Applying an update downloads and swaps a binary and takes minutes; the
doctor showed a spinner inside a disabled button for the whole time.
SystemActions gained a streaming CLI runner that hands each line to the
caller as it arrives, with CHAIN_PLAIN=1 set for the child so the CLI
emits one line per transition instead of its redraw-in-place block.
The update card now shows an indeterminate bar labelled with the
running step and an elapsed counter on its own timer, so the counter
keeps moving between lines that can be minutes apart. No invented
percentage: the CLI reports steps, not a measurable total.
The streaming path still defers to debugRunFaiOverride, so tests stay
hermetic and never spawn a process; three guards pin that.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Both surfaces showed an indeterminate spinner for the whole install,
because the hub only offered a unary call. They now follow
HubAdmin/InstallModuleStream:
- The store's install dialog shows the phase in plain language, the
overall percentage, megabytes while the size is known, and a seconds
counter. The counter is the second, independent signal: it keeps
running even if the hub goes quiet, which is what tells 'slow' apart
from 'stuck'.
- The setup wizard shows the same bar per module instead of a spinner
inside the button.
- New strings in both locales for the phases, the byte line and the
elapsed counter.
progress_bar_visual_test writes the four waiting states to
build/progress/ in either theme, so the result can be judged without
launching the desktop app. Writing it turned up a real trap worth
recording: awaiting toImage() directly in a widget test leaves the
binding waiting forever - the file passed in six seconds, then sat
there until the ten-minute timeout failed the whole suite.
tester.runAsync() is the fix. Full suite: 186 passed.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
A waiting user asks two questions that a plain LinearProgressIndicator
answers as one: how far along is it, and is anything still happening.
The widget keeps them apart - the value may stand still for minutes
while a highlight driven by 'busy' keeps crossing the bar, so a
stalled percentage no longer reads as a frozen app.
Also: glides to new values over 600 ms so the polling interval behind
it stays invisible, refuses to walk backwards, renders indeterminate
rather than an empty bar when no value is known, stops animating once
the work ends, and honours reduce-motion.
Eight guards in test/progress_bar_test.dart cover each of those rules.
Writing them found a real defect: with busy: false the late-final
controller was first constructed inside dispose(), where the ticker's
context lookup is already unsafe.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The switcher used to be embedded per page (Flows/Runs/Audit/
Approvals) — invisible on the other five pages and sitting in a
different corner depending on the page (persona review 2026-08-27,
consensus finding). It now lives ONCE in the sidebar, above the
destinations: active project/area always visible, opens the same
menu everywhere, Cmd+P from anywhere. The shell listens to the
workspace, so the sidebar endpoint label can no longer lag a
sealed switch until the next health tick.
Also in this rebuild:
* Stopped sealed areas ask before starting ("Start area X?") —
a context switch must never boot a hub daemon as a click
side-effect; running areas keep switching with one click.
* The switcher tooltip told a wrong scope ("filters this view") —
it now says the choice applies everywhere and stamps new runs.
* The aggregated sealed row explains itself in place (names can
reveal client identities) and links to the Settings toggle
(Settings dialog gained an initialCategory jump).
* The active entry carries a checkmark in the menu.
* The Cmd+K palette knows projects and areas, ranked by recent
use; sealed names honour the privacy setting — while hidden,
the palette offers the guarded picker instead of the names.
* The runs empty state names the active project filter as the
cause ("No runs in project X" + show-all action) instead of
claiming the feature is off.
Tests updated to the anchor and made hermetic (scriptable
projects on the fake hub, sealed-area fake); new coverage for the
checkmark, the why-line, and the start confirmation.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Two findings from the workspace persona review, both rated high:
* Switch race: the hub client re-pointed at a sealed area's hub
before the workspace announced the sealed context, so the 2 s
page pollers (runs, audit) could fetch and render that hub's
data without the sealed marking. Switches now run inside an
explicit switching window: opened before anything touches the
connection, announced optimistically in the identity bar
("switching…" + spinner, leave button hidden), pollers and the
shell health tick pause inside it, and pages drop replies whose
context epoch changed mid-flight. The sealed context is
announced only after the new hub answered healthy.
* Filter loss: entering a sealed area cleared the shared-hub
project filter and returning restored only the endpoint. The
filter is now parked on entry and restored on return; prefs
keep the parked value throughout, so live state and prefs agree
after the round trip (and after a mid-session relaunch).
Guard: workspace_switch_race_test pins both invariants
state-matrix-style against scripted hub + sealed-area fakes —
reconnects may only happen inside an open switch window, pollers
must stay silent inside it, and the filter must survive the round
trip. SealedAreaService gained a debugSetInstance seam so the
suite never scans a real ~/.chain.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The hub copies the reviewer string a client sends straight into
decided_by (DecideApproval, ClearEventLog); nothing on the wire ties
it to the authenticated caller. Studio filled it from the OS account,
so an export read like non-repudiation while being an arbitrary
client claim — the legal finding of the 2026-07-26 usertest panel.
