Wires the window-awareness modules to the product for the first time.
uia.rs and capture.rs were fully written and had no Tauri command at all;
their only caller anywhere was an #[ignore]d smoke test, so nothing in a
shipped build could reach them.
Six commands, all behind the ADR-0008 policy: the privacy switch, the
session list, start and stop, a window frame, and the window's UI
Automation text. The read takes a closure so the window handle never
leaves the policy module — there is no accessor that skips the switch, the
session lookup, the expiry check, or the rate limit.
Starting a session has an interaction problem worth naming: reading the
foreground window is the only selection mechanism these modules have, and
at the moment the user clicks the button, the foreground window is this
application. So the button gives them a few bounded seconds to bring the
target forward, and a window belonging to this process is refused rather
than captured — which is the most likely outcome if the delay elapses
while they are still on the settings page.
The panel sends either the window's text or its picture to the companion
slot. Text first in the UI order, deliberately: it is far less revealing
than a screenshot and usually enough to answer "what am I looking at".
The indicator in the shell is ADR-0008 item 5, which no project document
required. Reading someone's screen with no visible sign is
indistinguishable from spyware from the outside, and a stop control the
user has to go looking for is not a stop control. It polls rather than
waiting to be told, because an authorization can lapse on its own — and
when it cannot read the status it renders nothing, claiming neither that a
capture is live nor that one is not.
The app's own commands need no capability entries: only plugin commands go
through the ACL, which the existing commands working under `core:default`
alone already demonstrates.
Rust tests 110 -> 113, frontend 43 -> 47.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
F-33. A voice reply produced nothing at all: no marker, no event, no
error. Four independent gates each dropped it, and fixing any one alone
would have changed nothing observable.
The mapper reached `map_media` for image, record, file and video alike,
but its only emitting branch was gated on the frame being an image, and
the `attachment_saved` fallback rejected every other type outright. Both
now carry whatever type arrived, with the frame type as the marker's alt
text so the renderer knows what to expect before any bytes exist.
`pending_media` was also a single slot, so a second attachment in one
reply overwrote the first before it was paired; it is a bounded queue
now, and an `attachment_saved` with no id no longer consumes the media
frame still waiting for one.
The download judged a candidate path successful only if the bytes sniffed
as an image — this server answers unknown paths with its own text under
HTTP 200, so sniffing is load-bearing and cannot simply be dropped.
Audio and video signatures are recognised now, and a declared
Content-Type is trusted when it names media rather than only images.
SILK is included deliberately: it is what AstrBot relays for QQ and
WeChat voice, no browser decodes it, and recognising it is what lets the
UI say so instead of rendering a player whose play button does nothing.
The CSP had no `media-src`, so it fell back to `default-src 'self'` and
an `<audio src="blob:">` was refused — in release builds only, since the
dev server serves no CSP at all. That is the same divergence as F-04 and
the inline-script trap in pet.html, so the policy now has a test of its
own that pins each directive the app depends on, including that
script-src still refuses inline.
The renderer chose `<img>` unconditionally. The element now follows the
media type of the bytes that actually arrived, falling back to the
marker's declared kind for the placeholder and the failure message.
`map_media` also kept a private copy of the message-id handling that
included the reopen-suffix overwrite fixed earlier in `adopt_message_id`;
it shares the one implementation now.
Protocol tests 22 -> 26, Rust client 99 -> 101, frontend 35 -> 43. All
six mutations of the new assertions were confirmed to fail.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Repairs two commits that I pushed with a failing gate.
64dab45 treated every `streaming: false` plain frame as a new segment.
That is wrong half the time: AstrBot sends both a block that restates
what the delta stream already carried and a block carrying text the
deltas never delivered, sometimes in the same reply — which is exactly
what the frame-trace fixtures in astrbot_client.rs were built to tell
apart, and I had written those fixtures myself. Treating a restatement as
new content would have shown it twice, and completing each block on
arrival left the two R2 sentinels with no open message to finalize, so a
truncated body and a cancellation both reported "stop".
Both the non-streaming plain frames and the complete frame now go through
one `reconcile`: emit only the byte-exact tail beyond what was shown, or
nothing if the text has already been shown allowing for different
whitespace, or open a message of its own if it diverges. Fixes a real
defect this exposed in `adopt_message_id`, which assigned the server's
bare id back over an active id that carried a reopen suffix, merging
blocks that had just been separated.
