* feat(chat-ui): tool-diff follow-ups from PR review
Persist bounded applied-diff details in the transcript cache so cold
opens keep inline diffs; parse apply_patch envelopes into per-file diff
sections with stats; thread tool-result isError so failed calls suppress
proposal diffs and read as failures; replace per-line truncation with
horizontal scrolling for long diff lines.
* chore(i18n): sync native inventory for tool-diff follow-ups
* fix(chat-ui): render move-only patches and drop zero diffstats
* refactor(sessions): keep late-media attachment text out of persisted transcripts
Late-resolved media appended a second user turn whose content was literal
"[media attached: <path>]" lines, so clients rendered wire scaffolding
under the image — same leak class as #111204. Persist those turns bare
(media fields + a protected __openclaw.lateMedia marker, image blocks
kept), and project the attachment lines at the LLM boundary and the
history-image-prune stage instead, byte-identical for providers across
string, array, and legacy forms. Also dedupes media-field sniffing into
a shared hasPersistedMedia helper.
* fix(sessions): satisfy type gates for late-media boundary projection
* fix(agents): narrow prune content cast to array element type
* test(sessions): split transcript persistence suite under max-lines cap
* feat(ui): author avatar chips on user messages with gravatar and initials fallback
Renders a small author avatar on user-authored chat messages, including
queued/pending sends, so multi-user gateways show at a glance who wrote
which prompt. Resolution: profile avatar URL (when available) -> Gravatar
by SHA-256 email hash (d=404) -> deterministic initials with stable color.
Assistant/tool output stays unmarked, and messages without sender identity
render unchanged, keeping single-user gateways visually identical.
* fix(ui): clear avatar fallback state on image load; document gravatar tradeoff
* fix(ui): gate email-hash avatars behind a gateway proxy base URL
Without a same-origin proxy base the email tier is disabled outright, so
rendering a transcript can never disclose hashed sender emails or viewer
IPs to a third-party avatar host. The gateway-side proxy route (follow-up)
supplies the base and becomes the only avatar fetch path.
* fix(ui): restrict profile avatar URLs to same-origin relative paths
* fix(ui): validate profile avatar URLs with the URL parser, not string prefixes
Backslashes normalize to slashes and tabs/newlines are stripped by browser
URL parsing, so startsWith checks can be bypassed into cross-origin fetches.
Resolving against a probe origin and requiring the origin to survive is
parser-equivalent and closes the class.
hasInboundAudio previously accepted <media:audio> placeholder bodies,
[Audio transcript headers, and filename sniffing alongside MediaTypes.
It now consumes only the structured facts channels already carry
(MediaType/MediaTypes, kind audio or audio/*), removing the string
reverse-engineering ahead of the placeholder-elimination program.
Discord — the one channel with a coverage hole — now carries its native
audio classification (MIME, voice fields, constrained filename fallback)
as structured kind: "audio", with definitive fetched/declared MIME
precedence so generic container types cannot erase declared audio and
filename hints cannot override definitive non-audio MIME.
* feat(chat-ui): inline diffs for edit and write tool rows
Port the web Control UI's kind-aware diff rendering to the shared SwiftUI
tool rows: persisted details.diff parses into numbered add/del/ctx lines,
edit/write/text-editor args fall back to a local capped line diff, the
collapsed row shows a +N/-M diffstat, and the expanded surface renders the
diff with the result summary or error diagnostic beneath it.
* fix(chat-ui): preserve tool-result details through message reconciliation
Canonical-adoption rebuilds enumerated message fields and silently dropped
details, so inline diffs vanished after cache-warm history reconciliation
(every app relaunch). Carry details through adoptingCanonicalMessage and
messageWithTimestampIfNeeded and lock it in with regression tests.
* chore(i18n): sync native inventory for tool diff rows
* fix(chat-ui): render diffstat and line numbers as verbatim text
* fix(chat-ui): use theme tokens for diff colors
* refactor(chat-ui): drop test-only parseDetailsDiff wrapper
* fix(chat-ui): address diff-row review findings
Interpret details.diff only for edit/write/patch tool names so plugin
detail schemas stay opaque; number write previews from line 1 with an
80-line cap and a subtractive char budget while keeping the exact stat;
bound per-line rendering to 2000 unicode scalars; announce add/del rows
and collapsed skip markers to assistive tech.