The real fix is hub-side (derive decided_by from CALLER_IDENTITY);
that contract is written down in docs/reviewer-identity.md and needs
a hub release. Until then Studio does the one thing it can do
honestly and marks its own claim as a claim, inside the record:
- data/reviewer_identity.dart is the single place that produces and
reads the value; wire() is idempotent, so page and HubService may
both normalise. Every write path funnels through HubService, so no
surface can send a bare handle.
- The inbox states before the decision who will be recorded, what
that attribution is worth on this hub (from AuthStatus), and the
literal string that lands in decided_by. An unreadable auth policy
stays unreadable — never optimistic.
- Reading back: a marked value shows its plain name plus an
unchecked flag; an unmarked one (legacy row, CLI decision, or a
future hub-derived identity) is not classified either way.
- The audit wipe seeds the same kind of marked attribution into its
chain.reset marker.
When the hub starts deriving the value it overwrites the field and
the prefix disappears by itself — no Studio release needed.
Guards: reviewer_identity_test (the value) and
approvals_reviewer_identity_test (every surface that writes or
renders it, against the hermetic fake hub). Visual proof for both
themes via the dialog-shot harness.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
A popup menu is width-capped at ~280px, so the full-sentence German
entries in the audit overflow menu ran past the edge. The clipped
tail of the reset entry was exactly the "(nur local/dev)" scope
that keeps it from reading as "delete evidence", and the label's
centre landed outside the hit box.
Guard in responsive_test: the page-level overflow sweep cannot see
popup menus (their width does not follow the window), which is why
this went unnoticed.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Panel findings against the reworked approvals page, fixed in place:
- Never fabricate the request time: ApprovalRecord.createdAt is
nullable now; a missing created_at omits the line instead of
rendering DateTime.now() (which drifted on refresh). Guard:
approvals_origin_test pins the omit-on-null invariant.
- Copyable errors on approve/reject/batch via showChainErrorSnack
(the hard project rule) — batch surfaces the first real cause.
- Reject requires a reason: ChainInlineHelp strip + confirm disabled
while empty, no more silent close-and-nothing-happens.
- Batch approve applies the same no-data confirmation as the single
path, naming how many selected requests carry no show: data.
- Plainer language: glossary "Vorgang (Flow)", history label FRAGE
(was PROMPT), no-data hint drops developer jargon.
- One-click copy of the run id; history payload pretty-prints like
the card.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Approvals: the pending card now shows its full origin — flow, step,
run id (previously dropped at the Dart mapping layer), project, and
requested-at — under an ORIGIN heading, led by a one-line intro strip
that says what the inbox is and what Approve/Reject do. Approve/Reject
buttons carry tooltips; the history dialog gains project + run id.
Fixes the approvals doc drift (title/details/reviewer ->
prompt/show/timeout_seconds). Guard: approvals_origin_test renders the
card via the hermetic fake hub and pins every origin fact.
Runs: the "hub too old" state now leads with an in-place update button
(same `chain update apply` path as the Diagnose page), the Diagnose
deeplink demoted to secondary, with a CLI-absent fallback. Guard: two
new RunsLoadErrorView widget tests.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Doctor: when the system AI uses an Ollama endpoint that no declared
host service covers (host:port match, /v1 suffix stripped), the
services panel says so in one sentence with a one-click 'declare as
host service' via the new DeclareService RPC — and states honestly
that it takes effect after a daemon restart (the restart button sits
on the same page). Pure suggestOllamaServiceEndpoint pins every
branch.
Shell: one slim, dismissible banner after connecting when the
release manifest offers a newer hub version; dismissal is persisted
per version so each release hints exactly once (pure
shouldShowUpdateHint + a widget test through the fake hub).
Deliberately manifest-based — Studio and hub versions are
independent counters, so a direct comparison would be wrong;
unreleased dev skew stays with the per-page classified states.
The probe stays inert under the test probe override: its timeout
timer leaked into hub_banner_test (the hermeticity class again).
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The detail sheet always answers 'who maintains this?': one
selectable line per maintainer from the store index's new
maintainers list, or an honest 'not specified' when the index
names nobody (StoreMaintainersSection, public for the widget
tests). The install trust gate carries the same fact when
present and stays terse otherwise. DE+EN; dialog-harness proof
captured with the maintainer row.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Even with the honest install badge the hub error stays reachable
(stale store snapshot, race with a store refresh, older hub). The
friendly-error mapper now gives it its own headline plus a hint
naming the three acquisition paths — local module install, adding
the providing store, configuring the MCP/n8n integration — in EN
and DE, with the verbatim hub message kept copyable. Matcher unit
tests EN+DE guard the classification.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The capability set behind the flow editor's Install quick-fix and
the flow list's install badge ingested every store entry's
requiresCapabilities (dependencies, not provided capabilities) and
ignored entry status/kind. Clicking Install on such a capability
ended in the hub's "no store entry for '<name>'" error.