9b2ec64 added petEntry.test.ts importing node:fs. tsconfig.app.json sets
an explicit `types` array so Node globals are absent, and typecheck and
build both failed. Now imported as raw text through vite, rather than
adding Node's types to the compilation that covers the browser code.
Protocol tests 12 -> 19. Every new assertion was checked against a
deliberate regression, and the two that survived their first check —
dropping the byte-exact prefix case and dropping the whitespace-only
guard — now have tests that fail without them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
An audit of the destroy→hide change across four lenses raised 17
findings; 14 survived adversarial verification. Two of them were
regressions that change introduced, and both had the same shape: code
that only worked because the pet window used to not exist.
show_pet_context_menu picked the pet window and fell back to main.
While the pet rendered inside the app its window had been destroyed at
startup, so the fallback happened to select the visible main window.
Now that the window merely hides, the menu was being anchored to a
hidden HWND, which Win32 cannot bring to the foreground. The
visibility test is explicit now rather than incidental.
DesktopPetView read the connection state in a mount effect and nothing
emitted connection events, so destroy-and-rebuild was what re-read it.
Without a remount the floating pet showed whatever was true at startup
for the rest of the session. reveal_pet_window now emits
pet-window-revealed and the view re-reads on it.
The rest were already there. Every pet setting was decorative: the
preferences were applied and then dropped, so reveal_pet_window had to
invent an answer and hard-coded the first monitor, the safe edge and
always-on-top — reverting the user's choices on every show. The pet view
re-sent a fully hard-coded payload whenever game mode toggled, and the
settings page initialised its controls from literals, so returning to
the page and changing one thing pushed three stale defaults back onto
the window. There is a session-scoped store now, a command to read it,
and both frontends read before they write. Persisting it across restarts
is F-75.
Always-on-top is off by default, which is what the Q6 acceptance
criterion asked for (F-35), and the window no longer takes focus when
shown — it was activating on show and swallowing keystrokes meant for
the main window. focusable is left alone on purpose: clearing it adds
WS_EX_NOACTIVATE, which would re-break the context menu now anchored to
that window. That is F-79.
The pet also painted a solid dark rectangle before its assets arrived.
The transparency rule is gated on a class that an effect inside the
5.2 MB lazy chunk added, so until that chunk ran the app's opaque
background won. This is why the earlier hunt for "the rule that beats
!important" found nothing: the class was not applied yet, so specificity
was never the question. The pet now has its own entry document carrying
the class statically. An inline script would not do — tauri-utils only
nonces script[src^='http'], so it would pass in dev and be blocked by
script-src 'self' in a release build.
Rust tests 92 → 97, frontend 30 → 35. All 11 mutations of the new
assertions were confirmed to fail.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The floating pet still does not appear, and the two explanations offered
so far are not sufficient. The rebuild path genuinely never called
show(), which is fixed and correct on its own reading of the code, but it
did not resolve the symptom. A missing Suspense boundary around the lazy
pet view looked like a second cause and is not one: removing the wrapper
still renders the pet, so React 19 tolerates it. That claim is retracted
in the comment rather than left standing.
The window is transparent and undecorated, so "did not appear" covers a
window placed off-screen, a window that reports hidden, and a window
present with nothing painted in it. Log the monitor work area, the
position and size after placement, and the visibility the window
reports, at default level so no filter is needed to read it.
The Suspense wrapper stays. Relying on React tolerating a lazy component
with no boundary is not a contract, and the bubble and quick views
already have their own.
Refs: N-11, F-61.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Switching the pet to the floating desktop mode made it vanish.
close_pet_window destroys the window rather than hiding it, and the app
closes it on startup whenever the pet renders inside the UI, so opening
the floating pet always went through the rebuild path — which honours
the config's `visible: false` and never showed what it had just built.
Only the already-exists path called show() and placed the window. Both
paths now share one helper.
The in-app pet also sat on an opaque card: .app-pet-overlay
.pet-window-embedded painted a 92% gradient with a border and a drop
shadow over a base class that is transparent. Verified against the real
stylesheet — the backdrop now computes to no image, no colour, no
border and no shadow, with the drag box unchanged at 188x226.
Also record the state of voice chat, window awareness and game
companionship, all three of which are in the declared product scope but
largely unwired: the voice path uses a Web Speech API that WebView2 does
not provide and points away from the architecture's server-side STT
design, the window capture and UI automation modules exist in Rust with
no command registered and no route to the companion, and game mode
toggles a window preference and produces no data at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The chat panel opened with three stacked blocks above the messages: a
title and subtitle naming the panel, a strip repeating the profile id,
session id and connection state, and the session pills. Only the last
carried anything the reader needed.
