Claude Code · Daily Drop
VOL 03 · 2026-05-09 · APOGEE

Plot the burn,
then light the engine.

Every orbit you've ever seen began on a whiteboard. Today's drop is a postcard from low Earth orbit — and the Claude Code feature that, like a flight director, refuses to start an irreversible maneuver without a plan everyone has signed off on.

Field noteWhy the maneuver matters more than the engine

A spacecraft does almost nothing the way intuition says it should. To rendezvous with a target ahead of you, you slow down. To slow down, you sometimes burn forward. The intuitive move — point at the target and accelerate — sends you into a higher, slower orbit, and the gap grows. The right move is counterintuitive, often non-local, and always cheaper than the wrong one.

That's why every flight plan goes through review long before any propellant moves. A Hohmann transfer to geostationary orbit is roughly two engine burns and several hours of coasting — but weeks of math, simulation, and red-team review.

NumbersWhat it actually costs to get up there

Reaching low Earth orbit requires about 9.4 km/s of delta-v — most of which is spent fighting gravity and atmosphere on the way up. From LEO to geostationary is another ~3.9 km/s. From LEO to a lunar transfer is a relatively cheap ~3.2 km/s. Every kilometre per second is bought with mass, and every kilogram of mass requires more propellant to lift it.

9.4
km/s to LEO
~90
min per orbit
408
km · ISS altitude
Today's Claude Code feature · Plan Mode
STATUS: READ-ONLY · ENGINES SAFED

Show your flight plan before you light the candle.

Plan Mode puts Claude Code into a strictly read-only loop: it can explore the codebase, run searches, and read whatever it needs — but it cannot edit files, run commands that mutate state, or commit anything. It uses that quiet to write a step-by-step plan you review and approve before any work begins. Toggle it on with Shift+Tab twice, or launch a session with --permission-mode plan.

It's the difference between "trust me, I'll fix it" and a flight director reading a checklist back to the room. Especially useful before refactors, migrations, multi-file edits, or anything you'd hate to undo from a backup. 

# Engage plan mode for the session
$ claude --permission-mode plan

# Inside a session, toggle with Shift+Tab (twice)
> /plan  # review-only · proposes a step-by-step plan, executes nothing

# When the plan looks right, exit plan mode and let it run.
> approve  # Claude switches to edit/execute and follows the approved plan

Pairs well with: subagents (split investigation across parallel readers), /clear (start the next maneuver with a clean context), and writing rollback notes into the plan itself.