The set now mirrors the hub's install resolver (entry names only,
no planned or federated entries) via installableStoreCapabilities()
with unit tests for both classification states. Unresolvable
capabilities render the editor's 'not in store' state, which
explains the three recovery paths before any click; the editor pin
moves to 0.25.0 (commit-pinned until its tag exists) and the dialog
harness captures the badge states light+dark. Studio 0.78.0.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The doc-verifier pass over the new trust/exposure surfaces came back
PASS with five improvements, all applied:
- the store policy notice gains a 'Learn more' into the security doc
(the notice named security.require_signatures but not where it
lives)
- 'blocked' disables the trust gate's install button — an active
button contradicted the 'install would be refused' statement right
above it (guard test added)
- the unknown-exposure tooltip now says what the operator can do
(check where the name resolves)
- dead l10n key verifPillUnverified removed (unverified is the
page-level notice, never a card pill)
- stale header comment in install_verification.dart corrected
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The trust-gate dialog replaces its generic 'no per-entry status yet
(alpha)' note with the hub's classified verification statement —
pinned store key / trusted publishers (green), installs without
signature checking (amber), install would be refused (red), bridge
entry (neutral) — via one shared describeInstallVerification mapping.
Against a pre-0.23 hub the field is empty and the old honest wording
stays (pinned by test).
Information architecture: policy-off is a GLOBAL fact, so it appears
as ONE ChainInlineHelp notice above the store grid instead of a
warning pill on every card (card noise); only 'blocked' — a genuine
per-source anomaly — earns a card pill. Doctor's host services show
the hub-classified network reach per endpoint (local only / private
network / publicly reachable with a protect-it hint / reach unknown).
Verified end-to-end against the live dev hub (guide harness): the
wire field arrives, the store page shows exactly one policy notice
and quiet cards; trust-gate variants captured via the dialog
harness. Suite 123 green.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
New ChainInlineHelp (intro strip: what this is + what will happen, with
an optional 'Learn more' into the doc sheet) and ChainFieldHelp /
ChainFieldLabel (a '?' affordance per field). First applied to the
add-satellite dialog, which asked for a bare 'name' with no hint of
what a satellite is or does (usertest): it now leads with a plain
explanation + a federation 'Learn more', and the name field carries a
'?'. Both widgets are quiet by design.
Verified: field_help_test covers the widgets + that the dialog explains
itself; dialog_shots_test.dart (a reusable headed dialog-capture
harness) proved the layout in light + dark. Studio 0.76.0.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The Federation and Runs help buttons opened the architecture doc:
federation.md/runs.md existed as assets but had no _DocEntry, so
showFaiDoc resolved the unknown slug to _kDocs.first. Register both
(onWelcome: false, so they don't clutter the newcomer grid but are
reachable), split the Welcome grid onto the curated subset, and make
the unknown-slug fallback assert in debug instead of silently opening
the wrong topic.
Guard (no-bugfix-without-a-guard): doc_help_wiring_test.dart scans
lib/ for every showFaiDoc('slug') call and asserts each has a
registered entry AND both assets/docs/<slug>[_de].md files. Exposes
kKnownDocSlugs for the test.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The switcher listed sealed areas by name ('lbs', 'stromnetz') on
any glance or screenshot — but the names themselves often carry
client/mandate identity (usertest security finding). The sealed
section now renders one aggregated row ('2 sealed areas') with a
deliberate 'Show names' reveal per menu opening; selection still
pops the regular s:<slug> value. Settings -> Security gains 'list
sealed areas with their names right away' (WorkspacePrefs,
SidebarPrefs pattern, default off).
The aggregate row wraps to two lines — popup menus cap their
width and action texts must never be truncated (the first cut
showed '1 abgeschotte…' in the proof shot). Guard: switcher tests
cover aggregated-until-reveal and the Settings toggle; the old
direct-listing test now asserts the reveal contract. DE+EN.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The same single-select choice pattern appeared as four widgets:
audit's hover pills, the store's SegmentedButton, the store filter
dialog's ChoiceChips, and the approvals TabBar (usertest finding
#14 / night-log decision 'pill segment as canon'). The audit
pattern is promoted to a shared ChainSegments widget (optional
icons, hover, selected border, button+selected semantics) and all
four sites use it; approvals switches lists via IndexedStack so
both stay alive and switching does not refetch.
