What happens after Claude edits code Diagram

The short answer

• It records each edit, starts a detached review when the turn ends, and shows the result on the next prompt
• Claude never waits for it

Not shown here

• The turn log, router, ranker, settings and ledger, which the hooks and worker use along the way, and the tests, which run only on HARD changes
• Connected but out of scope: Turn log and jobs, Settings, Router, Project tests, Ranker, Ledger

What it guarantees

• Claude never waits for a review: the Stop hook starts a detached worker and returns; results arrive on the next prompt
• Only confirmed findings go back to Claude, at most two turns in a row: an unconfirmed claim sent as an instruction would have Claude fix code that may be correct, and a cap stops fix loops

What happens after Claude edits code

What does polywatch do between one Claude turn and the next prompt?

polywatch has to stay out of Claude's way. A review takes seconds to minutes, and a hook that waited for it would stall every turn. The design question is where the review runs and when its result reaches anyone.

The long read

Read the diagram top to bottom. During the turn, every time Claude edits or writes a file, Claude Code runs the PostToolUse hook. The hook appends the edit to a turn log on disk and returns; it does no review.

When the turn ends, Claude Code runs the Stop hook. The hook writes a job (the turn's edits and the last thing the user asked) and spawns the review worker as a detached process. Then it returns, and Claude is free. This is the only place the two timelines split: from here on, Claude and the worker run independently.

In the background, the worker reads the job and sends the changed files to DeepSeek Flash, which lists the defects it sees. The worker sends up to two of the most serious claims, one at a time, to Claude Opus 5.5, which says whether each holds against the code and gives its evidence. Refuted claims are dropped, the rest are scored, and the ranked result is written to the results folder.

The result waits there until the user sends the next prompt. The UserPromptSubmit hook asks the delivery module for undelivered results and prints two things back to Claude Code: a short message for the user with at most two findings, and, for Claude, only the confirmed ones, so Claude can fix them in its next turn.

Terms used here

turn: One round of Claude Code work: the user's prompt and everything Claude does in answer, up to when it stops.
hook: A command Claude Code runs at a fixed event, such as after a file edit, when a turn stops, or when the user submits a prompt.
tier: polywatch's label for how hard a change is to get right, from its size and whether it touches concurrency, state or protocol logic.
claim: One specific, checkable statement a reviewer makes about a defect: which file, where, and what is wrong.
finding: A claim that survived confirmation (it was not refuted), with a score. Confirmed findings carry Opus's evidence.
confirmed: Opus checked the claim against the code and said it holds.
refuted: Opus checked the claim against the code and said it does not hold. Refuted claims are never shown.
ledger: An append-only file of every review and every outcome, used to calibrate the ranking.
additionalContext: Text a hook returns that Claude Code adds to Claude's context for the next turn; polywatch uses it for confirmed findings.