Add wenyan-lite/wenyan/wenyan-ultra levels for max token compression,
sync SHARED_EXAMPLES/AUTO_CLARITY/PERSISTENCE across all levels, and
expose 3 wenyan buttons in endpoint settings UI.
Co-authored-by: Cursor <cursoragent@cursor.com>
MiniMax requires reasoning_content echoed back on assistant messages in
multi-turn/tool-call conversations. Add minimax and minimax-cn to
PROVIDER_RULES (scope all), same fix as DeepSeek (#1543).
Co-authored-by: Cursor <cursoragent@cursor.com>
Kiro requires a non-empty currentMessage tools array whenever history
references any tool use, else returns "Improperly formed request" (400).
Clients trip this by omitting tools on follow-ups after client-side
compaction.
- flattenToolInteractions(): no client tools -> collapse tool_use/result
to text so the "tools required" rule never fires
- reconcileOrphanedToolResults(): client tools -> salvage orphaned
results as text, keep matched ones, guard co-located tools array
- safeJSONParse(): guard tool-call argument parsing against bad JSON
- merge consecutive user userInputMessageContext; null-guard
currentMessage for assistant-only input
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Add shared OAuth credential lifecycle manager with provider-aware refresh
decisions. Implement CodexExecutor.refreshCredentials so 401/403 retry
refresh works for Codex, track lastRefreshAt and refresh before the
upstream stale-token window, preserve omitted idToken, and add
per-connection single-flight refresh to avoid refresh-token rotation races.
Merged from PR #1664.
Co-authored-by: Cursor <cursoragent@cursor.com>
- Add qmodel_latest to QODER_MODEL_MAP
- Expose qmodel_latest in static Qoder provider catalog (qd)
- Generalize executor comment so model set does not go stale
- Add unit coverage for the new model key + catalog
Closes#1638
Co-authored-by: Cursor <cursoragent@cursor.com>
- Add gemini-3.5-flash-extra-low across CLI menu, provider models, usage, pricing
- Add MITM synonyms (high/medium/extra-low) and split pattern so Low no longer falls through to Medium
- Strip models/ prefix in getMappedModel for AG public name normalization
Co-authored-by: Cursor <cursoragent@cursor.com>
Add mimo-v2.5-pro-claude alias routing to the Xiaomi TokenPlan Anthropic-compatible
/anthropic/v1/messages endpoint. Logic lives in a dedicated XiaomiTokenplanExecutor
(config-driven via targetFormat) instead of the shared DefaultExecutor.
Co-authored-by: Cursor <cursoragent@cursor.com>
Route image model tests to /api/v1/images/generations and STT to
/api/v1/audio/transcriptions instead of forcing all non-embedding
models through chat completions. Adds kind-aware pingModelByKind,
hf->huggingface alias, and silent WAV sample for STT reachability.
Scoped to dashboard/internal model testing only; runtime inference
routing is unchanged.
Author: yicone <yicone@gmail.com>
Closes#1628
## Features
- Add Qoder provider: device-flow OAuth, COSY signing, WAF-bypass body encoding, live model catalog, dashboard quota tracker, 11 models (#1372)
- Add new models: Claude Opus 4.8 (Claude Code), GPT 5.4 Mini (Codex)
## Fixes
- DeepSeek thinking mode: echo `reasoning_content` back on follow-up/tool-call turns so OpenCode-free and custom providers no longer 400 with "reasoning_content must be passed back" (#1543)
- Reasoning injector: match deepseek/kimi model ids case-insensitively (covers custom providers using capitalized model names)
- OpenCode suggested-models: include free models without the `-free` suffix, e.g. `big-pickle` (#1535)
## Improvements
- Codex: trim sunset models, keep gpt-5.5 / gpt-5.4 / gpt-5.3-codex family, add gpt-5.4-mini
- volcengine-ark: refresh model list (add DeepSeek-V4-Flash/Pro, drop EOL entries)
- Lower stream stall timeout 35s → 30s for faster hang detection
Wire Qoder credits into the Quota Tracker card grid:
- Add `qoder` to USAGE_SUPPORTED_PROVIDERS so the connection passes the
isUsageEligible filter at /api/providers/client and shows up in
providerOptions on the dashboard.