Guard per the no-bugfix-without-a-guard rule: widget tests for
selection + semantics, plus a canon sweep that bans
TabBar/TabBarView/TabController/SegmentedButton/ChoiceChip from
lib/ (comments exempt). Deliberately out of scope: the flow
editor's Graph/Text/Run tabs live in the separate editor package.
Studio 0.75.0; guide images regenerated, dark + light verified.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The widget suites used to talk to whatever listens on the real
endpoint — results depended on the operator's machine (a running
hub fed real data into a11y/responsive runs and its gRPC channel
timers caused the historic flake). Hardening round:
- HubService.instance is now injectable (debugSetInstance);
FakeHubService (test/support/fake_hub.dart) answers every member
the pages touch with healthy-empty defaults and scripts per-RPC
failures (UNIMPLEMENTED / UNAVAILABLE / detached gate) through a
GrpcError-shaped fake. Unimplemented members are recorded and
fail the sweep with the exact list.
- state_matrix_test.dart pins the app-wide invariants for every
sidebar page x hub condition: healthy => no unreachable claims
and no raw error text; hub gone => honest unreachable states;
UNIMPLEMENTED => never 'not reachable' while the sidebar shows
connected; detached gate => plain-language feature-off state.
- a11y + responsive sweeps now inject the fake (hermetic); the
6-minute idle-timer drain workaround is gone with the cause.
Real bugs the new sweep caught immediately:
- every data page (store, doctor, audit, approvals, federation)
folded ANY load failure into 'hub not reachable' — the runs-page
bug class; they now share HubLoadErrorView, which classifies
into unreachable / needs-newer-hub / load-failed-with-copyable-
detail (new generic DE+EN strings)
- the approvals page's hidden tab had no future listener: a load
failure there surfaced as an uncaught async error
- the audit status bar rendered the raw gRPC error wall verbatim;
it now shows the classified friendly headline (still selectable)
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Doctor findings used to be dead-end statements — 'approvals
waiting for review' left the operator to find the approvals inbox
on their own. Every finding with a dedicated surface is now one
tap away from it:
- summary tiles: modules -> store, approvals -> approvals inbox,
audit chain -> audit log (chevron affordance, tooltip + semantics
button; the services tile stays plain — no dedicated page)
- modules/approvals panel rows link the same way
- the event-log headline opens the audit page next to the
existing verify button
- host-services empty state gains a 'view the configuration'
button opening the in-Studio config viewer the hint refers to
- the update banner's release-notes URL is now an underlined,
clickable link instead of dead text
Tiles and the panel are public callback-driven widgets so the
widget tests pump them without a live hub. New DE+EN link labels.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Finally caught with a creation stack trace: when the last gRPC
stream closes, Http2ClientConnection arms the channel's 5-minute
idleTimeout timer — even on a shut-down connection — so the
test's 1-minute drain never covered it and the framework's
pending-timer invariant tripped whenever the arm landed inside
the test window (frequent while a real hub listens on 50051).
The suite now closes the channel in real-async space at the end
of the body (new @visibleForTesting HubService.debugResetChannel;
shutdown is deliberately not awaited — it wedges on a mid-connect
socket, but cancels its timers synchronously) and pumps past the
idle timeout so the timer fires inside the test. 6 consecutive
full-suite runs + 3 isolated runs green; before, roughly 1 in 3
full runs failed.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The runs monitor folded every load failure into 'hub not reachable',
contradicting the sidebar's green connected dot whenever the hub
answered but the RPC failed — most visibly against a pre-0.22 hub
whose version predates the ListInvocations RPC (UNIMPLEMENTED).
Classify the failure instead (top-level, unit-tested):
- UNIMPLEMENTED -> 'this view needs a newer hub version' with a
doctor-page link (the update banner lives there)
- FAILED_PRECONDITION from the detached gate -> the regular
feature-off empty state with the guide button
- UNAVAILABLE / DEADLINE_EXCEEDED / socket-level failures -> the
honest 'hub not reachable' state (unchanged)
- everything else -> a load-failed state with the friendly error
and a copyable detail box
The error view is a public callback-driven widget so the tests pump
each variant without a live hub. New DE+EN strings for the too-old
and load-failed states; grpcCodeOf/grpcMessageOf exposed from the
friendly-error mapper instead of duplicating the duck-typing.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The menu so far only exported the current view (type + search
filter over the page 100-event window). A second action now fetches
the complete event history of the active project scope in one
EventLog call (the RPC has no cursor and no server-side cap) and
writes it as JSONL. Serialisation extracted to a top-level function
with unit tests; a stale comment advertising the never-shipped
"chain audit export" command now names "chain admin events --json".
Studio 0.73.0.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
- Store hero: 'TODAY' badge only for an operator-accepted story; the
rotating compiled-in fallback deck now says 'FEATURED' (no false
freshness claim). Policy as a top-level function with unit tests.
