verdande could not be added to Claude at all. The documented setup —
POST /api/v1/mcp with an Authorization header — cannot be configured from
Claude's custom-connector dialog, which asks for a name, a URL and OAuth
client credentials and has no field for a bearer token. The instructions
described something the interface does not offer.
Worse, the obvious guess failed silently in the most misleading way
available: /mcp was not a route, so the SPA fallback answered 200 with a
page of HTML. The connector reported a successful connection and then
failed to parse the response, which reads as a broken client rather than
a missing endpoint.
So /mcp exists now and takes its credential in the query string, exactly
as the calendar feed does and for the same reason: a client that cannot
be told to send a header means the address is the credential. The
request logger records paths without their query strings, so the key does
not reach the log.
It is deliberately outside /api/v1 and accepts the key and nothing else.
Reading a session cookie here would make it a state-changing POST outside
the CSRF check — a cross-site request that acts as whoever is signed in —
and there is a test that fails if that ever becomes true.
The API path is unchanged for anything that can send a header. The tokens
page now shows the connector address next to the token it belongs to,
since that is the one moment the token exists.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two corrections after checking against Yggdrasil's own behaviour rather
than guessing at it.
The advice to delete the failed server and make another was wrong. The
image reference is read from the rune at container-create time, and both
Start and Restart recreate the container — so once the image can be
fetched, pressing Start on the server you already have is enough.
And a rune does not pick up a changed manifest on its own. The panel
writes the rune row from a UI action only; neither a restart nor a timer
does it. So a new manifest needs Runes → verdande → Update, and the
catalogue list is cached in memory for ten minutes, which is long enough
that two attempts in a row can both hand back the old version.
The host-pull fallback now says what it costs: the panel re-pulls on
every recreate, fails quietly, and carries on with the local image, so a
new release looks like it rolled out when it did not. It is a stopgap for
keeping the package private, not a way to run this.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Installing the Rune still failed after v0.2.0 was published, with the same
message as before:
create container: Error response from daemon:
No such image: ghcr.io/kristianwind/verdande:latest
The advice in these pages — log the host's Docker daemon in — cannot
work, and Yggdrasil's source says why. Its PullImage calls the Docker
SDK's ImagePull with an empty image.PullOptions{}; RegistryAuth is never
set, and there are no registry credentials anywhere in the panel. A
private image answers 401. pullImageRetry discards that error on purpose
("second failure surfaces at create"), and ContainerCreate — which only
ever looks locally — then reports the image as missing. The message is a
liar: the tag is in the registry the whole time.
docker login does not change this. It is client-side: the daemon stores
no credentials, the CLI sends them itself with each request, and
Yggdrasil is a different client that sends none.
So the package has to be public, and the pages now say that rather than
offering a login that does nothing. Keeping it private is still possible
but is now described as what it is: putting the image on the host by hand
before creating the server and again after every release.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The suite passed on a laptop and failed on the runner with zero links
found. locator.all() is a snapshot and does not auto-wait: calling it
straight after goto() measures how fast the machine is, not what the page
contains.
An assertion on the first link waits for hydration, and the hrefs are
collected once and re-resolved on each pass rather than held as element
handles — navigating re-renders the sidebar, so a handle from before the
click points at an element that no longer exists.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The tool exists so the images can be regenerated when the interface
changes, and the interface just gained six pages that nothing in the list
would ever photograph.
The integrations page rather than the account one: it shows the calendar
feed and the mail-to-task address, which are the parts people ask what
they look like. Account settings are a password form.
The PNGs in site/screenshots are not regenerated here — that wants a
running instance with realistic data, which is a decision about
presentation rather than a build step.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Installing the Rune failed with
create container: Error response from daemon:
No such image: ghcr.io/kristianwind/verdande:latest
which was two things wearing one message. No semver tag had ever been
pushed, so release.yml had never run and the image did not exist. And now
that v0.1.0 is published, the package is private because the repository
is — and an anonymous pull reports that as "no such image" rather than as
a permission error, which sends you looking in the wrong place. The
Docker daemon on the host has to be logged in once, as whichever user
runs it. Said in rune.md, self-hosting.md, the README and the handoff,
because whichever one you are holding is the one that has to say it.