- Reshape getQoderUsage so quota records (user, organization) live under
`quotas` and scalar metadata (totalUsagePercentage, isQuotaExceeded,
expiresAt) are siblings — the parser used to walk Object.entries(quotas)
and would have rendered `totalUsagePercentage: 0.42` as a "0/0" row.
- Surface Qoder's expiresAt as resetAt on each quota record so the card
shows when credits reset.
- Add a parser branch in ProviderLimits/utils.js: rename internal keys
(user → "Personal", organization → "Organization"), drop empty org
buckets so personal accounts don't render a misleading "0/0 Organization"
row, and forward remaining/unit so the QuotaProgressBar can use them.
- Add Qoder's brand color (#EC4899) to ProviderLimitCard's color map.
42 tests still pass; build clean.
Adds 18 new tests covering the bugs fixed in the previous commit so they
can't silently regress:
- parseExpiry (7 tests): numeric ms-epoch input, numeric strings handled
before Date.parse so "1700000000" doesn't get year-interpreted, RFC3339
strings, expires_in:0 honored as already-expired, 30-day fallback only
when both inputs are missing/invalid
- normalizeMessages (4 tests): system hoisting, multipart text flatten,
multiple system joining, empty input
- wrapQoderSSE (6 tests): the fixed cases — trailing partial line drained
in flush(), no chunks forwarded after [DONE], embedded newlines stripped
from inner body, error envelope produces error chunk + [DONE], non-ok
responses returned unchanged
- expose parseExpiry from auth.js, expose normalizeMessages/wrapQoderSSE
via __test__ from the executor (internals only — not part of the public
API). Marked with comment so the surface is intentional.
42 tests total (24 original + 18 new). Build still clean.
Correctness:
- testUtils: drop checkExpiry so the userinfo URL probe actually runs (revoked
tokens used to look "active" until local 30-day expiry passed)
- auth.parseExpiry: handle numeric expiresAt, swap parseInt before Date.parse
so "2026" doesn't get interpreted as year-2026, treat expires_in:0 as
already-expired instead of fabricating a 30-day default
- providers.mapTokens: synthesize email from userId when fetchUserInfo fails
so OAuth dedup works (re-logins no longer accumulate "Account N" rows)
SSE wrapper:
- wrapQoderSSE: add !doneEmitted guard on success branch (chunks could leak
past [DONE] when an error envelope shared a TCP packet with a valid one)
- flush(): finalize TextDecoder + drain trailing buffer so the chunk carrying
finish_reason is delivered when upstream closes without a final \n
- sanitize literal \n inside inner OpenAI body so SSE framing stays intact
Robustness:
- executor: wrap buildCosyHeaders in try/catch so a missing accessToken
returns 401 (re-auth) instead of bubbling as 500
- executor: short-circuit on missing accessToken before signing
- executor: plumb proxyOptions/signal through buildQoderRequestBody so
proxy-only networks can fetch the model_config catalog
- qoderModels: dedupe concurrent first-time misses with an in-flight Promise
map (parallel chat windows now do 1 upstream fetch instead of N)
- qoderModels: check signal.aborted before addEventListener so a pre-aborted
parent signal cancels the inner fetch immediately
- auth: AbortController + 15s timeout on pollDeviceToken / fetchUserInfo to
prevent hung sockets when openapi.qoder.sh stalls mid-response
UX:
- OAuthModal: derive polling deadline from device-code expires_in (qoder
publishes 300s; the previous fixed 120s caused timeouts when users took
more than 2 minutes on the consent page)
Cleanup:
- delete src/lib/oauth/services/qoder.js — referenced removed config fields
(clientId/clientSecret/tokenUrl/authorizeUrl) and was re-exported from
services/index.js, so any future caller would TypeError on first use
GitHub Copilot's /responses endpoint only serves OpenAI (gpt/codex)
models. gemini-3.1-pro-preview was failing on /chat/completions with a
"not supported" error, getting cached as a codex model, then escalated
to /responses where it 400s with "does not support Responses API".
Add GithubExecutor.supportsResponsesEndpoint() and gate both the cached
/responses route and the 400-fallback on it, so Gemini/Claude always
stay on /chat/completions and the real upstream error surfaces.
Adds tests/unit/github-responses-routing.test.js (5 tests).
