Claude Code plan mode vs. a plan you can comment on

[thisisnsh]

Claude Code plan mode vs. a plan you can comment on

npm install --global @thisisnsh/planx

Plan mode is the pause Claude Code takes before writing code. It's genuinely good and I use it every day. This post is about where it stops, and what to do after.

What plan mode does well

It makes the agent research first and propose second — stopping it from editing files while it's still working out what to do, and putting an accept/reject gate in front of the work. That's the hard part solved. Getting an agent to think before it acts is the whole battle, and plan mode wins it.

What it hands you at the gate

A message. Long, arrived all at once, sitting in the same scrollback you're using for everything else, with an accept button under it. That creates four problems:

• Not addressable. You can't point at a line — disagreement has to become prose ("in the third section, where it says the migration runs first, actually...") — so small objections aren't worth typing and go unraised. A 200-line plan with four wrong lines gets accepted whole.
• Not comparable. Ask for a revision and you get another long message. Nothing shows which fifteen words changed, so re-reading the whole thing is the only honest option and skimming is the realistic one.
• It disappears. The plan you approved is somewhere back in scrollback — there's no copy to check the result against.
• All-or-nothing. Approve or reject. No gesture that means "this, but not line 40."

None of this is a bug in plan mode — it's what happens when a plan is a message instead of an artifact.

What changes when the plan is a file

PlanX stores the plan as a version with a number instead of a message.

Problem	What replaces it
Not addressable	v to select lines, f to comment on exactly those lines
Not comparable	A word-level diff against the previous version
Disappears	A file in ~/.planx, every version kept
All-or-nothing	Comment, rewrite a line yourself with e, or approve one exact version

One more thing chat can't do at all: because the plan is a file rather than a conversation, the agent that builds it doesn't have to be the agent that wrote it.

They aren't alternatives

PlanX expects you to leave plan mode — the skill exits it with a one-line stub before doing anything else. The accept/reject gate is where PlanX starts, not something it replaces.

Plan mode gets you a considered plan. PlanX gets you a decision about it.

When plan mode alone is enough

Often — if the change is small, you can hold the whole plan in your head on one read, and you're building in the session you planned in, the gate is the whole review. A stored artifact is overhead you don't need there.

Reach for PlanX when:

• the plan is too long to review in one pass
• you want to disagree with part of it, not all of it
• the agent that plans isn't the agent that builds
• you're coming back to it later — tomorrow, after lunch, after a meeting

That last one is easy to underrate. Plan mode assumes the review happens now, while the message is on screen. Real reviews get interrupted.

What the diff does to revision

This is the biggest change, more than the comments:

• in chat, asking for a change means re-reading the whole plan to see what moved
• so you ask for fewer changes than you want
• and approve the second version more readily than the first because you're tired of reading

When the revision arrives as a diff instead, re-reading costs seconds. You end up asking for the changes you actually wanted.

More: Versions and diffs · Feedback and edits

Try it on your next real plan

npm install --global @thisisnsh/planx

Then /planx <whatever you were about to ask Claude Code to build>, and open the review. The comparison means more on a plan you actually care about than on a toy one.

More: Plan mode vs. a reviewable plan · Why review an agent's plan

---

MIT, nothing hosted, plans are local files. If this changed how you review a plan, a star helps the next person find it.