truncateText shortened the cached lastMessage preview with value.slice(0, max - 3), which can cut a surrogate pair in half and emit a lone surrogate into the codex CLI session list JSON. Use the shared truncateUtf16Safe helper so truncation falls back to a whole code-point boundary. Add regressions for both the history.jsonl and sessions/**/*.jsonl preview paths.
generateJobName truncated reminder content with String.slice(0, 20) on
a UTF-16 code-unit index, so an astral character (e.g. an emoji) landing
on the boundary was cut into a lone surrogate, producing a malformed
cron job name.
Truncate with the shared truncateUtf16Safe helper so a surrogate pair is
never split, keeping the existing 20-unit budget and ellipsis suffix.
Adds a test asserting the truncated job name contains no lone surrogate.
buildReflectionPrompt truncated the thumbed-down response with
String.slice(0, 500) on a UTF-16 code-unit index, so an astral
character straddling the 500-char cap was cut into a lone surrogate in
the reflection prompt built for the LLM.
Use truncateUtf16Safe so truncation never splits a surrogate pair,
keeping the existing 500-char budget and '...' suffix.
Adds tests asserting the prompt stays UTF-16 well formed when truncating
and that a boundary emoji is preserved when it fits.
Route xAI OAuth through device-code sign-in so remote and headless hosts do not need a localhost callback. Preserve the legacy manual `xai-device-code` auth choice/method as a compatibility alias to the same device-code flow.
Also migrate stale xAI token endpoints on refresh and fail fast on structured refresh errors while keeping retries scoped to detected HTML/Cloudflare challenge responses.
Verification:
- `node scripts/run-vitest.mjs extensions/xai/index.test.ts extensions/xai/xai-oauth.test.ts`
- `node scripts/run-vitest.mjs src/cli/models-cli.test.ts -t 'maps --device-code'`
- `node scripts/run-vitest.mjs src/commands/auth-choice.test.ts -t 'removed provider auth choice'`
- Crabbox local-container live smoke on exact head `fef3cb24afb01cd1f69cf04ef67ed11d71dfadb3`: xAI discovery and device authorization returned 200.
- `$autoreview` after the live smoke: clean.
Co-authored-by: Jaaneek <Jaaneek@users.noreply.github.com>
Co-authored-by: fuller-stack-dev <263060202+fuller-stack-dev@users.noreply.github.com>
wrapNoteMessage measures every fit decision in visible columns, but its
splitLongWord fallback (for a single word longer than the line budget) sliced
the word into groups of maxLen code points. maxLen is a visible-column budget,
so a run of wide characters (CJK / fullwidth / emoji, 2 columns each) produced
lines up to twice the budget — e.g. a 24-char CJK word at maxWidth 20 emitted a
40-column line.
Accumulate grapheme visible width (the same unit visibleWidth uses) and start a
new segment when the next grapheme would exceed maxLen, so every wrapped segment
fits the budget; a single grapheme wider than the budget still emits, preserving
progress. ASCII wrapping is unchanged.
Co-authored-by: ly-wang19 <ly-wang19@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The live-frontier turn-timeout resolution only matched `anthropic/` models,
so `claude-cli/*` routes fell into the 120s fallback bucket instead of the
180s Anthropic floor, and claude-cli Opus missed the 240s Opus floor — even
though claude-cli serves the same Anthropic Claude models.
Recognize claude-cli as Anthropic-family via a small `isAnthropicFamilyModel`
helper, mirroring the existing `provider === "anthropic" || provider === "claude-cli"`
precedent in the aimock and mock-openai servers.
Co-authored-by: clawSean <260045960+clawSean@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
summarizeParentMessage truncated the parent body with text.slice(0, PARENT_TEXT_MAX_CHARS - 1), which can cut a surrogate pair in half and emit a lone surrogate into the injected "Replying to @sender: …" system event. Use the shared truncateUtf16Safe helper so truncation falls back to a whole code-point boundary. Add a regression asserting the summary stays isWellFormed() when the limit lands inside an emoji.
markdownToWhatsApp only handled **bold** and __bold__, so combined GFM
strong+emphasis such as ***bi***, __*y*__ or **_x_** was reduced by the
plain bold rule first and left a literal ** around the inner emphasis
(e.g. ***bi*** -> **bi**), which WhatsApp renders as plain characters.
