[LUCKYREDDY31]

Replaces #170, which GitHub closed when I accidentally deleted my fork. The review discussion and the before/after screenshots are on #170.

What changed

Sending a saved draft left it in Drafts, so the email looked unsent and was easy to send twice. Now the draft is removed once its email is sent.

• Review email sends the draft's ID with the review. propose only accepts a draft that belongs to the same person, and links the review to it.
• When the send succeeds, the draft is removed, but only if it still points to that review. A declined, expired, failed or uncertain send keeps the draft. If removing the draft fails, the error is logged and the send still reports success.
• Saving a draft cancels its pending review, and a new review cancels the older one, so a draft never has two sends waiting.
• Every save gives the draft a new revision, and so does linking a review. A review links only if the revision is unchanged since it started, so a save or a second review in the meantime cancels it with a 409 instead of letting it replace, and later delete, the newer draft. This is the race @jerelvelarde found on fix: remove a saved draft once its email is sent #170, plus the same race between two reviews started at once.
• Edit details on a review that was replaced shows that outcome instead of opening an editor for a draft that no longer exists.

Verification

• Tests cover each case from both reviews on fix: remove a saved draft once its email is sent #170. Each new test fails without its fix: the first five on e01080e, the revision tests on a816931, and the two-reviews test without the revision bump on link.
• pnpm lint, pnpm typecheck, pnpm --dir apps/worker typecheck, pnpm test (494/494) and pnpm build:server pass on Node 24. On top of current main (a4944f4) the same checks pass at 531/531, along with pnpm test:browser, both Docker suites and the web, iOS and Android exports.
• In the web app in sample mode: sending a draft removes it; rewriting after starting a review cancels that review and keeps the rewrite; two reviews of one draft leave the first cancelled and send one copy; an open review screen switches to cancelled when a newer review starts in another tab.

Integration limits

• Tested in sample mode only, not with a real Gmail account.
• Drafts left behind by sends before this change are not cleaned up.

[caspianchan31]

Read the full flow against main — the shape is right. Cleanup after a confirmed send, a guarded delete keyed on actionId (a draft re-saved after review no longer matches and survives), and a cleanup failure downgraded to backgroundFailure instead of reading as a failed send: all three are the correct calls. Two things it rests on that I checked: double-approve cannot double-send because decide already atomically claims awaiting_review → executing, and compareAndSwap is a jsonb shallow merge (data = data || patch), so linking a draft never clobbers its body.

One reachable edge this leaves open. On outcome_unknown the draft is intentionally kept — correct, since the email may already have gone out — but in the editor it still looks like an ordinary unsent draft; it only carries actionId. Someone landing back there from an "unknown result" review can reasonably assume nothing was sent and press Send again, producing a duplicate. Worth either surfacing a marker on a draft linked to an outcome_unknown action ("result unknown — check Sent in Gmail before resending") or gating the next send from that draft behind a confirm. It sits right next to this fix, so deciding it now saves a real double-send.

Minor: cleanup runs exactly once, after the terminal put. A transient DB error there leaves the draft linked to a succeeded action with no later trigger to retry. Fine if backgroundFailure is log-only; just confirming nothing is expected to reconcile it.

Not blocking — the core change is solid.