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
## Fixes
- Codex: auto-retry when upstream drops mid-stream (no more hangs)
- Codex: fix random 400/404 errors, tool-calling failures, and unstable prompt cache
- MITM: support Antigravity 2.x (updated IDE version detection and DNS/cert flow)
- Sanitize Read tool args to prevent retry loops from non-Anthropic models (#1144)
- Implement json_schema fallback for OpenAI-compatible providers without native Structured Output (#1343)
- Strip empty Read pages argument in OpenAI-to-Claude translator (#1354)
- Forward Gemini output dimensions for embeddings (#1366)
- Resolve setState-in-effect errors in dashboard components (#1362)
- Gemini CLI: reuse stored OAuth project IDs for quota checks and show clearer setup guidance when the project is missing (#1271, #1428)
When json_schema response_format is sent to models that do not natively
support Structured Output (e.g. DeepSeek/Ollama/local LLMs via
openai-compatible-* providers), they often return empty or malformed
content.
Detect response_format.type === "json_schema", inject the schema into the
system prompt, and downgrade response_format to json_object so Structured
Output works transparently across openai-compatible providers.
Gated to provider.startsWith("openai-compatible-") so providers with
native Structured Output support are not downgraded.
Co-authored-by: Cursor <cursoragent@cursor.com>
* fix: sanitize Read tool args to prevent retry loops from non-Anthropic models
* fix: sanitize invalid Read pages from tool args
Non-Anthropic models sometimes emit optional Read args like pages: "" for
non-PDF files, which Claude Code rejects before the tool runs. Drop invalid
pages values, keep valid PDF page ranges, and coerce numeric string bounds
before clamping limit/offset.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Some OpenAI-compatible providers emit optional string tool parameters
as empty strings (e.g. pages: "") instead of omitting them. Claude
Code rejects pages: "" as invalid, breaking the Read tool for
non-PDF files routed through 9router.
Add sanitizeToolArguments() that parses tool-call arguments and
removes known optional empty-string fields before emitting
input_json_delta back to Claude format. Currently handles the
Read tool pages field specifically.
Includes regression test.
Fixes#1278
Co-authored-by: JoJo <noreply@github.com>
## Features
- Xiaomi MiMo Token Plan: region selector (Singapore / China / Europe) — keys are cluster-specific
- Antigravity: risk confirmation dialog before first connection
- Gemini CLI: surface upstream retry delay on 429 errors
## Fixes
- MITM: cannot kill process on macOS under sudo (lsof not found in PATH)
- Stream: false-positive stall timeout on Claude reasoning / Kiro responses
- Tunnel: cannot re-enable after disable (stuck state)
- Tunnel: cloudflared error messages now include log tail for easier debugging
- Language switcher: applies selected locale immediately on close (#1234)
- Antigravity OAuth: metadata now matches the official client
## Improvements
- Gemini CLI: bump engine to 0.34.0
- Re-hide `qwen` (OAuth EOL) and `iflow` (not ready) providers
Fixes#1226
The Antigravity OAuth flow sent inconsistent client metadata between
the token acquisition phase and the API usage phase. String enum values
(IDE_UNSPECIFIED, PLATFORM_UNSPECIFIED) were used during OAuth token
exchange + loadCodeAssist + onboardUser, while numeric enums (ideType: 9,
platform: <computed>, pluginType: 2) were used in runtime API calls.
Google detected this fingerprint mismatch and blocked 9router accounts.
Replace all string enum occurrences with the correct numeric values:
- src/lib/oauth/constants/oauth.js: loadCodeAssistClientMetadata now
uses getOAuthPlatformEnum() for platform and numeric 9/2 for
ideType/pluginType, matching getOAuthClientMetadata()
- src/lib/oauth/services/antigravity.js: getMetadata() now delegates
to getOAuthClientMetadata() instead of returning hardcoded strings
- src/lib/oauth/providers.js: postExchange metadata now uses
getOAuthClientMetadata() instead of inline string enums
- open-sse/services/usage.js: getGeminiSubscriptionInfo body now uses
CLIENT_METADATA (already imported from appConstants.js) instead of
inline string enums
* fix: enhance stall detection in stream handling for improved disconnect management
* fix: improve stall detection handling in pipeWithDisconnect to prevent stale aborts
## Features
- Add Vercel AI Gateway provider support (#1183)
- rtk: Kiro format tool result compression — handle conversationState.history & currentMessage, preserve error results, ~13.6% savings (#1194)
## Fixes
- openclaw: normalize agent.model object form `{primary, fallbacks}` before .startsWith → fix TypeError & 'not configured' status (#1216)
- Usage Details pagination: stay inside mobile viewport <640px (#1218)
- Fix test model error
- Fix MIMO provider in Codex
- Disable log file creation when using MITM AG
Deepseek API (and likely other providers) reject messages with
role: 'developer' — only accept system, user, assistant, tool.