- Audit filter chips: standard label typography instead of mono —
mono stays reserved for paths and identifiers.
- DE chain wording: 'Hash-Kette geprüft' as the one confirmation term
(audit header now matches the doctor pill); 'intakt' stays the
state headline. EN was already consistent.
- Workspace switcher: contrast bump for the sealed-area 'stopped'
label.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Three doctor-page findings from the usertest panel:
- Module tile no longer counts the hub's built-in 'system'
pseudo-module — a fresh hub shows 0 modules, matching the
welcome checklist's definition of an install. Counting is a
top-level function with unit tests.
- When the audit DB lives in an OS-cleanable temp directory
(/var/folders, /tmp, Windows Temp), the daemon-files panel
says so instead of presenting the state as healthy. The
classifier is a top-level function with unit tests.
- The daemon card states who-talks-to-whom-how in one line:
endpoint, transport security (TLS / unencrypted-local /
unencrypted), and whether a bearer token is attached (length
only) — the auditor's baseline the green dot cannot answer.
Also syncs pubspec.yaml (0.70.0 -> 0.72.0) with kStudioVersion,
which had drifted to 0.71.0 while pubspec stayed behind.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Text widgets render markdown backticks verbatim (usertest: reads
as typos / raw dev output) — stripped across both catalogs; code
terms stay inline in plain type. 'Zeig dem Hub' / 'Nimm das CLI'
now use the formal address; the guard test learns both stems.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
- search field over flow/step/module/error/detail/project/id backs
the list and the export ('current view' semantics); match logic
is a top-level function with unit tests
- export writes one JSON object per line via the save dialog; the
CLI stays the canonical WORM-grade export
- the bare trash icon on the audit toolbar read as 'delete
evidence' (security-auditor finding) — the dev-only reset now
sits in a labeled overflow menu next to the export action
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
- 'Fähigkeiten' is the German lead term everywhere (welcome step 2,
integrations panel, federation) — English jargon only as a
parenthesized technical term on first mention
- 'Enrollment-Token' -> 'Registrierungs-Token', bootstrap label in
plain words, CA spelled out, enrollment hint now formal address
- formality guard also rejects reader-addressed imperatives
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
One-click installs showed no trust signal at all (top security
finding of the usertest panel). Every install path — store card,
detail sheet, and the flow list's quick fix — now routes through
one confirmation dialog showing what the hub actually knows
before download: origin store, version, license, maturity, and
required services/capabilities, plus the sandbox model and an
honest note that per-entry signature status is not in the store
index yet (verification happens hub-side at install). Widget
tests cover content and confirm/cancel semantics.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The icon-only rail forced first-time users to guess (usertest:
Senior, a11y, UX personas). Three changes:
- Nav tooltips appear instantly and carry the page shortcut
(Cmd+1..9, Ctrl on non-mac — Ctrl activators added); expanded
labels show the same hint. Explicit button semantics for
screen readers on every destination.
- A visible 'Search & commands' row above the footer opens the
existing Cmd+K palette, which nothing in the UI advertised.
- Settings -> Appearance gains 'Keep the navigation expanded':
pins the rail with permanent labels (persisted preference).
Footer strip and pillar toggle made overflow-safe for the
animating rail; responsive test scrolls the by-design scrollable
destinations list.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Store search, security panel, System-AI fix hint and the
approvals payload hint addressed the operator informally while
the rest of the app uses Sie — the panel flagged the break on
every second screen. All DE strings now use the formal address,
and a guard test rejects any future informal form in app_de.arb.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Stefan's live findings, all four addressed at the root:
- 'In 3 Fragen loslegen' read like ad copy → the entry is now plainly
'Einrichtung starten' / 'Start setup'.
- The setup button sat permanently on the Welcome page of a running
app ('setup after the app runs is backwards' — reported twice). A
fresh install now starts INSIDE the setup: SetupGateScreen hosts
the wizard embedded as the page (new embedded/onFinished modes on
GuidedSetupDialog), with an explicit 'Später einrichten' skip.
Welcome loses the setup button entirely and stays a calm intro.
- Re-running the setup later lives in Settings → General ('Run setup
again…'), the single post-first-run home.
- 'You must grant access first and only then see what will be done':
the preview used to spawn the chain CLI, whose first run could pop
the macOS permission prompt BEFORE the plan was ever shown. The
preview now calls the new PlanSetup RPC over the live hub
connection (no subprocess, nothing granted); the CLI remains only
a fallback when no hub is reachable — and applying stays the
explicit, separate step.
Widget tests: gate hosts the wizard + skip/cancel leave it; CLI-path
tests drive the fallback through the new hub-preview test seam.
