README: rewritten to the brand voice, with the new logo, Pilot as the
recommended install and a Frappe Cloud button. The old PNG logo is gone.
iPad: the primary rail now pads the status bar inset itself, since it
sits beside <main> and never got it. The bridge view controller overrides
supportedInterfaceOrientations because Capacitor builds its mask from the
iPhone list in Info.plist only, which pinned the iPad to portrait.
Android release: a signing config read from a gitignored
keystore.properties, and scripts/playstore.sh, which builds the signed
bundle the same way testflight.sh archives for iOS. Both are wired to
yarn native:playstore.
Build number 314 for both platforms.
Store screenshot tooling and the demo site seed are ignored; they are
local tooling, not part of the app.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The chosen site users were filtered afresh each render, so the preview
effect fired every time, set state in the page, and re-rendered the step
again. Memoized on the list and the selection.
Giving someone Raven access saves their User, which needs System
Manager. Setup admits Raven Admins too, and everyone starts ticked, so
their first Add raised for the whole batch, including people already in
Raven who only needed a membership. People still needing access are now
offered but not addable to a caller who can't write Users, and the row
says so.
Install no longer adds every system user it can and skips the rest.
Raven ships Raven User and Raven Admin role profiles, and the team step
opens a searchable picker, with Select all, over everyone on the site;
a single person gets a checkbox. One helper now gives Raven access
everywhere: a user whose roles come from a role profile gets Raven's
profile alongside, which the Framework keeps across saves; on Frappe 15,
where a user has one profile, they're shown but not addable. Someone
whose Raven access was taken away is offered again.
The setup funnel is reported as the Framework wizard reports its own:
each step completed or skipped with the time spent, going back, starting
and leaving. The web app posts them to one endpoint that passes known
events and properties to the Framework's capture.
A public photo on the User is used as it is on the Raven User; the old
copy fetched the file from the site itself, which fails on a dev server
and was swallowed. Setup also falls back to the User's photo.
The HR step's copy is rewritten.
Found while walking a new site through setup:
- The DM sidebar showed only "No conversations yet" when there were no
DMs. The start-a-conversation rows were the virtual list's footer, so
they never rendered for an empty list. They now show in that case too;
the empty state is kept for when there's no one to suggest.
- The command menu's "Users" heading rendered over nothing on a site with
no DMs, as cmdk only hides an empty group while searching.
- "Select a conversation" now sits in the page's island, where the chat
would be, instead of floating on the canvas.
- Icons on "Select a conversation" and "No messages yet".
- The channel sidebar's empty state is sized for the narrow column, held
near the top, and offers a Create a channel button.
After the last step the form gives way to a check that draws itself and
"All done", which fades out and is replaced by "Now, send a raven." in
Newsreader, the rest of the line following the "Now,". Both lines share
one slot so nothing shifts. The preview then grows into the app as
before. The finale timers now run through one helper.
The app's slim boot carries setup_complete only under sysdefaults, so
the gate sent pending admins to the Desk wizard on sites already set up.
The slim boot also gains is_fc_site, which the gate and the password
step read.
The app serves every site from one origin, so the done flag, the saved
progress and the boot-cache update were shared across sites. All three
now use the site-scoped key. The workspace step also prefills the company
in the fetch's onSuccess instead of an effect.
Frappe Cloud opens a new site on /raven before the Framework wizard. The
gate now lets those sites into Raven's flow, the profile step asks the
first admin for a password since no wizard did, and finishing Raven's
setup completes the Framework wizard with the region Frappe Cloud stored.
Sites with ERPNext or HRMS still go through the real wizard first.
The web app, the native bundle and the store versions move to 3.1.0.
A 3.1.0 site asks apps older than that to update, and the app refuses
sites below 3.1.0 in the picker: the native boot endpoint and the CORS
hook only ship from there, so an older site could not open anyway.
Every share wrote a new folder into the app group and nothing removed
it. The page now deletes each copy once its bytes are in memory, and the
extension clears the previous share before writing a new one, as the
Android shell already does. Only the latest share is ever readable, so
nothing reachable is lost.
The share extension loaded each file-backed item into Data before
writing it to the app group, which breached the extension's memory
limit on large videos and ended the share. The copy now streams through
the file manager; Data is kept only for items that exist nowhere on disk.
The share page stashed files globally as soon as it loaded, so backing
out left them for the next composer to attach in any conversation. The
page now keeps the files until a conversation is picked and stashes them
for that channel alone; a composer takes only files addressed to it.