Handle the combined strong+emphasis variants before the plain strong
rules and emit WhatsApp bold+italic (*_text_*). Plain bold, italic,
strikethrough and code-span handling are unchanged.
Adds it.each cases for the GFM bold-italic variants and a regression
case ensuring bold-italic markers inside inline code are preserved.
* fix(irc): sanitize internal tool-trace lines from outbound text
* fix(irc): sanitize internal tool-trace lines from outbound text
---------
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
* fix(validation): preserve null in anyOf unions instead of coercing to empty string
Fixes#96716
* fix(validation): preserve null in anyOf unions instead of coercing to empty string
* fix(validation): preserve null in anyOf unions instead of coercing to empty string
---------
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
* fix(opencode-go): re-arm idle timer on block-boundary events to prevent false stalled-stream abort
When the opencode-go model finalizes a tool call and deliberates before
the next one, the provider emits real block-boundary SSE events
(text_end, thinking_end, toolcall_start, toolcall_end) that prove the
socket is alive, but the watchdog's isProviderProgressEvent only
returned true for token deltas (text_delta, thinking_delta,
toolcall_delta). This caused the idle timer to fire and falsely abort a
live stream, replacing a completed answer with a stalled error and
dropping the provider's real done event.
Fix: include block-boundary events in isProviderProgressEvent so the
idle timer is re-armed on any forward-progress provider event.
text_start and thinking_start are intentionally excluded because they
are synthetic preamble events that should not shorten the first-event
window.
Closes#96518
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* test(opencode-go): satisfy lint in stream regression
* test(opencode-go): satisfy lint in stream regression
* test(opencode-go): satisfy lint in stream regression
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
* fix(provider-transport-fetch): bound SSE buffer to prevent OOM
* fix(provider-transport-fetch): appease oxlint curly rule in test
* fix(provider-transport-fetch): drain events before cap + cancel reader on overflow
* fix(provider-transport-fetch): remove unused encoder from coalesced chunk test
Co-Authored-By: Claude <noreply@anthropic.com>
* fix(transport): tighten SSE buffer guards
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
* fix(openai-chatgpt-responses): bound streaming success-body SSE reads at 16 MiB
Apply createSseByteGuard to src/llm/providers/openai-chatgpt-responses.ts
(parseSSE) so the ChatGPT Responses SSE parser cannot be exhausted by a
hostile or malfunctioning endpoint that streams an unbounded SSE body.
The 16 MiB cap matches the non-streaming readProviderJsonResponse cap.
What changed:
- src/llm/providers/openai-chatgpt-responses.ts: wrap body.getReader() in
createSseByteGuard before the existing parseSSE loop. The function is
also re-exported as parseSSEForTest for direct test access.
- Inline test added to openai-chatgpt-responses.test.ts: hostile 1 MiB
pull stream, asserts canonical overflow message, cancel(reason) was an
Error instance, pullCount bounded to 17-20.
Reuses the createSseByteGuard helper from src/agents/streaming-byte-guard.ts
(introduced in #96701 for the anthropic-transport-stream path; same
helper, same 16 MiB cap, same overflow-error shape). This PR is the
third of the planned per-surface rescue series for the previously closed
PR #96666, after PR #3b (#96723).
No SDK surface change. No repro script committed (proof in this body).
* fix(openai): bound ChatGPT error body reads
* fix(openai): bound chatgpt error parsing
---------
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
* fix(agents): truncate exec command detail on code-point boundaries
compactRawCommand in src/agents/tool-display-exec.ts middle-truncated the
one-line command with raw String.prototype.slice. When the head or tail
boundary fell between the two UTF-16 code units of a surrogate pair (e.g. an
emoji like U+1F600), the slice kept a lone surrogate, which renders as the
replacement character in the tool-call summary shown in chat/transcripts.