filterToOpenAIFormat() normalizes content blocks but never touched
message roles, so developer passed through unmodified and caused
400 errors (issue #773).
Fix: add one-line developer → system mapping in filterToOpenAIFormat()
before role-specific logic. This is the common normalization point
called for all targetFormat=openai providers (Deepseek, Groq, Mistral,
Perplexity, Together, Fireworks, Cerebras, xAI, NVIDIA, etc.)
Closes#773
## Features
- Add Cline & Kilo Code tool cards
- Tailscale TUN mode for stable Funnel TLS
- Sort APIKEY providers by usage, collapse to top 20
## Improvements
- Local Material Symbols font (no Google Fonts)
- Docker base: Bun → Node 22-alpine
- MITM reads aliases from JSON cache (no native sqlite)
- Stream stall timeout (2 min) in open-sse
## Fixes
- Fal.ai key test: use stable models endpoint
Two findings, neither blocked by anything else:
1. src/app/callback/page.js — the OAuth callback page posted the
{ code, state } payload to window.opener with targetOrigin "*", so any
page that opened the popup against the well-known redirect_uri received
the live OAuth code. The expectedOrigins list was already computed but
never used. Iterate over it and pass the origin per send.
2. open-sse/utils/proxyFetch.js — createBypassRequest() set
rejectUnauthorized: false on the HTTPS request that runs after the
Google-DNS-resolved real-IP fallback (used for cloudcode-pa.googleapis,
GitHub Copilot, Cursor, AWS LLM endpoints). Combined with servername:
parsedUrl.hostname this gave SNI-correct connections that nonetheless
ignored cert validation, so an on-path attacker could swap in their
own cert and read the user's API tokens / prompts. Drop the flag.
Detected by Aeon + semgrep (javascript.browser.security.wildcard-postmessage-configuration
+ problem-based-packs.insecure-transport.js-node.bypass-tls-verification).
Severity: HIGH (#1) / MEDIUM (#2).
CWEs: CWE-1385 (#1), CWE-295 (#2).
Co-authored-by: aeonframework <aeon@aeonframework.dev>
Update all DeepSeek model prices to match current V4 Flash pricing
($0.14/$0.28 per 1M input/output tokens), and add V4 Pro model with
its own pricing ($0.435/$0.87). Also add deepseek-v4-pro to the
provider model list.
Co-authored-by: smarthomeblack <truongbber@gmail.com>
- Introduced OllamaLocalExecutor to handle requests for the "ollama-local" provider.
- Removed the direct URL construction for "ollama-local" from BaseExecutor.
- Updated index.js to include the new OllamaLocalExecutor in the executors mapping.
- Enhanced the ProvidersPage component to support dynamic addition of OpenAI/Anthropic compatible providers.
Add input_audio and audio_url content type handlers to
convertOpenAIContentToParts() in geminiHelper.js, converting
OpenAI audio format to Gemini inlineData format.
Also add audio types to VALID_OPENAI_CONTENT_TYPES in
openaiHelper.js so they are not stripped by filterToOpenAIFormat().
Fixes#912
- BaseUrlSelect: add cloud endpoint option, custom URL local state, always
default to first option; new cliEndpointMatch helper; CLI tool cards refactor
- API: new /v1/audio/voices and /v1/models/info; /v1/models filters disabled
models, drop unused timestamp
- initializeApp: guard tunnel/tailscale auto-resume to once-per-process
- geminiHelper: ensureObjectType for schemas with properties but no type
- skills: minor SKILL.md tweaks (chat/embeddings/image/stt/tts/web-*)
- Introduced a caching mechanism for in-flight token refresh requests to prevent race conditions and reduce unnecessary API calls.