Suite 76 green, analyze clean.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
A black window on launch, no error anywhere: main() awaited
loadPersistedEndpoint before the first frame, and
SharedPreferences.getBool threw 'int is not a subtype of bool?' —
the store is writable from outside the app and hub.secure had been
written as int 0. Pref reads now go through defensive typed helpers
(int coerces to bool, wrong types fall back to defaults), and every
pre-frame restore step is failure-isolated: a broken store can cost
a preference, never the first frame. Regression tests stage the
corrupt store (the exact observed value and worse).
Also removes the CocoaPods leftovers from the macOS project
(Podfile, [CP] script phases, Pods framework references, xcconfig
includes): the project builds via Swift Package Manager, and the
dual wiring ran both dependency managers on every build — Flutter's
persistent 'removing CocoaPods will improve build time' warning.
Verified: clean profile build produces a launchable bundle (plugins
statically linked via SwiftPM), suite green, analyze clean.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Review follow-ups on the auth-status work:
- Both daemon-start paths classified an auth-rejected hub as "daemon
dead" via healthy() and showed a start-failure dialog while the
shell banner above correctly blamed the token. They now share
daemonAnswers(): only an unreachable probe counts as down.
- An auth-rejected poll now re-reads ~/.chain/hub-auth-token and
reconnects when the file changed, so a token fixed outside Studio
(CLI, editor) heals the connection without a restart — previously
the client kept the stale in-memory token forever and the banner's
own advice could not work.
- An endpoint switch resets the failure streak, so a stale in-flight
probe can no longer let the unreachable banner blame the new
endpoint for the old one's misses.
- The auth-policy panel re-queries when the hub token is saved or
cleared in the panel above (reloadTick), instead of keeping a
stale admin-denied hint; it also renders the hub's new
reload_required flag as a pending-reload warning (DE+EN).
- today-pipeline.md still documented ~/.fai/today after the rename;
the FAB theme comment now states the both-themes intent.
flutter analyze clean; 71 tests green including four new ones.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The wizard's widget tests all ran in German, leaving the English
variant unverified (open follow-up from the setup redesign). One
EN walk now pins the localized step counter, scenario cards with
explanations, environment step and Next/Back navigation, and
asserts no German strings leak through.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The sustained-failure banner treated every failed health poll as
'can't reach the hub'. With token auth active, a wrong or rotated
token gets UNAUTHENTICATED from a perfectly reachable hub — the
old wording sent the operator to fix the endpoint. The shell now
uses the SDK's probe() and, on auth rejection, switches the banner
to 'rejected the sign-in — check the access token' (key-off icon,
DE+EN). Two widget tests pin the wording per failure kind and the
banner clearing once the probe turns serving.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Accessibility/responsive audit pass with two new permanent test
gates (test/a11y_test.dart: WCAG text contrast + labeled tap
targets on every page in both themes; test/responsive_test.dart:
no layout overflow at 800/960/1280/1920 px). Findings fixed:
- Light theme primary/tertiary sky-500 → sky-700: white text on
the lighter accent only reached 2.8:1 (welcome CTA, active
sidebar label); sky-700 clears WCAG AA at ~5.9:1. Dark theme
unchanged (already compliant). FABs now follow the same accent
instead of Material 3's washed-out tonal default.
- Audit page: filter chips collapse into a checkmark popup menu
below 900 px window width (app bar overflowed); the live-status
bar's left text is now Expanded with ellipsis so the row can
shrink, and the disconnected state's copyable error gets the
full remaining width.
- German strings now use formal address consistently (~20 strings
still used du-forms next to Sie-forms on welcome/setup), the
audit event-type chip "Step" is "Schritt", and the doctor
page's event count pluralises correctly in both languages.
flutter analyze clean, 64 tests green. Screenshot pass light+dark
via the guide-shots harness (verified parity, no overflows).
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Settings → Security now shows the hub's effective auth policy via the
new read-only AuthStatus RPC: active token validator (static / jwt-rs256
with issuer, audience, JWKS source), anonymous-access warning, per-token
cards with scope grants, env-var presence and rate limits, plus a
localized admin-denied story for non-admin tokens. Live-reloads on
endpoint change.
Also fixes a batch of fai→chain rename leftovers this panel's
verification uncovered: hub_auth_token.dart and registry_token.dart
read/wrote ~/.fai/ while the hub reads ~/.chain/ (stored registry
tokens never reached the hub), today_story_loader + tools/today used
~/.fai/today, chain_log legacy ~/.fai/logs migration removed per the
no-legacy-recognisers decision, and UI strings still advertised the
retired .fai bundle extension.
Includes 5 widget tests for the panel, an integration-test screenshot
harness (auth_policy_shots_test.dart, guide-shots style), and DE+EN
l10n. flutter analyze clean, 58 tests green.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Field test of the setup wizard surfaced three trust breaks in one run:
an unexplained macOS Documents permission prompt, a perceived crash,
and an error message whose copy button could not be reached.