The rotation listener was only installed when push was already on at
startup, so a session that enabled push for the first time kept a dead
token on the server until the next restart. One watcher now serves both
paths, and it only syncs while push is on for the site, so a rotation
after turning push off does not subscribe the device again.
* fix: do not allow sending messages to disabled users
* fix: message action toolbar position in left-right mode
* fix: show timestamp for continuation messages
* fix: do not enforce max length on password fields
* fix: Raven logo sizing and welcome text
* fix: limit raven document notifications to stop at cacading error logs
* fix: do not leave channel references in notifications when channel is deleted
* fix: handle raven user deletion by removing role
* fix: set dm_user_1 and 2 to None explicitly
* fix: in-desk integration
* fix: allow Cyrillic characters in channel name
* fix: cap width to container
* fix: enable user via roles
* fix(channel): say "left" when the only admin removes themselves
When the only admin left a channel or thread, the admin handover branch
always wrote "X was removed by Y", with Y set to the current user. A
member who removed themselves read as "Diptanil was removed by Diptanil
and Nabin Hait is the new admin of this channel."
The handover branch now checks whether the member removed themselves,
the same way the branch without a handover already does. A self-removal
writes "X left and Z is the new admin of this channel." with a
`user_left` event that carries `new_admin`. Removal by someone else is
unchanged. The web client renders the new sentence from the event, and
old messages keep their `member_removed` event and render as before.
Adds the first tests for Raven Channel Member: leaving as the only
admin, and being removed as the only admin.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* test: make two backend tests pass on current Frappe
Neither test is about this branch; both failed on develop's CI.
test_falls_back_to_utils_on_older_frappe deleted render_safe_globals
from frappe.utils.safe_exec to simulate an older Frappe. On current
Frappe, get_safe_globals is an alias of exec_safe_globals, which calls
render_safe_globals by name, so the fallback it was meant to exercise
raised NameError. The test now stands in a copy of the globals for
get_safe_globals before taking the name away, which is what an older
Frappe actually looks like.
test_failed_dispatch_marks_failed_and_continues scheduled a failing row
first. The sweep rolls back after a failed row, and that rollback also
discarded the test's uncommitted fixtures, so the next row no longer
existed. A row that sends now goes first: its commit saves the fixtures,
and the test still checks that the sweep continues past the failure.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* test: drop the older-Frappe fallback test for document notifications
The test deleted render_safe_globals from frappe.utils.safe_exec to
stand in for an older Frappe. On current Frappe, get_safe_globals is an
alias that calls render_safe_globals, so the simulation could not match
a real older version and the test failed on develop's CI. The fallback
in get_read_only_frappe stays; it is only reached on Frappe versions
that still have their own get_safe_globals.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* test: check ownership of scheduled rows by name, not by count
The sweep test now commits its rows, and the Failed one it leaves
behind is listed for its owner, which is correct. Counting rows made
the "only own" test depend on what ran before it. It now checks that
its own row is listed for its owner and that another user sees none.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
A Raven Admin could change the push server URL to their own server.
From then on, push requests, token syncs and the daily job would send
the API secret, every user's push token and notification text there.
Move the push server URL, API key and API secret to permlevel 1,
which only System Managers have. The Push Notifications panel shows
these fields to System Managers only. Raven Admins keep the Sync
button, which uses the saved server.
Also drop the AI settings test that only re-checked field permlevels.
test_llm_configuration sent the stored OpenAI Compatible key to any
URL the caller passed, and returned the response from any URL. Raven
Admins could call it, and could also save a new LLM URL, so they
could capture a key a System Manager set or probe internal services.
Move the OpenAI and local LLM provider fields (URL, provider, keys,
org and project) to permlevel 1, which only System Managers have.
For everyone else, test_llm_configuration tests the saved server and
ignores the URL passed in.
The AI settings panel shows these fields to System Managers only.
Agent model listing now loads whenever local LLM is enabled.
- firebase 10.9.0 -> 12.19.0: clears the protobufjs, websocket-driver,
undici and busboy alerts. The messaging APIs Raven uses are unchanged.
- vitest 3 -> 4.1.11: clears the tinypool alert.
- Refresh yarn.lock within existing ranges for the other vulnerable
packages (tiptap, prosemirror-view, react-router, vite, axios, turbo
and others).
- engines.node >=22.12.0 in the root package.json. Frappe Cloud checks
it against the bench's Node version before a build. Vite 8 already
needs 20.19+ or 22.12+.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>