* chore(i18n): sync native artifacts for diff-row labels
* fix(chat-ui): announce diff line numbers to assistive tech
* feat(ui): live agent activity subtitles on running sidebar sessions
* refactor(ui): lazy-load narration controller off the startup chunk, unexport test-only symbols
* chore(ui): raise startup JS budget to 312 KiB for the lazy narration feature
* test(ui): restore real timers after narration controller cases
* feat(browser): send pages to the main session from the Chrome extension
One-click page share in the OpenClaw Chrome extension: toolbar popup with an
optional note, page/selection context menu, and Alt+Shift+S. Capture is
selection-first with a readability heuristic, X/Twitter thread extraction, and
Google Docs plain-text export via the user's session cookies. Payloads ride the
existing paired relay WebSocket as a new pageShare message; the gateway-only
page-share sink wraps page text in the external-content safety boundary, then
enqueues a main-session system event and requests an immediate heartbeat
(hooks/wake semantics). Node-hosted relays report a clear unsupported error.
Capture heuristics adapted from Nat Eliason's MIT-licensed send-to-openclaw.
Co-authored-by: Codex <codex@openai.com>
* fix(browser): keep page-controlled metadata inside the share safety boundary
Review findings: move title/URL inside wrapExternalContent (a hostile <title>
must not become trusted header text), prefer the user's selection over the
full Google Docs export, and pass the context-menu selectionText through so
iframe selections and selections cleared during relay reconnect still win.
* fix(browser): bind context-menu shares to the click-time document
Selection shares from the context menu now send the click snapshot directly
(no recapture), so navigations during relay reconnect cannot mislabel the
source and iframe selections are preserved. The Google Docs selection probe
scans all accessible frames before falling back to the full-document export.
* test(browser): expect the page-share handler in relay server args
* fix(browser): probe only the main frame for Google Docs selections
All-frame injection rejects wholesale when one frame is inaccessible and
returns child frames in nondeterministic order, so the probe now reads the
main frame only. Child-frame selections still share correctly through the
context menu's click-time selectionText; toolbar/shortcut entry sends the
full page for that case (named tradeoff in the code comment).
* fix(browser): satisfy page-share CI gates
---------
Co-authored-by: Codex <codex@openai.com>
Live-append voice transcripts into the agent session and persist a per-agent SQLite call record across relay and client transcript paths.
Add run-scoped spoken confirmation for high-impact actions, mutation digests, bootstrap-context injection, talk.client.transcript and talk.client.close protocol methods, and Control UI adoption. This adds zero new configuration.
Co-authored-by: Clifton King <clifton@users.noreply.github.com>
The Nodes & Devices screen showed pending pairing counts but pointed users
at a CLI command on the gateway host. An already-trusted phone can now act:
approve/reject pending requests (gated on hello methods plus the
operator.pairing or admin scope, mirroring the gateway's second
authorization gate for requested roles/scopes) and remove paired devices
(admin-gated, matching the gateway's removal contract). The confirm dialog
is the consent surface: it renders the complete requested roles/scopes
unbounded and scrollable with control/bidi sanitization, keyed to the
gateway identity so a pending confirmation cannot survive a gateway switch.
Mutations capture the gateway scope at claim time, require the exact write
acknowledgement plus a canonical device.pair.list readback before claiming
success, surface definitive gateway denials verbatim, and take over the
refresh generation with the verified snapshot so stale refreshes cannot
republish pre-mutation state. Bootstrap-requested scopes stay on the
canonical limited profile; extending it with operator.pairing is a
deliberate gateway-side product decision left open.
Publicly-CA-valid gateway hosts (e.g. Tailscale Funnel with Let's Encrypt)
no longer force fingerprint pinning: a real platform-trust handshake with
HTTPS hostname verification and SNI connects without prompting, ending the
changed-pin re-prompt on every cert renewal. Candidate gating excludes
.local, IP literals, and single-label hosts so LAN endpoints keep pinning.