Root causes and fixes:
- chain init failures were shown as a SnackBar, which lands BEHIND the
wizard's modal barrier: dimmed, clipped, copy unreachable — and the
click aimed at it hit the barrier, dismissing the whole wizard with
all answers (the perceived crash). Errors now open a modal dialog
ABOVE the wizard via showChainErrorDialog with a copyable detail
block, and the wizard is no longer barrier-dismissible.
- When the resolved chain binary is older than Studio and rejects
--plan-json, the wizard now explains the version skew in plain
language (binary path + update path) instead of leaking a raw clap
usage error. A missing binary gets its own localized story.
- Step 3 announces which chain binary the preview will execute; when
that binary physically lives (symlinks resolved) in a TCC-protected
folder, the wizard pre-explains the macOS folder prompt.
Supporting changes: FriendlyError passes through friendlyError()
unchanged so call sites can ship precise localized stories through the
shared presentation; SystemActions gains resolvedChainBinary() plus
run/resolve test seams; ChainErrorBox hugs its content instead of
filling an unbounded dialog; the wizard's answers file is written
synchronously (the async dart:io variants never complete under the
widget-test fake-async zone).
Verified: flutter analyze clean, 53 tests green (6 new wizard error-
path tests incl. clipboard round-trip), plus a live GUI walk on macOS
in dark + light with a stale binary (skew dialog, copy verified via
clipboard) and with the real binary (TCC pre-explanation with the
resolved path, full plan preview).
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
- Regulated path finishes without a terminal: the signed-source
state offers 'Add a signed source…' (stores dialog with pin-a-key)
plus the per-module install buttons and a plain-language hint why
pinning the publisher's key matters — instead of a hint with no
affordance.
- Apply warnings (e.g. the empty-trusted-publishers caveat) surface
selectable in the done state instead of being swallowed.
- Truthful preview: new lines state which machine is being set up
(server/container targets configure THIS machine), that regulated
profiles always get the hash-chained audit log (even with WORM
off), and that the curated reading list is stored with the setup
record.
- Language pass: onboarding checklist in Sie-form + 'System-KI'
(was du-form + 'System-AI'), 'Audit-Sperre' jargon replaced,
answers file moved to a private per-dialog temp dir.
- Screenshot harness: GUIDE_SHOTS_THEME=light for light-parity
proof runs.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
integration_test/guide_shots_test.dart boots a hermetic hub
(HubFixture), seeds demo projects (open Bürgeramt; sealed
Ratsinformation and a setup.applied audit event only when the
runner confirms an isolated $HOME), launches the app in German +
dark mode, walks every sidebar page in Cmd order with
content-aware waits, opens the workspace switcher and the setup
wizard, and writes the guide PNGs via a driverless RepaintBoundary
capture. Driven by the platform repo's scripts/regen-studio-guide.sh.
Also fixes the hub fixture's binary resolution, dead since the
rename (it looked for 'fai' and ../chain_platform/): now $CHAIN_BIN,
'chain' on PATH, then ../fai_chain/target/{release,debug}/chain —
the integration tests actually run again instead of silently
skipping.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
test/nav_manifest_test.dart derives docs/nav.generated.json from
the _pages list (order = Cmd numbers), the labelOf switch, and both
.arb files. On any nav change it regenerates the manifest and fails
once with instructions to commit + mirror it to
fai_chain/docs/studio/, where the platform repo's docs_consistency
gate checks the operator guide against it — cross-repo nav drift
becomes a red gate instead of quietly rotting docs.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The wizard's first step now offers 'or just describe what you want
to do': the goal goes to the configured system AI, which maps it
onto the menu answers (validated against strict enum whitelists —
a hallucinated value can never reach the engine). The suggestion
comes back as an editable plain-language reflection ('this is how I
read your task') the operator can adjust step-by-step or take to
the same preview/apply the menu path uses. Trust rules per
guided-setup.md: suggestion only (never auto-apply), a privacy line
states whether the description is processed locally or sent to a
provider, and without a configured system AI the section explains
that the menu always works — no dead end.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Post-apply the wizard now renders real Studio actions instead of CLI
text: a start-hub button that polls until the daemon answers,
per-module install buttons (capability-name install via the hub's
store index) with done/progress states, and an open-the-starter-flow
button that navigates to the Flows page. Regulated plans explain in
plain language that modules come from a signed source; the preview
offers 'allow installing from the public store' as one deliberate,
reversible switch that re-assembles the plan (allow_unsigned_modules).
