* feat(android): durable offline chat with attachments and history-proof retirement
Every chat send is journaled to the per-gateway Room outbox before any
network attempt, so process death always has one durable recovery owner.
Rows survive gateway ACKs as 'accepted' and retire only once the turn is
proven in canonical chat.history by idempotency key; ambiguous outcomes
(lost ACK, kill mid-send, gateway restart before the transcript write)
park as delivery-unconfirmed for explicit retry, per the fail-closed
model #103273 landed. An unproven accepted head briefly holds only its
own session's queue, and retrying a parked head re-orders still-queued
successors behind it. Offline sends now accept picked images and voice
notes; attachment bytes persist as chunked BLOBs (512 KB chunks, 8 MB
per message, 48 MB per gateway) admitted and retired atomically with
their row. Pre-hello 'main' rows pin to the canonical session at first
dispatch so later default-agent changes cannot retarget captured input;
slash commands are connection-gated by a persisted epoch (legacy queued
command rows migrate to a never-matching sentinel) and never auto-replay
across reconnects. The Room store moves to schema v4 on top of v3's
data-parking migration, adding the epoch column and attachment tables
while preserving queued rows. The queued->sending claim is an atomic
compare-and-set shared by the direct dispatch and the flush loop, the
direct path waits for the startup recovery sweep before claiming, and
failed reconnect attempts republish outbox rows so queued sends stay
visible on offline cold starts.
Proof: 1151 Android unit tests and :app:ktlintCheck green (new
migration, storage, and controller coverage for admission, restarts,
claims, pinning, gating, ordering, and byte round-trips); a live
emulator scenario on the pre-integration build queued text+image
offline, survived force-stop plus a device reboot, resent exactly once
on reconnect (single <rowId>:user turn in the gateway transcript), and
drained the outbox after canonical-history confirmation.
Closes#104087.
* fix(android): rearm outbox recovery when direct-send state persistence fails
* fix(android): keep direct dispatch alive across caller cancellation
The direct send's network phase now runs in the controller scope, so a
cancelled UI scope (leaving the chat screen mid-send) can no longer
strand a claimed row in 'sending' with no user action available; the
dispatch completes and settles the row exactly once. Direct-path state
persistence failures also re-arm the startup recovery sweep, mirroring
the flush path's fail-closed handling.
* fix(android): hand claim-persistence failures to the flush lane
A direct-send claim that fails to persist no longer reads as a lost race:
the admitted row is handed to the flush lane so a healthy connection still
delivers it, and the flush path's fail-closed handling owns any repeated
storage failure instead of the UI reporting success with no active owner.
* fix(android): enforce connection gating and fail-closed parking on every send path
The direct dispatch now rechecks a slash command's connection epoch after
its durable claim, so a reconnect between admission and dispatch parks the
command instead of replaying it on the new connection. Parking a stale
gated row only counts once the write persists; a storage failure drops
health, re-arms recovery, and halts the flush pass instead of spinning or
dispatching the stale row.
* fix(android): recognize gateway-acknowledged run ids in outbox ownership checks
A chat.send ACK can return a run id that differs from the row's
idempotency key (the direct path transfers local run ownership to it).
Ownership, in-flight, backlog, and timeout-park checks now consider both
ids, so reconciliation can no longer park — and Retry can no longer
duplicate — a turn that is still running on the gateway.
* fix(android): close remaining ack-ownership and persistence-failure gaps in the outbox
Flushed sends acked under a divergent run id now transfer local run
ownership like the direct path, so live runs cannot time out and surface
spurious errors for delivered turns. A session pin that cannot persist
stops the dispatch while the row is still safely queued. Reconcile and
timeout parking only report changes their writes actually persisted,
failing closed on storage errors.
* docs(android): record why acknowledged run ids stay in-memory
* [AI] fix(session-memory): forward hook config model to LLM slug generator
When llmSlug is enabled with a model override in the session-memory hook
config, that model was ignored — slug generation always resolved the
agent default model via resolveDefaultModelForAgent.
Add an optional model parameter to generateSlugViaLLM and pass
hookConfig.model from the session-memory handler. When a hook model is
provided, only the model is forwarded (provider is omitted) so
runEmbeddedAgent's built-in resolveInitialEmbeddedRunModel chain handles
alias, provider-qualified ref, and bare-name resolution through a single
source of truth — avoiding the routing drift that blocked prior PRs.