Move the slot tabs into the window title bar and delete the rest. The
tabs go there through a portal rather than by lifting state into the
shell, which keeps the slot selection and its status polling in the
panel that owns them; the quick chat and bubble windows have no title
bar, and they are slot-fixed already, so the tabs simply do not render
there.
The connection actions move onto the session row, which costs no height
because that row already exists. 刷新 is dropped: the status refreshes on
a 500ms poll, so the button did nothing the panel was not already doing.
重连, 断开 and 新对话 stay, since losing them would leave no way to
recover a dropped connection or start a conversation.
The panel now puts 82px above the messages instead of three blocks. The
title-bar rules carry [role="tab"] on purpose — the panel's own
.chat-slot[role=tab] rule sets a 2.75rem minimum and sits later in the
sheet, so an unqualified selector loses to it and the tabs overflow a
44px bar.
Verified against the real stylesheet: tabs 33px inside a 44px bar,
vertically centred, clear of both the title text and the window
controls; the removed blocks are absent from the DOM.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The bubble header repeated what the avatar and the side of the bubble
already say. Removing it leaves the header holding only the generating
indicator, so it is rendered only while a reply is streaming.
The name moves into the avatar as a clipped label rather than
disappearing: the avatar's glyph is aria-hidden, so without it a screen
reader would announce nothing at all for the speaker.
Verified against the real stylesheet: avatar and bubble tops differ by
0px for both roles, the glyph stays centred in a 32px avatar, the label
measures 1x1, and the bubble no longer contains the name.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
An attachment downloaded and then rendered as a broken image. A blob URL
is decoded according to the blob's type, so handing the webview
application/octet-stream makes img refuse bytes that are a perfectly
good PNG — and octet-stream is exactly what the client substitutes when
the server names no type.
Prefer a concrete image type from the server, otherwise identify the
format from its leading bytes. Every signature checked is a fixed marker
at a fixed offset, so this reads no user content, and a JSON error body
served with HTTP 200 is deliberately not mistaken for an image.
Whether that was the cause is not yet established, so the alternative is
made self-diagnosing rather than guessed at. The download outcome —
status, the type the server gave, the type the bytes indicate, and the
length — is now traced, and the frontend reports why an attachment did
not render instead of leaving a bare broken-image icon: a failed
download shows its error, and a payload the browser cannot decode says
so. Both paths were previously swallowed.
Refs: N-6.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The message list had no scroll code at all. Opening a conversation
landed at the top of the history, and every message sent or received
appeared below the fold, so reading a reply meant scrolling down by
hand each time.
Scroll to the tail when the rendered messages change and when the slot
or session changes. Following stops as soon as the reader scrolls more
than 80px away from the bottom and resumes when they return, so reading
back through history is not interrupted by an arriving delta.
scrollTo carries the smooth behaviour but is not universally available —
jsdom has none — so it falls back to assigning scrollTop.
Refs: F-44 (audit P2-28).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Adding an AstrBot connection asked for eleven fields. Three of them
reach nothing: the SSE URL has no consumer anywhere in networking (the
desktop transport builds its own endpoint and AstrBot never reads it),
the certificate fingerprint is stored but TLS pinning is not
implemented, and the role only decides a word in the profile list.
The form is now the four things an activation actually requires — server
address, adapter id, API key, slot — with everything else behind a
collapsed advanced section.
The adapter id is the remote username. It was previously entered twice,
once in the profile form and again in the credential card, and it is
what keeps two slots on one server from sharing a conversation, so the
field now says that.
Three values are derived rather than asked for. The display name falls
back to identity@host. The role follows the chosen slot. The insecure
HTTP approval appears only for an http:// address and is required there,
because normalize_url rejects such an address without it — showing it
unconditionally invited enabling it for an https:// endpoint that never
needed it.
Choosing a slot now does the whole job: it assigns the slot, sets the
role, and points startup auto-connect at it, replacing a separate slot
card interaction and a separate auto-connect slot dropdown that could
disagree with each other. The API key is enrolled in the same submit.
Both steps need the profile to exist, so they run after the save and
report their own failures rather than failing the save.