- Added error handling for unrecoverable refresh errors, ensuring that the application can gracefully handle token reuse and invalidation scenarios.
- Updated the MITM server management to handle port 443 conflicts, allowing users to kill processes occupying the port before starting the server.
- Improved user feedback in the MitmServerCard component regarding port conflicts and admin privileges.
- Refactored the ComboList component to streamline the display of media provider combos.
This update aims to enhance the reliability and user experience of the token management and MITM functionalities.
- Removed the isCloud check from proxyFetch.js, simplifying the fetch patching logic.
- Updated MediaProviderDetailPage to include a new section for API key retrieval, improving user experience with clearer layout and additional notice text.
- Enhanced ConnectionRow to better handle email display names.
- Improved ProviderDetailPage to conditionally render provider notices and API key links.
- Refactored localDb, requestDetailsDb, and usageDb to remove unnecessary isCloud checks, streamlining database interactions.
- Updated OAuthModal to combine waiting and manual input steps for a more cohesive user flow.
- Added API key URLs to several providers in providers.js for better accessibility.
- Introduced Xiaomi MiMo as a new provider in providerModels.js and providers.js.
- Updated model alias mapping in model.js to include Xiaomi MiMo.
- Enhanced validation route to support Xiaomi MiMo API endpoints.
- Added Xiaomi MiMo to APIKEY_PROVIDERS with relevant details.
This update expands the range of supported providers, improving integration capabilities.
- Refactored handleChatCore to include Caveman functionality, allowing for terse-style system prompts to reduce output token usage.
- Updated APIPageClient to manage Caveman settings, including enabling/disabling and selecting compression levels.
- Adjusted AntigravityExecutor to consolidate function declarations for compatibility with Gemini.
- Removed unnecessary console logs during translator initialization across multiple routes.
- Updated refreshCredentials methods in various executors (Antigravity, Base, Default, Github, Kiro) to accept optional proxyOptions for improved proxy handling.
- Modified token refresh logic to utilize proxy-aware fetch for better network management.
- Enhanced usage retrieval functions to support proxy options, ensuring seamless integration with proxy configurations.
- Updated ModelSelectModal and ProviderInfoCard components to incorporate kind filtering for improved user experience in model selection.
- Added validation for API keys in the provider validation route, including support for webSearch/webFetch providers.
- Introduced Cloudflare AI as a new provider with specific configurations in providerModels.js and providers.js.
- Updated DefaultExecutor to handle account ID resolution for Cloudflare AI connections.
- Enhanced AddApiKeyModal and EditConnectionModal to include account ID input for Cloudflare AI.
- Implemented validation for Cloudflare AI API key connections in testUtils.js and route.js.
- Updated UI components to reflect changes in provider management and connection handling.
- github.js: split thinking vs reasoning_effort stripping
- thinking (Claude-native format) still stripped for all Claude on Copilot
- reasoning_effort now passed through for Opus 4.6 and Sonnet 4.6
- still stripped for Haiku 4.5 and Opus 4.7 (rejected upstream)
- reasoning_effort "none" stripped for all models (not all support it)
- openai-to-claude.js: map reasoning_effort → thinking.budget_tokens
for direct Anthropic backend (none→skip, low→4096, medium→8192,
high→16384, xhigh→32768)
Previously reasoning_effort was stripped for ALL Claude models,
meaning Opus 4.6 via Copilot never received thinking configuration.
AI-generated commit by Claude Opus 4.6 (Anthropic)
- Added new image models for GPT 5.2, 5.3, and 5.4, including capabilities for text-to-image and editing.
- Updated embedding handling to include optional dimensions in requests.
- Introduced support for custom embedding providers, allowing dynamic fetching and validation of custom nodes.
- Improved image generation handling with Codex integration, including progress tracking and error handling.
- Enhanced UI components to support adding custom embeddings and displaying their status.
Add Volcengine Ark as a first-class API key provider with official model presets, endpoint configuration, API key validation, model discovery, connection testing, provider logo, and runtime alias mapping for `ark/*` model IDs.
Made-with: Cursor
Co-authored-by: kingsy <kingsylin@vip.qq.com>
- Introduced OpenCode Go provider with relevant configurations.
- Enhanced model management by allowing users to add and delete custom models.
- Updated UI components to support model selection for image types.
