The pre-flight checklist
Most prompts are easy. ls — yes. rm -rf / — no. The interesting ones live in the middle: a command you don’t recognize, a file edit that touches more than you expected, an action that sounds fine but you can’t quite tell.
That middle is where this level pays off. You’re about to run through a four-question checklist that takes maybe ten seconds and catches almost every prompt worth catching.
The checklist
When a prompt appears and your finger hovers, ask these in order:
1. What is this actually going to do?
Not what kind of action — the actual action. Read the command. Read the file path. If you can’t tell what it does, you’re not ready to approve it.
2. What’s the blast radius? (Lesson 6.3)
Where on the ladder does it sit? Your terminal, your scratch folder, your project, your machine, or someone else? The higher the rung, the more carefully you read.
3. Is it a one-way door? (Lesson 6.4)
If this is wrong, can you get it back? Two-way: approve and move on. One-way: read it again before clicking.
4. Do I have my safety net? (Lesson 6.8)
If it’s a file change, is git up to date? Did you commit before letting Claude loose? A safety net doesn’t make you reckless — it makes the read on the previous three questions less stressful.
The fifth question, for the truly weird ones
Sometimes a prompt looks fine to a careful reader but you can’t tell why Claude wants to do that specific thing.
When in doubt, ask Claude to explain what it’s about to do and why — before you approve.
You can literally say “wait — what’s that command for?” and Claude will explain. Two things often happen next:
- The explanation makes sense, and you approve.
- The explanation reveals Claude is about to do the wrong thing for a reasonable-sounding reason, and you save yourself a mess.
This is the single most underused habit in Claude Code. Use it whenever the prompt makes you pause.
After this week, you won’t do this consciously
The checklist looks like four questions on a page. In practice, after the first week or two, it collapses into a single instinct: you glance at a prompt, the right answer is already obvious, and the only ones that slow you down are the ones that should slow you down.
That’s the goal of this level. Not “be paranoid about every prompt.” Not “trust Claude and click through.” Just: notice the prompts that matter, and let the others fall through.
What’s next
You’ve finished Trust & Safety. Up next: bigger work — tasks that don’t fit in a single prompt, plan mode in its own home, and the to-do system Claude uses to keep its head straight on long jobs.