The handoff's seven gaps are down to the ones that need real credentials
and real services: Gmail against an account, Web Push against a push
service, PNG icons for iOS, the AI adapters against a key. Added in their
place are the traps this round cost the most time on — that the embedded
frontend is a copy that npm run build does not touch, that Playwright
re-runs its config in every worker, and that svelte-check is not a gate
in a codebase with no type annotations.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The brief asked for one. docs/api.md is complete and hand-written, and
there was nothing a tool could read.
Written by hand rather than generated from annotations, which would have
meant every handler carrying a second copy of itself in a comment. The
risk with a hand-written spec is that it goes stale, so it does not
depend on anybody remembering: openapi_test.go walks the chi router and
fails both on a route nothing describes and on a description with no
route behind it. CI separately checks it is a valid OpenAPI document.
The test reads the file with a forty-line scanner rather than pulling in
a YAML library. It needs two things — which paths are listed and which
methods each has — and both sit at fixed indentation in a document the
test also owns, which is worth more than a dependency in a project whose
whole shape is one static binary.
The CalDAV routes are excluded deliberately: PROPFIND and REPORT are
outside what OpenAPI can describe, and docs/caldav.md covers them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The brief asked for these under Definition of done and they were never
written. 610 Go tests, none of which opens a browser.
Six flows: create the first account and sign in, quick add and tick a
task off, click every link in the sidebar, open a task and edit it, walk
every settings tab, and check that the manifest's icons and the service
worker can actually be fetched. Every assertion is on rendered content
rather than on a URL — the binary serves the SPA shell for anything it
does not recognise, so toHaveURL passes on precisely the bug this exists
to catch.
The suite builds the frontend, copies it into cmd/verdande/webbuild and
starts the real binary, every run. Merely starting the server would test
whatever frontend was copied there last, which during development is
nearly always the wrong one, and a green report on code that is not
running is worse than no report.
Chromium only. A smoke test catches a build that does not work at all;
running it in three engines finds the same breakage three times.
They found four bugs the first time they ran, fixed in the commits before
this one: concurrent writes returning SQLITE_BUSY, a completed task
keeping its row, the same task landing in the list twice, and a database
failure presenting as everybody being logged out.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The server side has been finished for a while — VAPID keys, RFC 8291
encryption, delivery on reminders and assignments — and nothing in the
browser ever asked for a subscription. web/static/ had a manifest and an
icon and no sw.js, so a subscription could not be created at all.
The worker handles push and notificationclick and nothing else. There is
deliberately no offline cache: the app is a live view of a list other
people are editing, and a cached shell serving yesterday's data is worse
than the browser saying it cannot connect. Adding one later means
versioning and invalidating it, which is a feature rather than a few
lines here.
Unsubscribing tells the server first. If the browser unsubscribed and
then the request failed, the server would keep pushing to an endpoint
nothing is listening on, and only a 410 from the push service would ever
clean it up.
Still not verified against a live push service — this is the client half
of a chain whose other half has only ever been tested against the RFC.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The board has had it since it was written; the list has not, and
POST /tasks/{id}/move was already tested and working.
Where the board shows the target column, a list has to show the target
gap — the whole decision is "above this one or below it" — so a line is
drawn where the row would land, chosen by which half of the row the
pointer is in rather than by counting rows, which is what keeps it where
the eye expects it when rows have different heights.
The move sends both neighbours. The API's after_id is the task above,
which reads as the opposite of "after" in a vertical list, so the locals
here are named for where they end up on screen.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
TaskRow has had an onedit prop that nothing passed. There was no way to
see a task's description, sub-tasks, comments, files or reminders, all of
which the API has supported from the start.
A drawer rather than a page, so the list you came from stays behind it
and ticking a sub-task off and closing lands you where you were rather
than at the top of a re-fetched list. Mounted in the root layout and
driven by an id in the store, so a task opens the same way from Today, a
project, a label or a saved filter — and so an edit arriving over the
socket reaches the drawer without a second source of truth to keep in
step.
Text fields save on blur. A PATCH per keystroke is forty requests for one
sentence, and the last to arrive is not necessarily the last one sent.