Fresh installs (no config, no setup-plan.yaml) auto-open the wizard
once per run — the wizard IS the onboarding — and it steps back once
a setup exists. The welcome CTA is framed honestly ('get started in
3 questions'), and after the wizard closes the onboarding checklist
remounts, re-probes, and says what the assistant already covered
(profile line from setup-plan.yaml) instead of acting as a second,
disconnected onboarding surface.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Reworks the Setup-Assistent toward the zero-learning-curve bar
(docs/architecture/guided-setup.md, phase 1.1):
- Every scenario/intent/target choice is now a localized option CARD
with a one-line plain-language explanation of what it configures
(DE+EN, Sie-form) — replacing the bare dropdowns whose labels were
English enum humanizations ('Regulated Production', 'This Laptop').
- Three explained steps with a 'Schritt n von 3' progress line
(stakes → task → environment); the two adaptive toggles move to the
last step in plain language (no 'air-gapped' jargon).
- The review step renders a localized PLAIN-LANGUAGE summary built
from 'chain init --answers --plan-json' (the structured SetupPlan) —
'Ch∆In richtet einen regulierten Betrieb ein: signierte Module
verlangt · … · geändert wird nur ~/.chain/config.yaml' — instead of
echoing the CLI's English prose. Warns when an existing config will
be overwritten. After apply: a plain 'Fertig' + next steps.
flutter analyze clean; widget tests for the step flow + German option
labels. Remaining per plan: clickable follow-up actions, signature
dead-end fix, placement/auto-open, and the LLM free-text path.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
The workspace switcher now lists the operator's sealed areas (read from
~/.chain/sealed/ manifests, the same source the CLI uses) below the
shared projects, each with a lock icon and a running/stopped status.
Selecting one is a real connection switch: Studio reconnects its hub
client to the area's own port with a full state reload — one window,
one truth. A stopped area is started first (chain project start) with a
visible notice; a failure surfaces as a copyable error and rolls back to
the shared hub.
While in a sealed area an identity bar under the AppBar is painted in
the area's accent colour and names it, with a one-click Leave back to
the shared hub. The area colour is marking, not theming — Studio's blue
stays the app accent. Selecting a shared project from inside an area
switches the connection back first. The sealed connection is never
persisted across restarts.
New: SealedAreaService (manifest + PID discovery), Workspace sealed
switch logic, ChainSealedIdentityBar, SystemActions.chainProjectStart.
l10n DE+EN. flutter analyze clean; 33 tests green (switcher lists sealed
with lock+status, pill shows active area, identity bar renders in the
area colour). Runtime plumbing (discovery, start, endpoint, reach)
verified headlessly against real sealed instances under a redirected
HOME; the identity-bar screenshot is deferred (display click-automation
failed after sleep on the shared desktop — an environment issue, not a
code gap; the visible components are widget-tested).
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
New Runs sidebar destination listing detached invocations (detach:true)
with phase, current step, project and a Cancel button while
pending/running. Workspace-scoped like Audit/Approvals, polls every 2s.
Detached runs are opt-in (detached.enabled) — the empty state explains
how to enable them. Inline help doc DE+EN. DetachedRun model +
listDetachedRuns/cancelDetachedRun in HubService, backed by the SDK's
listInvocations()/cancelInvocation(). flutter analyze clean; 29 tests
green (sidebar Y-stability updated for the new destination, model
mapping unit-tested).
Screenshot verification (light+dark) deferred — shared desktop in use.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
Multi-project stage 1 against the shared hub (platform design
docs/architecture/projects.md, § Studio):
- ChainWorkspaceSwitcher in the Audit + Approvals AppBars: lists the
registry (colour dot per project, shield for protected, honesty
tooltip), 'All projects' stays reachable — a filter, not a jail.
Selection is persisted and shared via the Workspace notifier.
- Audit page: list query AND live stream re-scoped hub-side on switch.
- Approvals page: pending + history scoped; the sidebar badge counts
the active workspace's pending approvals.
- Flow runs are stamped with the active workspace; a flow file
carrying its own project: keeps it (file wins, CLI semantics).
- Data layer: listProjects/ProjectRef; project fields on AuditEvent,
PendingApproval(+Record), SavedFlow; project params through
HubService. l10n DE+EN. Widget tests for the switcher contract.
Visual verification (light+dark screenshots) still pending — the
shared desktop was in active use; code paths are covered by
flutter test (26 green).
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
A Welcome 'Setup assistant' button opens a wizard that collects
scenario / intent / target (+ approval & data-local toggles), then calls
`chain init --answers` to preview the assembled plan and `--apply --force`
to write the config — reusing the Rust deterministic engine, no logic
duplicated. New SystemActions.chainInit; copyable errors via
showFaiProcessError; EN+DE l10n. analyze clean; smoke test + existing
welcome/sidebar tests pass.
Signed-off-by: flemming-it <stefan.a.flemming@googlemail.com>
