Code review follow-ups Claude Code picks up next time

Claude Code and Pando, for code review follow-ups.

You run /code-review before opening the pull request. It "reports correctness bugs and reuse, simplification, and efficiency cleanups", and "The findings arrive in your conversation when the review completes." Then you choose. Some belong in this change. Others wait: a race that was there before the branch, a duplicated parser the diff only touched, a cleanup that would double the diff. A finding you leave stays in that conversation, and "Each Claude Code session begins with a fresh context window."

After /code-review, tell Claude Code which findings to fix in this change, and have it file each of the others in Pando as a todo under a Follow-ups bullet, with the file, the line and why it waits in the note. A later session, on this computer or another, searches Follow-ups for what is still open and ticks each follow-up when its fix is committed.

Set it up

In a terminal: claude mcp add --transport http --scope user pando https://pando.ink/mcp
Start claude, run /mcp, select pando and choose Authenticate.

Make a bullet called Code in Pando, with Follow-ups inside it: Claude Code has to name the bullet it writes under, and Pando gives it no default. When Pando asks, choose Read and write your outline, or hold Claude Code to Code as below, which lets it write there either way. With --scope user the server is "available across all projects on your machine". Another computer needs the two steps and the habit below too, and its sign-in is a second agent with your whole outline, so hold it as well. Then put the habit in ~/.claude/CLAUDE.md, the file for "Personal preferences for all projects":

After a code review, when I say a finding waits, file it
in Pando as a todo under Follow-ups, in this repository’s
bullet: the finding on the line; the file and line, the
branch, and why it waits in the note.
At the start, search this repository’s Follow-ups for
layout:todo is:incomplete. Before fixing one, read the
file; if the finding no longer holds, tick it and say why.
When you commit a fix, tick it and append the commit.

The trigger is your word because the review itself runs as a subagent that "follows your CLAUDE.md like any Claude Code session", and what waits is your call. Claude treats CLAUDE.md "as context, not enforced configuration", so a session may still miss the habit; then ask.

After the review

When the findings arrive, say which to fix and which wait: “Fix the nil check and the off-by-one now. The race and the parser wait.” Claude Code fixes the first two and files the other two. Follow-ups afterwards, with one from last week already ticked:

- Code
  - Follow-ups
    - inventory-service
      - [ ] Stock can go negative when two orders reserve at once
            note: internal/stock/reserve.go:88, branch fix-sku-lookup.
                  Was there before this branch; needs a row lock,
                  and this change is a one-line fix.
      - [ ] Merge the two SKU parsers
            note: internal/sku/parse.go:14 and api/order.go:203,
                  same branch. A reuse cleanup; out of scope.
      - [x] Cache the warehouse list per request
            note: … Done in 9b7d31e.

The box belongs to each follow-up, so Code, Follow-ups and inventory-service stay plain bullets. A ticked one stays in its place, struck through. On the phone, at pando.ink/app, tap a box to tick one, or type [] and a space on an empty line to add your own.

The next session

The session runs the search from the habit, focused on the repository’s bullet. layout:todo is in it because is:incomplete alone also matches every plain bullet nobody ticked. Each hit arrives with its note, so the file, the line and the reason come with the task, and appendNote adds the commit after them without touching the rest.

What it can reach

Approved on Pando’s consent page, Claude Code reaches your whole outline, whichever level you picked. To keep it to your code notes, open Code, open Agents and API keys in Settings, tap its row under Your agents, and press Let it reach only “Code”, the bullet you are in. From then on it reads and writes Code and everything under it, even on a read-only approval, and nothing outside. The press keeps one bullet and drops any other, so it goes on Code, not Follow-ups: Code has room for a bug list too, and what Claude Code remembers lands there, beside the follow-ups rather than among them. Its old memory bullet goes out of reach.

Where this is not the answer

Follow-ups in one repository. A TODO comment at the line, or a file in the repository, is enough: Claude Code reads and edits both, and git carries them to every computer.
Follow-ups your team should see. Put them on the pull request or in the issue tracker: /code-review --comment "posts the findings on a GitHub pull request as inline comments", and on GitHub "You can open a new issue from a comment in an issue or pull request".
Review comments on an open pull request. /autofix-pr starts a cloud session that "pushes fixes when CI fails or reviewers leave comments".

Pando is for the case in between: your own follow-ups from several repositories, next to the rest of your notes, readable and tickable on the phone with no checkout at hand.

Deeper

Connect an agent, about two minutes · The twelve tools · Which note apps an agent can reach · Who runs this