The navigation shortcuts are suppressed while it is open: with a task in
front of you, "t" means the letter t.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
There was no /indstillinger route at all. The Gmail OAuth callback
redirects there and had been answering 404 since it was written.
Six pages rather than one long scroll, because the sections have nothing
to do with each other and somebody opening API tokens should not have to
walk past their own password to get there: the account and its second
factor, notifications, integrations, AI, tokens, and data. Everything
listed as having no interface in the handoff is reachable from one of
them — the calendar feed, the mail-to-task address, Gmail, the AI
provider, import, export, templates and the update notice.
Two decisions worth naming. The AI key field starts empty and says why:
a settings page that repopulates a password field is one that will
eventually leak a key into a screenshot, and an empty value means "leave
the stored one alone" rather than "clear it". And the form chrome lives
in the layout as :global rules, because scoped styles do not reach a
child route and six copies of the same twelve selectors is six places for
them to drift.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Ticking a task off never removed its row. The optimistic update ran, the
request went out, the server answered 200 — and the view did not move.
The same applied to a change arriving over the websocket, which is how
somebody else's edit was supposed to show up. It had never worked.
The common factor was tasks[i] = value. push and whole-array assignment
both reached the views; index assignment did not. Rather than depend on
which mutations the reactivity happens to see through, every write now
goes through app.replace() or app.upsert(), which build a new array. The
lists are at most a few hundred rows and the copy is not worth a moment's
thought next to the class of bug it removes.
upsert exists because the same task arrives twice: once as the response
to the request that created it, and once over the socket, which publishes
to the whole project including whoever did it. Either can land first, and
appending both puts two rows with one id into a keyed {#each} — not a
duplicated row but a thrown error that stops the view rendering.
complete, reopen and update were three copies of the same twenty lines;
they are now one private helper.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
docs/api.md has told people to make a token under Settings since it was
written. The store has held CreateAPIToken, ListAPITokens and
DeleteAPIToken since the first migration. There were no routes: nothing
between the two, and no way to make one.
GET, POST and DELETE under /tokens, all behind a new requireSession
middleware. A bearer token is refused there with 403 even though it is
accepted everywhere else — otherwise a leaked token is permanent, because
whoever has it issues a second one and revoking the first changes
nothing. The plaintext comes back from POST and never again; only the
hash is stored, so the interface has to say so at that moment rather than
offering a reveal that cannot work.
PATCH /auth/me alongside it. Name, timezone and language could not be
changed anywhere — not through the API and not through the interface —
and an account created from an invite inherits a default timezone. That
is not cosmetic: every date in the app is resolved in it, so somebody in
another country had every due date land an hour out with no way to say
so. The zone is validated against the running system's database rather
than a list kept here, which is the same database the dates are resolved
with. The email address stays read-only: it identifies the account and is
what invites were sent to.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
authenticate() cannot tell "no such session" from "the session lookup
failed" on its own — both arrive as an error. Answering 401 for both
meant a disk or file-permission problem presented as every session
ending at once: the frontend clears its state and shows the sign-in
screen to everybody, and the logs record a wave of failed logins rather
than the fault.
Now only a missing or unknown credential is 401. Anything else is a 500
and is logged as what it is.
Found while writing the smoke tests, where a deleted database file
produced SQLITE_CANTOPEN on the connections that had not been opened
yet — so most requests succeeded and a few reported the user as signed
out, which reads exactly like a flaky auth bug.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two people saving at the same moment got a 500 after two milliseconds.
The DSN has set busy_timeout since the first commit, and it covered none
of the contention it looks like it covers. With a plain BEGIN a write
transaction starts as a reader and asks for the write lock at its first
UPDATE — and SQLite refuses that upgrade with SQLITE_BUSY immediately,
without applying busy_timeout to it. Backing off cannot help: the
transaction is already holding a read snapshot the other writer is
invalidating.
_txlock=immediate takes the lock at BEGIN instead, where the timeout does
apply and the second writer waits its turn. The driver leaves read-only
transactions deferred, so readers still run concurrently, and every one
of the twenty callers of db.Tx is a write.
Found by opening a task and saving a description, a sub-task and a
comment at once — three writes, one of which came back a 500. The test
fails without the flag.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