Fixes#89551
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(session-memory): centralize slug model resolution
Honor the session-memory slug model override while keeping default, alias, and provider-qualified model selection in the embedded runner.
Co-authored-by: SunnyShu <shu.zongyu@xydigit.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* feat(ui): add gateway browser panel with annotate and inspect modes
* chore(i18n): translate browser panel strings
* fix(ui): address browser panel review findings (stable tab handles, stale view guard, inspect click suppression)
* fix(ui): make the browser panel reload button reload the remote page
* fix(ui): tile the browser dock against the terminal dock and re-probe evaluate per connection
* fix(ui): label page-reported prompt text as untrusted and accept host:port URL entries
* fix(ui): sanitize inspected selector fragments in the annotation prompt
* fix(i18n): retranslate annotation prompt keys so the page-reported provenance label survives localization
* docs: regenerate docs map for the browser panel section
* chore(i18n): reconcile locale bundles with latest main after rebase
* style(ui): format chat-pane imports after rebase
* chore(i18n): refresh locale metadata after rebase
The workspace rail and message detail panel keyed layout off viewport
media queries, so narrow split-view panes and compact windows got
crushed side-by-side columns, and below 1120px the rail was
display:none with no way to reach workspace files.
Chat panes now measure their own width with a ResizeObserver: under
800px the rail presents as the existing bottom-dock strip (side-dock
controls hide), and when the chat + detail split has under 680px the
detail panel stacks below the thread with a horizontal resize divider.
The phone full-screen detail takeover is unchanged. Bottom-strip path
and summary rows no longer flex-shrink into clipped text.
Closes#104023
* feat(webui): reintroduce opt-in AI purpose titles for tool calls
Restores the chat.toolTitles path removed in #103821, gated behind the new
gateway.controlUi.toolTitles opt-in (default false) so tool rendering stays
fully deterministic with no background model calls unless an operator enables
it. Disabled gateways answer { titles: {}, disabled: true } without loading
the completion runtime, and clients stop asking for the session.
When enabled, titles use canonical utility-model routing: an explicit
utilityModel (operator-chosen provider, like every utility task), else the
session provider's declared small-model default, honoring per-session model
overrides and auth profiles; utilityModel "" disables titles and malformed
refs fail closed — never the primary model. Tool inputs are redacted with the
tools-mode redactor before cache keys or prompts, caller ids are bounded and
never reach the model, and results cache in the per-agent SQLite
cache_entries so repeat views never re-bill.
Also completes two crestodian model-input mock factories that leaked into
sibling tests under shared-registry CI shards.
Fixes#103987
* fix(webui): redact tool-title inputs before truncation
* feat(browser): import Chrome-family system-profile cookies into managed profiles
Import cookies from a real Chrome/Brave/Edge/Chromium system profile (macOS)
into a fresh OpenClaw-managed browser profile so the agent can browse as the
signed-in user. Reads the source Cookies DB via a coherent VACUUM INTO snapshot,
decrypts v10 AES-128-CBC values with the Safe Storage Keychain key (one Touch ID
consent), maps rows to Playwright cookies (FILETIME expiry, SameSite, M124+
domain-hash prefix strip, CHIPS skipped), and best-effort injects them via
addCookies into a mock-keychain profile so they persist without further prompts.
Decrypted values are never logged or returned.
Exposed as agent tool action=importprofile, CLI system-profiles/import-profile,
and POST /profiles/import; action=profiles surfaces importable systemProfiles.
Listing and import are pinned host-local at every surface (gateway, browser
tool, node proxy) since they read the local Keychain and Chrome profiles.
Malformed domain filters fail closed via a shared validator. Gated by
browser.allowSystemProfileImport (default on). Imports cookies only.
* fix(browser): satisfy CI lint (OpenClaw temp dir, Unicode control-char class)
Use resolvePreferredOpenClawTmpDir() instead of os.tmpdir() for the cookie DB
snapshot (messaging/channel runtime tmpdir guard), and match control characters
via the Unicode \p{Cc} class instead of a literal control-char range so the
CLI table sanitizer passes the no-control-regex lint.