- Adjusted sidebar visibility to include image media kinds.
Add a shared resolveOllamaLocalHost() helper and wire it through the
executor, models/validate/test routes, so users can point ollama-local
at a remote Ollama instance instead of being locked to localhost:11434.
Also expose the host as an "Ollama Host URL" field in AddApiKeyModal
(empty = default localhost:11434), making the option reachable from the
dashboard without hand-editing db.json.
Co-authored-by: anuragg-saxenaa <anuragg.saxenaa@gmail.com>
Made-with: Cursor
The previous blanket strip in GithubExecutor.transformRequest removed
`thinking` AND `reasoning_effort` for every GitHub-routed model to avoid
Claude-on-Copilot 400s from OpenClaw. That regressed GPT-5 family support
(gh/gpt-5-mini honors reasoning_effort: low/medium/high).
Make supportsThinking(model) model-aware — return false only for Claude
models, so the strip fires only where the upstream actually rejects these
fields.
Benchmarks on /v1/chat/completions via GitHub Copilot:
effort=(none) → 64 reasoning_tokens, ~2.0s
effort=low → 0 reasoning_tokens, ~1.55s
effort=medium → 64 reasoning_tokens, ~1.9s
effort=high → 128 reasoning_tokens, ~2.2s
Made-with: Cursor
GitHub Copilot /chat/completions endpoint does not support the thinking
or reasoning_effort fields. OpenClaw sends thinking: { type: "enabled" }
for Claude models which causes a 400 Bad Request.
Added supportsThinking() and strip both fields in transformRequest before
sending to the upstream endpoint.
Co-authored-by: anuragg-saxenaa <anuragg.saxenaa@gmail.com>
Made-with: Cursor
* fix: add multi-model support for Factory Droid CLI tool (closes#521)
* fix: show quota auth expired message for Kiro social auth accounts (closes#588)
Remote HTTP(S) image URLs are fetched and inlined as base64 data URIs
in a new prefetchImages() step run before super.execute(), so the body
sent to Codex contains resolved image bytes instead of URLs the backend
cannot access.
Scope is limited to the Codex executor — base executor and other
providers are untouched.
Co-authored-by: anuragg-saxenaa <anuragg.saxenaa@gmail.com>
Made-with: Cursor
* fix: add multi-model support for Factory Droid CLI tool (closes#521)
* Add Claude Opus 4.7 to cc and cl provider lists
RESEARCH confirmed GA release April 16, 2026. Adding to:
- cc (Claude Code): claude-opus-4-7
- cl (Cline): anthropic/claude-opus-4.7
Refs: TICKET-20260416-ENG-O4.7-001
* fix: add Blackbox AI as a supported provider (closes#599)
- Integrated Google TTS languages from a separate module for better maintainability.
- Updated local device voice fetching to support both macOS and Windows, improving cross-platform compatibility.
- Enhanced dashboard route protection by adding dynamic settings for login requirements and tunnel access.
- Introduced UI elements for managing security settings related to API key requirements and dashboard access via tunnel.
- Added default TTS response example in the media provider page for better user guidance.
- Updated constants to reflect changes in TTS provider configurations.
This commit improves the overall user experience and security of the TTS features.
sseToJsonHandler.js unconditionally deleted reasoning_content from all
non-streaming responses (added for Firecrawl SDK compatibility). This
breaks thinking models (Qwen3.5, Claude extended thinking, etc.) where
the model may use all tokens for reasoning, leaving content empty.
When reasoning_content is stripped in that case, the response appears
completely empty to the client.
Fix: only strip reasoning_content when the response also has non-empty
content, so that reasoning output is preserved when it is the only
useful output.
Co-authored-by: Agent Zero <agent@agent-zero.local>
Cursor's API now rejects requests with outdated client versions,
returning [400]: Update Required for Composer 2. Bump
x-cursor-client-version from 2.3.41 to 3.1.0 across all three
locations where it is set.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add clientDetector utility to identify CLI tools (Claude Code, Gemini CLI,
Antigravity, Codex) from request headers. When the CLI tool and provider
are a native pair, skip all translation — only swap model and Bearer token.
Made-with: Cursor
Replace empty reasoning_content with explicit </think> closing tag when exiting thinking block to properly signal end of reasoning section in streaming responses.
