Apr 15 • 📚 Resources & Finds
Claude Code Modes - Auto mode < GSD YOLO < --dangerously-skip-permissions
this one is inspired by your great question.
We all want to optimize our efficiency and our experience with using tools.
Few things to note:
1. Using Claude Code in VS Code terminal.
2. On Claude Max plan.
3. Using Linux in a windows virtual machine.
4. Blast Radius defined
4.1. Small - a mistake affects one file, function, or is easily reversible (git checkout, undo, local-only)
4.2. Large - a mistake affect production systems, shared infrastructure. Examples: force-push to main, rm -rf, dropping a database, leaking a secret).
Out of the box claude code comes with several modes that you can press "Shift TAB" to cycle thru.
- Accept edits on
- Default
- Auto mode on
- * bypass permission on ( this has to be enabled when you start new claude code session)
I added 1 more from "get shit done" (GSD) repo
**************don't download github repos without doing your own research************
5. GSD YOLO mode.
------------------------------------------------------------
Blast Radius (smallest → largest)
Accept edits on < Default < Auto mode < GSD YOLO < --dangerously-skip-permissions
----------------------------------------------------------
Three Execution Modes — High-Level Comparison
1. --dangerously-skip-permissions
What: Claude Code CLI flag. Skips every permission prompt for tool use — file edits, bash, network, everything.
How to use: claude --dangerously-skip-permissions at launch.
Scope: Entire session, all tools.
Pros: Zero friction. Useful for sandboxed VMs, ephemeral containers, CI runners.
Cons: Disables the main safety layer. One bad command = real damage. Your CLAUDE.md forbids it. Security hooks (install-guard,
gsd-security-guard, credential-sandbox) still fire, but nothing else gates tool use.
When: Never on your host machine. Only in throwaway environments.
---
2. Auto Mode (Claude Code native)
What: Continuous-execution mode where I proceed without asking clarifying questions on low-risk work.
How to use: Toggle via /auto or by exiting plan mode into auto. Exit with /auto again or a course correction.
Scope: My behavior — I make reasonable assumptions instead of pausing. Permissions, hooks, and the security perimeter are fully intact.
Pros: Fast iteration on well-defined tasks. Still safe — hooks catch everything dangerous. Good for multi-step builds where you trust the
plan.
Cons: I may assume wrong on ambiguous calls. Not appropriate for exploratory or high-stakes work where you want checkpoints.
When: Executing an approved plan, routine edits, build-and-verify loops.
---
3. GSD YOLO Mode
What: A GSD-internal setting (mode: yolo, auto_advance: true) that makes the GSD orchestrator auto-advance phases without user verification
checkpoints.
How to use: Create bypass token first (token expires in 30 min):
echo "$(date +%s):$(openssl rand -hex 32)" > ~/.claude/.yolo-bypass
Then set YOLO in the GSD config or invoke with auto-advance. Without the token, the GSD Security Guard hook blocks it.
Scope: Only affects /gsd:* commands — skips /gsd:verify-work and phase checkpoints.
Pros: Rips through multi-phase GSD projects without stopping. Paired with ralph-loop for fully autonomous builds (per your build-workflow
preference).
Cons: You lose per-phase human review — bugs compound across phases. Destructive if the plan is wrong. Not a substitute for a good plan.
When: Trusted plan, well-scoped phases, recoverable git state, you're monitoring output.
___________________________________
We all want to be efficient and effective so make choices that work for your use case.
4
7 comments
Claude Code Modes - Auto mode < GSD YOLO < --dangerously-skip-permissions
skool.com/cliefnotes
Jake Van Clief, giving you the Cliff notes on the new AI age.
Leaderboard (30-day)
1
+1407
2
+1272
3
+649
4
+649
5
+641
Powered by