* fix(auth): preserve token health after OAuth migration
After a user migrates from OAuth to a token/setup-token credential, the
gateway model-auth rollup reported the provider as missing, producing a
false "model auth expired" warning.
Rename aggregateOAuthStatus → aggregateRefreshableAuthStatus and
extract aggregateProfileStatus helper. OAuth remains authoritative when
present; token credentials are the fallback when no effective OAuth
profile exists. Empty effectiveProfiles stays authoritative (missing).
Token fallback applies regardless of expectsOAuth flag.
Fixes#97996
Co-authored-by: SunnyShu0925 <SunnyShu0925@users.noreply.github.com>
* test(auth): use fake token fixture
* fix(doctor): repair stale auth profile orders
* fix(doctor): inspect retained auth profile stores
* fix(doctor): harden retained auth store proof
---------
Co-authored-by: SunnyShu0925 <SunnyShu0925@users.noreply.github.com>
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* fix(agents): apply stale-run liveness to aborted subagent orphan recovery
Skip stale unended subagent runs during orphan recovery and registry
restore, even when they carry abortedLastRun. Previously, restart-aborted
runs were exempt from the stale-unended age check, allowing hours-old
aborted child sessions to be resurrected after long downtime.
Fixes#90766
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(agents): finalize stale aborted runs instead of only skipping them
Previously the stale-run check in recoverOrphanedSubagentSessions only
incremented the skipped counter. Stale active runs were left unended
because scheduleOrphanRecovery only retries failedRuns, not skipped runs.
Now stale runs are finalized via finalizeInterruptedSubagentRun so they
don't remain orphaned in the registry.
Ref: #90766 review feedback
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(agents): await stale-run finalization in orphan recovery
Await finalizeInterruptedSubagentRun for stale aborted runs and report
failedRuns when finalization does not update the registry, so the
scheduler retry path can recover from finalization failures.
Co-authored-by: Cursor <cursoragent@cursor.com>
* test(agents): faithful restart-path proof for stale orphan recovery
Drive the real recoverOrphanedSubagentSessions against the real subagent
registry, the real isStaleUnendedSubagentRun policy, and a real on-disk
session store, mocking only the outbound gateway transport and transcript
reader. Proves finalizeInterruptedSubagentRun actually ends the stale
aborted run in the registry (endedAt set, outcome error) instead of
resuming it, while a fresh aborted run still resumes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore: amend author email
* fix(agents): scope orphan finalization to run generation
* fix(agents): preserve stale restart recovery ownership
* test(agents): isolate restart recovery scheduling
* fix(agents): make stale restart finalization durable
* fix(agents): keep stale retries generation-scoped
* test(agents): await restart recovery scheduling
* fix(agents): verify interrupted finalization
* fix(agents): defer interrupted finalization during restart
* fix(agents): preserve lifecycle type narrowing
* style(agents): use nonmutating recovery ordering
* chore: keep release note in PR body
---------
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Pick-cat <266665499+Pick-cat@users.noreply.github.com>
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* feat(ui): worktrees page owners and creation, structured snapshot failures, preserved-checkout cleanup
- Worktrees settings page gains an Owner column (manual/Workboard/session
with a link into the owning chat) and a New worktree form with repo
default, optional name, and a worktrees.branches-backed base picker
- non-forced worktrees.remove snapshot failures are a structured
{ removed: false, snapshotError } result (typed WorktreeSnapshotError in
the service) instead of error-string sniffing in the UI
- deleting a session whose dirty/unpushed checkout was preserved now
reports worktreePreserved through the session capability and offers an
explicit force removal instead of silently orphaning the checkout
Part of #103431
* fix(ui): report preserved checkouts from batch session deletes
deleteMany now aggregates worktreePreserved results, and the Sessions
page surfaces one notice pointing at Settings -> Worktrees instead of
silently orphaning dirty checkouts during bulk deletion.
Part of #103431
* fix(ui): reconcile generated locale artifacts after rebase
Open Settings from anywhere in the Control UI with Shift-Command-Comma while preserving browser-owned Command-Comma and existing shortcuts.