Verified in a browser against the real stylesheet: four fields plus a
collapsed advanced section, the insecure checkbox appears only for
http:// and carries required, and the display name placeholder resolves
to identity@host.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The file held 603 CRLF and 85 LF-only lines, while .editorconfig and
prettier both pin LF. Any substantive edit therefore produced a
whole-file diff that buried the actual change, which is why the fix
queue carries a standing warning about this file.
Isolated on purpose: `git diff --ignore-cr-at-eol` on this commit is
empty, so it can be reviewed as whitespace and the form rework that
follows reads as logic alone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The minimize, maximize and close buttons did nothing. They were wired
correctly; the ACL rejected them. core:default resolves to
core🪟default, which grants only read-only queries — is-maximized
is in it, which is why the maximize icon tracked state correctly, but
minimize, toggle-maximize, close and start-dragging are not. Both
windows are decorations: false, so with close denied the main window
could not be closed at all except through the tray.
Add two capabilities rather than widening the shared one: the main
window gets the four controls it draws, and the pet window gets dragging
alone, since it is moved by dragging its body and must not be able to
close or resize itself.
Every one of these calls was a bare void expression, so the rejection
was swallowed and looked identical to an unwired button. Route them
through a helper that reports the failure in the shell error banner.
This is the second time a missing capability has surfaced as a silent
no-op — the pet window previously had no entry at all. Add tests that
read the shipped capability files and assert each permission the UI
calls is granted to the window that calls it.
Also translate the TypeError the Tauri shim throws outside the
application into a sentence a user can act on. The dev server is
reachable at 127.0.0.1:1420, so opening it in a browser and hitting save
produced "Cannot read properties of undefined (reading 'invoke')".
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Right-clicking the pet showed the native WebView menu and never the
application's own menu. The handler was a bare void invoke(...) with no
catch, so any failure to build or pop the menu was completely invisible
and indistinguishable from preventDefault not taking effect.
Report the error in the pet window and stop propagation as well as the
default action. This does not yet explain which of the two is happening
— it makes the difference observable, which is the prerequisite for
fixing the right one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three defects reported from real-machine testing, all rooted in the same
few conflicting CSS blocks, so they are fixed together — separating them
would leave intermediate states that render worse than either end.
The action buttons showed wrapped vertical text inside a 2rem box.
styles.css already defined .message-action-label, a clip-rect
screen-reader-only class clearly written for these buttons, but nothing
in the tree ever used it: ChatPanel rendered a bare span. Wire the class
up so the buttons are icon-only, and drop the border so they read as
plain icons rather than pills.
.chat-message-actions and its buttons were defined five times across the
stylesheet, three of those styling them as bordered pills that the final
block then partially overrode. Remove .chat-message-actions button from
the three pill rule groups and the small-screen grid rule, leaving one
definition next to .message-action-icon. .chat-runtime-actions and
.chat-stop-button keep the pill treatment; they are different controls.
The avatar rendered below its bubble instead of beside it.
.chat-message-row used align-items: flex-end, and .chat-message-content
is taller than the bubble because the action bar reserves a row, so the
avatar was pinned to the bottom of that taller box. Align to the top.
Removing visibility: hidden from the hidden action bar also closes F-40
(audit P2-24): visibility removed the buttons from the tab order, so
:focus-within could never fire and a keyboard user could never reveal
them. Opacity alone keeps them focusable.
Verified against the real stylesheet in a browser: avatar and bubble
tops differ by 0px, the button computes to 34x34 with 0px border, the
label measures 1x1, and focusing a button takes the bar from opacity 0
to 1.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The composer draft, the pending attachment list and the error banner
were single values shared by both slots. Uploading on companion and
then switching to operations routed the companion instance's
attachment ids into the operations instance, the operations error
banner stayed visible on the companion tab, and a half-typed message
followed the user across tabs.
Make all three Record<ChatSlot, ...>, mirroring the sending,
unreadCounts and storedSessions state that was already per slot. Every
asynchronous action now pins its slot at the start, so an upload or a
send that resolves after the user switched tabs still lands on the slot
it belonged to.
Splitting the error state is required rather than incidental. refresh()
runs every 500ms and wrote the shared error on failure and cleared it on
success, so any send or upload error was silently wiped within half a
second. The polling failure describes the IPC channel rather than a
slot, so it keeps its own runtimeError; slotError is owned solely by
user actions and takes precedence when both are present.
All three new tests were checked against a mutation that reads the
companion slot regardless of selection, and all three fail under it.
