Skip waits as weft's own concept, let the install register built versions, and let a waiting run leave the database asleep

.agents/skills/releasing/SKILL.md

@@ -22,6 +22,16 @@ into `main` through a PR, and the merge is what publishes.
    to the stores. The `extension versions bumped` CI check refuses a
    PR that forgot; `scripts/check-extension-bump.sh origin/main` is the
    same check locally.
+   In the same PR, write what changed into `CHANGELOG.md`, for the
+   people who update their weft: what they will notice, what they have
+   to do (a `./setup.sh` that asks, a command to run again, a setting
+   that moved), and anything breaking first. Each release gets its own
+   section, newest first, headed with its date and its version, the VS
+   Code extension's version the PR bumped to (`## 2026-10-09 (0.2.267)`);
+   a release that bumps no extension names its merge's short commit
+   instead. Work not merged yet stays under `## Unreleased`, and the PR
+   that ships it renames that heading. Write it plainly, for a person
+   reading it before they run `./setup.sh`, never as a list of commits.
 2. `gh pr create --base main --head <branch>`.
 3. Never `--auto`. Wait for every check to finish (`gh pr checks <n>
    --watch`), then read what the PR collected besides the checks before

.claude/skills/releasing/SKILL.md

@@ -22,6 +22,16 @@ into `main` through a PR, and the merge is what publishes.
    to the stores. The `extension versions bumped` CI check refuses a
    PR that forgot; `scripts/check-extension-bump.sh origin/main` is the
    same check locally.
+   In the same PR, write what changed into `CHANGELOG.md`, for the
+   people who update their weft: what they will notice, what they have
+   to do (a `./setup.sh` that asks, a command to run again, a setting
+   that moved), and anything breaking first. Each release gets its own
+   section, newest first, headed with its date and its version, the VS
+   Code extension's version the PR bumped to (`## 2026-10-09 (0.2.267)`);
+   a release that bumps no extension names its merge's short commit
+   instead. Work not merged yet stays under `## Unreleased`, and the PR
+   that ships it renames that heading. Write it plainly, for a person
+   reading it before they run `./setup.sh`, never as a list of commits.
 2. `gh pr create --base main --head <branch>`.
 3. Never `--auto`. Wait for every check to finish (`gh pr checks <n>
    --watch`), then read what the PR collected besides the checks before

.kilo/skills/releasing/SKILL.md

@@ -22,6 +22,16 @@ into `main` through a PR, and the merge is what publishes.
    to the stores. The `extension versions bumped` CI check refuses a
    PR that forgot; `scripts/check-extension-bump.sh origin/main` is the
    same check locally.
+   In the same PR, write what changed into `CHANGELOG.md`, for the
+   people who update their weft: what they will notice, what they have
+   to do (a `./setup.sh` that asks, a command to run again, a setting
+   that moved), and anything breaking first. Each release gets its own
+   section, newest first, headed with its date and its version, the VS
+   Code extension's version the PR bumped to (`## 2026-10-09 (0.2.267)`);
+   a release that bumps no extension names its merge's short commit
+   instead. Work not merged yet stays under `## Unreleased`, and the PR
+   that ships it renames that heading. Write it plainly, for a person
+   reading it before they run `./setup.sh`, never as a list of commits.
 2. `gh pr create --base main --head <branch>`.
 3. Never `--auto`. Wait for every check to finish (`gh pr checks <n>
    --watch`), then read what the PR collected besides the checks before

CHANGELOG.md

@@ -3,7 +3,95 @@
 What changed between versions of weft that you need to know about when you
 update. To update, pull and run `./setup.sh` again.
 
