Claude Code Settings & Config Productivity Developer Workflow ~14 min

01 Notification Sound 0:15

The most common fail with Claude Code: you set it on a big task, come back 20 minutes later, and it's been sitting there waiting for permission the whole time.

  • Fix: use /config → enable local notifications (terminal bell)
  • Works with terminals like Ghostty that support bell notifications
  • All settings live in ~/.claude/settings.json
💡 Bonus: Claude Sounds library by Dave Schumacher adds fun voice notifications — search GitHub for dave-schumacher/claude-sounds

02 Mobile Push Notifications 1:53

The remote control feature lets you take away sessions on your phone. Claude pings your phone when it's waiting for input — approve permissions while you're making tea.

  • Add "agent_push_notify_enabled": true to your settings.json
  • Enables push notifications to your phone when Claude needs approval
  • Approve permission prompts directly from your phone

03 Smart Permission Allow Lists 2:35

Not just auto mode. The /fewer-permission-prompt command scans your 50 most recent transcripts, finds commands you always approve, and creates a tailor-made allow list.

  • Gets the speed of skipping permissions without actually skipping permissions
  • Think of it as a dialed-down auto mode, personalised to your workflow
  • Based on your actual usage history — not a generic whitelist
💡 This is the sweet spot between full manual approval and reckless auto-approve.

04 Deny Rules 3:47

Actions you never want auto-approved. Deny rules are checked first, then ask rules, then allow rules.

  • Deny reading .env files — never read API keys into context window
  • Deny git push * — commit freely, but publishing power stays with you
⚠️ Start with these two deny rules at minimum. Deny rules always take priority over allow rules.

05 Model Switching: Opus Plan + Sonnet Build 4:35

Use /model to switch. The pattern: big model where mistakes are expensive (plans, reviews), small model where volume is high (implementation, reading files).

  • Opus Plan: big model does thinking and planning
  • Auto-drops to Sonnet for execution
  • Best setup since Fable 5: Fable for plan, Sonnet for build
  • Half your usage with zero quality loss
💡 This is the single highest-ROI setting change you can make.

06 Effort Level Sweet Spot 6:09

Use /effort to set from low to max ultra code. Independent testing across 20 rounds per level revealed: below max, quality barely moves and correctness doesn't change.

  • Extra tokens purchase re-verification and caveats, not better answers
  • Default for routine work
  • High for genuinely hard problems
  • Max is almost never worth it
⚠️ Don't put Fable on max — it's overkill and burns tokens for marginal gains.

07 Privacy & Telemetry 7:40

Claude sends analytics and error logs by default. Disable with 4 environment variables:

  • CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1
  • DISABLE_TELEMETRY=1
  • DISABLE_ERROR_REPORTING=1
  • DISABLE_NONESSENTIAL_MODEL_CALLS=1
💡 Also go to claude.ai → Settings → Privacy → turn off "Help improve AI models".

08 History Retention 8:42

Default: only saves 30 days of history, then deletes forever. Extend it to keep your conversation history.

  • Add "cleanup_period_days": 365 to your settings.json
  • Keeps a full year of conversation history
⚠️ Setting this to 0 wipes everything — it means "keep nothing", not "keep forever".

09 Auto-Compact Override 9:28

Default compacts at 95% context window. But context rot starts at 70–80% — Claude forgets context and quality degrades well before hitting the wall.

  • Set CLAUDE_AUTO_COMPACT_PERCENTAGE_OVERRIDE=75
  • Compresses before rot, not after
  • 75% is the sweet spot per experts
💡 This prevents the gradual quality degradation that happens silently at high context usage.

10 Status Line 10:25

Use /status-line to show model name and context percentage with a progress bar. Monitor usage in real-time.

  • Shows which model is currently active
  • Displays context window usage as a percentage with progress bar
  • Know when to manually compact before auto-compaction hits

11 Prompt Stashing 11:11

Mid-prompt and realize you need to ask something first? Ctrl+S stashes your current prompt (like Cmd+X). Paste it back with Cmd+V when ready.

  • Ctrl+S — stash the prompt you're currently typing
  • Cmd+V — paste it back after handling the prerequisite
  • No more losing carefully written prompts
💡 Bonus: disable spinner tips with "spinner_tips_enabled": false in settings for a cleaner UI.

12 Remove AI Attribution 12:20

Claude Code was recently attributed to 4% of all GitHub commits. Every commit says "generated with AI" by default.

  • Blank the commit attribution field in settings.json
  • Blank the PR attribution field as well
  • Stop broadcasting AI usage on every single commit

★ The Prompting Shift 12:55

Anthropic's own guidance for the newest models: when the model has enough info, let it act. Stop writing hyper-specific prompts.

  • Give outcome + context + done criteria, NOT step-by-step instructions
  • Tarik's field guide: 2M views in 3 days
  • The bottleneck is now your ability to close the gap, not the model
  • Loop engineering > prompt engineering

🎯 Key Takeaways

  • Notification sound — never lose 20 min to a permission prompt again
  • /fewer-permission-prompt — auto-generates allow list from your actual history
  • Deny rules — always block .env reads and git push
  • Opus Plan + Sonnet Build — half usage, zero quality loss
  • Effort levels — default for routine, high for hard — max is almost never worth it
  • Disable telemetry — 4 one-line ENV settings for privacy
  • Extend history from 30 days to 365
  • Auto-compact at 75% to prevent context rot
  • Status line — real-time context monitoring
  • Ctrl+S stashes prompts mid-typing
  • Remove AI commit attribution — stop broadcasting on every commit
  • New prompting — outcome + context + done criteria, not step-by-step

📚 Resources

⏱ Timestamps

0:00 Intro 0:15 Notifications 1:53 Mobile Push 2:35 Allow Lists 3:47 Deny Rules 4:35 Model Switching 6:09 Effort Level 7:40 Privacy 8:42 History 9:28 Auto-Compact 10:25 Status Line 11:11 Prompt Stash 12:20 Attribution 12:55 Prompting Shift