Refs: F-16 (audit P1-11).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
All four listen() call sites assigned the resolved unregister function
unconditionally. When the effect was cleaned up before the promise
settled, that assignment landed on a dead closure and the listener was
never unregistered. React StrictMode mounts, cleans up and mounts again,
so the first listener leaked on every development mount. In ChatPanel
that meant applyRuntimeEvents ran twice per event and, because a delta
is appended to the existing content, every character of a streamed reply
was doubled.
Track whether the effect is still live and unregister immediately when
the promise settles late. Each site also gains a catch, since these were
all bare void listen(...).then(...) chains — a rejected listen, such as
the missing desktop-pet capability, became an unhandled rejection rather
than anything observable.
The listener mock in src/test/setup.ts resolved eagerly, which made the
race impossible to express. Replace it with a controllable double in
src/test/tauriEvent.ts that defers resolution until the test asks for
it, tracks each registration separately, and only dispatches to handlers
that have not been unregistered. The registry is plain module state, so
setup.ts resets it after each test.
ChatPanel had no test file; this adds one. Both new tests were checked
against a mutation restoring the old assignment, and both fail under it.
Refs: F-15 (audit P1-10).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Commit 41592e2 shortened the localized fallback string but left the
ternary wrapped across three lines, which prettier now collapses. This
file passed `prettier --check` at 785f125 and failed afterwards, so it
is the only formatting regression introduced by the R1/R2 batch.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Historical UTF-8 corruption left question-mark runs in user-visible copy and in
one accessible name: the sprite load failure message, the canvas context
failure message, the pet window's `aria-label`, and the page loading state.
A separate spot had the "desktop floating" option label hard-coded as numeric
HTML entities.
Refs P1-17 / F-09.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Rust serializes `Option<String>` without `skip_serializing_if`, so absent
fields arrive as `null`. Zod's `.optional()` accepts only `undefined`, so
`safeParse` failed and three chains silently died:
- `boundRuntimeEventSchema.sessionId/messageId`: desktop-v1 and mock events
carry no session id, so every one of them was discarded; AstrBot error and
heartbeat frames were discarded too.
- `remoteChatSessionSchema.displayName/updatedAt`: one null field threw for
the whole array, leaving the remote session list permanently empty.
- `astrbotModelOptionSchema.name`: a null name failed capability parsing,
which permanently disabled the file upload button.
The sibling `slotRuntimeSnapshotSchema` in the same file already used
`.nullish()` throughout, which is what makes this look like a miss during the
de-enveloping change rather than a deliberate choice.
`formatSessionTime` widens with it: `updatedAt` is now `string | null` and the
existing falsy guard already handled null.
Refs P1-7a / F-06.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Tauri rejects `invoke` with the deserialized Rust error value, which for
`RuntimeCommandError { code, message }` is a plain object rather than an
`Error`. Nine call sites used `reason instanceof Error ? reason.message :
String(reason)`, so every send, reconnect, stop, and settings failure showed
the user `[object Object]`.
`ProfileManager` already had a correct implementation; this lifts it into
`src/lib/errorMessage.ts` and routes the other call sites through it. The
module also exports `isCancelledError`, which a later batch needs to tell a
deliberate stop apart from a real failure.
Refs P1-16 / F-08.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Assistant messages come from a remote AstrBot instance and are untrusted, but
the markdown renderer passed the href through verbatim, so `javascript:` and
`data:` URLs reached an anchor. Inside this webview, script execution has
access to the full Tauri IPC surface.
Links are now resolved with `new URL()` and only http/https/mailto survive;
anything else renders as the original plain text. Relative hrefs are rejected
too rather than being presented as absolute links against the placeholder base.
`rel` gains `noopener`.
The helper lives in `src/lib/` rather than beside the component because a
non-component export in a component module fails `react-refresh` under the
repository's `--max-warnings 0` lint gate.
Refs P1-15 / F-05.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The manifest filename pattern used `\.` inside a regex literal, which matches
a literal backslash followed by any character, while the surrounding character
class already excludes backslashes. Every valid pack was rejected, including
the one bundled in this repository, so pack import was entirely unusable.
Adds a test that parses the bundled manifest so the pattern cannot regress.
Refs P1-12 / F-02.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Add configuration-only auto-connect rules, runtime operations slot lock
while tasks are active, and duplicate server-instance warnings without
silent Profile merge. Surface policy status to the settings UI and cover
the new boundaries with Rust and Vitest checks.