-## Unreleased: weft no longer needs Kubernetes
+## 2026-10-11 (0.2.267)
+
+**On Google Cloud, a run waiting on a person lets the database sleep.**
+
+A run that cannot pause (a route still answering its caller, a trigger with
+`recorded: false`, a live stream between steps) keeps its question in its
+worker while it waits. Until now that kept the database awake for the whole
+wait. Now nothing touches the database until the answer or a cancel arrives:
+the answer reaches the worker through a news topic weft creates in your
+Google project. Upgrade as usual (merge upstream into your fork and run the
+install workflow again): it creates the topic and the rights that go with it.
+
+A worker on Google Cloud can keep one connection to weft for at most an
+hour. If every run on it is still waiting at that point, the worker stops:
+runs that can pause carry on elsewhere, and a run that cannot pause ends,
+saying which limit ended it. To wait longer, make the run one that can pause.
+
+**Five instance nodes lost "Instance" from their names.**
+
+`StartInstanceInfra`, `StopInstanceInfra`, `TerminateInstanceInfra`,
+`ActivateInstanceTriggers` and `DeactivateInstanceTriggers` are now
+`StartInfra`, `StopInfra`, `TerminateInfra`, `ActivateTriggers` and
+`DeactivateTriggers`. Rename them in your `.weft` files. Their `instance`
+input is now optional: if you leave it unwired, the node acts on the
+program's own copies, the ones every instance shares.
+
+**GmailSend can send from another address.**
+
+The new optional `from` input sends from one of the account's "Send mail as"
+addresses set up in Gmail; any other address fails naming the ones that work.
+To read that list, the Google connection needs "Read mail" or "Organize mail"
+ticked. Replies are unchanged: they need one of those two or "Read mail
+headers".
+
+**A program built before this version needs one rebuild.**
+
+How a waiting step receives its answer changed, so a worker built before this
+version fails when a waiting step gets its answer. Resync each program once:
+`weft resync` if its triggers are on, `weft build` otherwise. Either one
+builds the worker again on this version of weft and moves the program to it.
+
+**Skipping a question ends the step that asked it.**
+
+When a person skips a waiting question (Skip in the browser extension, or
+`POST /signal/{token}/skip`), the step that asked now ends skipped: its
+outputs close and the steps reading them skip too. It used to answer the
+question with null, so if your program reads a null answer as "skipped", it
+now sees the step skipped instead. Skipping a trigger is refused with a `409`;
+it used to fire the trigger.
+
+**A Google connection can hold Gmail alone.**
+
+The connect check now asks Google who the account is, through two permissions
+every connection asks for ("Know who you are" and "See your email address"),
+instead of reading Drive. So a connection with only Gmail, or only Calendar,
+ticked now passes. If you run your own shared Google app, its `covers` does
+not need to list those two.
+
+**Node authors: inputs can be required only when wired.**
+
+`requiredWhenWired: true` on an input means: left unwired, the node runs
+without it; wired, a wire that delivers nothing skips the node. It is for an
+input whose absence means something, like the instance nodes' `instance`.
+
+**Builds no longer hold one request open.**
+
+`weft build`, `run`, `activate`, `resync` and `infra start` used to keep one
+request open to the install for the whole build. The install now answers as
+soon as the builds start, and the CLI follows each one to its end. If a cloud
+install does not answer, the error names that install instead of telling you
+to start the local daemon. Update the VS Code extension too: it needs this
+version.
+
+**A route's limits can overshoot briefly after a quiet spell.**
+
+When a worker copy wakes after a quiet spell, its first calls are decided on
+what it heard before it went quiet, until weft's broker answers it: under a
+second normally, a few seconds when the broker was asleep too. During that
+window, copies waking at the same moment can each let in up to the full
+`callsAtOnce`. We chose this so the first call after a pause never waits for
+the broker to start.
+
+**The editor's activate button names the triggers that are off.**
+
+When some triggers are off while the others run, the second button now says
+which ones ("Reactivate a, b" or "Reactivate a, b and 3 more") instead of
+"Reactivate the rest".
+
+## 2026-10-09 (extension 0.2.266): weft no longer needs Kubernetes
 
 **Breaking: your run history does not come along.**