Use the existing sliceUtf16Safe helper for both ends so the boundary falls on
a code-point boundary, dropping the whole emoji instead of half of it. This is
behavior-preserving for non-surrogate input.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* refactor: move surrogate-safe slice helpers to browser-safe module
ClawSweeper flagged that importing sliceUtf16Safe from src/utils.ts into
tool-display-exec.ts pulls node:fs/os/path into the Control UI browser-shared
bundle (ui/vite.config.ts treats tool-display-exec.ts as browser-shared).
Move sliceUtf16Safe/truncateUtf16Safe into a self-contained, dependency-free
module src/shared/text/surrogate-safe-slice.ts. src/utils.ts re-exports them
(zero churn for existing Node-side callers), and tool-display-exec.ts now
imports directly from the node-free module so no Node built-ins can leak into
the browser bundle.
* fix(agents): use shared utf16 helper in exec display
---------
Co-authored-by: ly-wang19 <ly-wang19@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
registerNodesCli unconditionally registered plugin CLI commands, so
lightweight built-in commands like `nodes status`/`nodes list` paid the
full plugin CLI/runtime load cost. Only resolve plugin-provided node
subcommands (e.g. `nodes canvas`) when the invoked subcommand is not
already a built-in, keeping the built-in nodes path fast.
Fixes#96697
* fix(googlechat): replace unbounded response.json() with readProviderJsonResponse
Replace the local readGoogleChatJsonResponse and
readGoogleChatCertsResponse wrappers with the existing SDK helper
readProviderJsonResponse (from openclaw/plugin-sdk/provider-http) so the
Google Chat API JSON responses are bounded at 16 MiB, matching the
non-streaming cap already used by 15+ other extensions.
What changed:
- extensions/googlechat/src/api.ts: readGoogleChatJsonResponse now
delegates to readProviderJsonResponse. Removed the local try/catch
wrapper.
- extensions/googlechat/src/auth.ts: readGoogleChatCertsResponse now
delegates to readProviderJsonResponse. Error message preserved.
Removed the local try/catch wrapper.
This PR applies the same pattern as Alix-007's #96042, #96038 (lmstudio,
provider JSON reads). No SDK promotion needed — readProviderJsonResponse
is already available in openclaw/plugin-sdk/provider-http.
* fix(googlechat): add inline bounded-read regression tests
Co-Authored-By: Claude <noreply@anthropic.com>
* fix(googlechat): remove unused variable flagged by oxlint
Co-Authored-By: Claude <noreply@anthropic.com>
* fix(googlechat): bound api error body reads
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
* fix(agents): truncate tool-display detail on code-point boundaries
coerceDisplayValue truncated a long first-line detail value with raw
UTF-16 slice() at half = floor((maxStringChars-1)/2). When an emoji
(surrogate pair) straddles the cut boundary, the head kept a lone high
surrogate and the tail could begin on a lone low surrogate, rendering as
the replacement character. Use the existing sliceUtf16Safe helper so the
whole code point is dropped at the boundary, matching the UTF-16-safe
truncation used elsewhere in the repo. Behavior-preserving for
non-surrogate input.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(build): import surrogate-safe slice from node-free leaf module
tool-display-common.ts is in the UI browser bundle graph; importing
sliceUtf16Safe from utils.js dragged node:fs (via infra/fs-safe) into the
bundle and broke build-artifacts/QA Smoke. Extract sliceUtf16Safe/truncateUtf16Safe
into src/shared/utf16-slice.ts (dependency-free) and re-export from utils.js to
preserve the existing import surface.
* fix(build): import surrogate-safe slice from node-free leaf module
* fix(build): import surrogate-safe slice from node-free leaf module
* fix(build): import surrogate-safe slice from node-free leaf module
---------
Co-authored-by: ly-wang19 <ly-wang19@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
* fix(mcp): emit image content with base64 source
* fix(mcp): keep plugin tool images in SDK schema
* test(mcp): exercise image bridge end to end
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