Existing pins stay authoritative: a valid public chain never silently clears
a stored pin - the changed-pin prompt gains an explicit one-time 'Use system
trust' switch instead. Discovered endpoints still require a stored pin for
reconnect eligibility (discovery data is attacker-controllable). When the
TLS probe cannot capture a certificate, the trust dialog now accepts an
out-of-band SHA-256 fingerprint paste (sha256:/colon tolerant, 64-hex
validated) which is then enforced on connect. Malformed stored pins now
fail closed instead of silently degrading. Both probe attempts share one
connect+handshake budget so stalled hosts cannot double the probe latency.
* feat(android): accept audio and document attachments via share sheet and composer
Extends Android share-sheet intake beyond text/images: audio plus a curated
document set (PDF, OOXML, CSV, Markdown) with code-side mime enforcement,
provider-type revalidation while the URI grant is live, bounded streaming
reads, per-kind size budgets aligned with the outbox command limit, and a
document/audio picker in the composer. Share parsing moves off the main
thread behind a bounded slot queue; draft-not-autosend and drop/failure
reporting behavior unchanged.
* test(android): cover audio and document share intake
* test(android): align composer draft test with attachment kinds
* chore(android): refresh native i18n inventory
* chore(android): rebuild native locale artifacts for inventory order
* fix(android): stop push-to-talk from dropping speech after mid-hold pauses
The SpeechRecognizer PTT path ran one session with a 2.5s silence window,
kept only the last final, and never restarted during a hold - speech after
a mid-hold pause was silently dropped. Capture now uses an ordered rung
ladder: API 33+ app-owned AudioRecord piped via EXTRA_AUDIO_SOURCE with
EXTRA_SEGMENTED_SESSION, then silence-keyed segmented sessions, then a
restart-during-hold single-session fallback for older devices; failed holds
only descend. Final segments accumulate and merge locale-aware (no ASCII
separators after CJK/Arabic punctuation), with a trailing live partial
deduped against the last final. Release waits one bounded grace for the
terminal recognizer callback so late finals are not truncated; rapid
re-press drains the prior release instead of cancelling it; cancellation
mid-release still tears capture down, resumes realtime talk, and resolves
the stop payload. Raw-rung release also stops the recognizer since
EXTRA_AUDIO_SOURCE is optional and a service may run its own microphone.
Input level meters from our PCM stream in the raw rung where onRmsChanged
never fires.
* fix(android): API-33 recognizer extras behind RequiresApi and inventory refresh
Min-SDK lint (InlinedApi) flagged the API 33 RecognizerIntent extras; they
now live in RequiresApi(TIRAMISU) helpers behind explicit SDK gates, and
the native i18n inventory is refreshed for the moved lines.
* chore(android): rebuild native locale artifacts for inventory order
* feat(codex): fork upstream-linked sessions at a message via thread/fork
* fix(gateway): fail closed for rewind and branch switch on upstream-linked sessions
* fix(codex): fail closed on first-message forks, image-only prompts, and orphan archival
* fix(codex): baseline retained history and reject paginated threads on upstream fork
* fix(codex): validate the full fork prefix and fail closed across crash windows
* fix(codex): treat all non-text inputs as unverifiable in fork drift checks
* fix(codex): support first-message forks as empty-history upstream cuts
* fix(codex): reject source-id reuse and unverifiable hidden inputs in fork boundaries
* refactor(codex): materialize upstream forks from verified thread read-back
* fix(codex): satisfy strict type lanes and knip for upstream fork
* fix(agents): preserve ANSI sanitizer state across bash chunks
* fix(agents): harden streaming ANSI sanitization
Keep incremental parser state in the canonical terminal owner, avoid a second sanitizer pass, and leave OutputAccumulator and public terminal APIs unchanged.\n\nCo-authored-by: Jicheng Xu <xu.jincheng@xydigit.com>
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