Prepared head SHA: 769a4fec5f
Co-authored-by: shakkernerd <165377636+shakkernerd@users.noreply.github.com>
Reviewed-by: @shakkernerd
* feat(ui): sessions sidebar redesign with agent sections, smart groups, and draft sessions
- agents become collapsible top-level sidebar sections; expanding an agent
browses its sessions without navigating, replacing the hidden scope select
- built-in smart groups: one section per channel (rows keep their chat
titles), a Work section for worktree/exec-node sessions with a
repo/branch/node subtitle, custom groups, and Chats
- session names never show raw keys or peer ids; DM fallbacks shorten ids,
dashboard sessions read 'New session', unnamed work sessions read as
their checkout
- one + opens the new-session draft dialog: agent, exec host (paired
system.run nodes), folder, worktree toggle with base-branch picker
(worktrees.branches) and optional name; the first message creates the
session and starts the run in one sessions.create call
- custom group catalog/order moves to the gateway (sessions.groups.*) with
a one-time localStorage migration; rename/delete update members
server-side instead of client-side paging
Part of #103431
* fix(ui): resolve dragged sessions across browsed agent sections
Dropping a row dragged out of a non-active agent section now finds the
session in the per-agent row cache instead of only the active scope, so
the category patch is applied instead of silently doing nothing.
Part of #103431
* fix(ui): propagate group catalog changes to open clients
sessions.groups.put/rename/delete now always broadcast a groups-change
event, and the session capability reloads the gateway-owned catalog when
one arrives, so another browser's group create/reorder/rename/delete no
longer leaves this client on a stale snapshot for the rest of the
connection.
Part of #103431
* docs: restore new session dialog section after rebase
* fix(ui): translate new sidebar/session strings and refresh docs map
Real locale translations for the new-session dialog and sidebar keys
(the fallback gate requires zero recorded English fallbacks), plus the
regenerated docs map for the new Control UI section.
Part of #103431
* fix(ui): repair locale metadata after rebase conflict resolution
* fix(ui): merge mainline locale keys with the sidebar redesign strings
* fix(ui): refresh raw-copy baseline for mainline tool-card strings
* fix(ui): reconcile generated locale artifacts after rebase
* fix(anthropic): delegate adaptive CLI effort
Strip static --effort args for adaptive runs so Claude Code resolves effort from its environment, settings, and model default. Preserve configured effort for off or absent thinking and replace it only for concrete OpenClaw levels.
Fixes#103245
Co-authored-by: Dan Korotin <korotin.daniil@gmail.com>
* fix(anthropic): make effort dispatch exhaustive
---------
Co-authored-by: Dan Korotin <korotin.daniil@gmail.com>
* feat(agents): derive a provider-declared default utility model when unset
When agents.defaults.utilityModel is not set, utility tasks (titles, progress
narration) now use the primary provider's declared small model
(modelCatalog.providers.<id>.defaultUtilityModel: OpenAI -> gpt-5.6-luna,
Anthropic -> claude-haiku-4-5). Auth is inherent because derivation follows
the agent's primary provider. Setting utilityModel to an empty string
disables utility routing entirely; narration turns on automatically when a
default resolves and stays off otherwise.
Formatting/lint/tests verified on Testbox (oxfmt, oxlint, plugins:inventory,
docs:map, import-cycles, check:test-types, 4 test files).
* fix(ai): drop the temperature parameter for models that reject it
The GPT-5.6 family 400s on temperature via the Responses API (live-verified:
gpt-5.6-luna/-terra reject it; gpt-5.5 and gpt-5.4-mini/nano accept it).
supportsOpenAITemperature gates all three OpenAI payload builders, with a
catalog compat override (compat.supportsTemperature) declared for the 5.6
family in the openai manifest. Without this, utility tasks (titles,
narration) would fail once gpt-5.6-luna becomes the derived default.
Gates on Testbox: oxfmt, oxlint, plugins:inventory, import-cycles,
check:test-types, 6 test files. Live proof on Testbox: gpt-5.6-luna and
claude-haiku-4-5 one-shot completions succeed with temperature requested.
* fix(agents): carry the primary model's auth profile onto the derived utility default
A primary like openai/gpt-5.5@work previously reached utility tasks via the
profiled fallback; the derived default shares the provider, so its trailing
auth profile carries over instead of silently switching to default
credentials (Codex review). Gates on Testbox: oxfmt, check:test-types, tests.
* docs: realign the manifest provider-fields table after adding defaultUtilityModel
