A stream that stalled or lost its upstream was closed with no terminal frame
at all, so clients saw "200 OK, a few chunks, then nothing" and could not tell
a truncated reply from a finished one. The Responses passthrough path already
synthesized response.failed; every other client format got nothing.
The watchdog now hands its reason ("stream stall timeout" or "upstream
connection lost") to onAbortTerminal, and buildStreamErrorBytes frames it per
client format: OpenAI-compatible clients get data: {"error":{...}} followed by
data: [DONE], Anthropic clients get `event: error`. The error frame always
precedes [DONE] (openai-python raises APIError on any data payload carrying an
error key), and no synthetic finish_reason is ever emitted — a truncated
stream must not look like a clean stop.
Co-Authored-By: Claude Code <noreply@anthropic.com>
The merged Qoder work also rewrote shared translator/handler code so that
/v1/responses clients got token usage on response.completed. That changed
behaviour for every provider, not just Qoder: proxies saw input tokens
rise by the 2000-token context buffer, and the plain token mapping was
replaced by one that always adds input_tokens_details.
A probe confirms the Qoder benefit does not depend on those edits: the
executor's coalescer already emits one include_usage-style finish chunk, so
a Claude client receives input_tokens and cache_read_input_tokens with
every shared file at its original state. Only the Responses path relies on
the shared translator, and that path has no Qoder-owned seam to put it in.
Reverts the shared files to their pre-PR state and drops the Responses
usage test. The Cline envelope unwrap in nonStreamingHandler.js, which
landed after the PR in the same file, is kept.
Codex's /responses validator has no Unicode property escapes, so a tool
`pattern` containing `\p{...}` 400s the whole request with `Invalid schema
for function ... is not a 'regex'` — identically on every account, costing a
full combo failover per turn (#3922).
- Add open-sse/utils/codexToolSchema.js: copy-on-write walk that drops only
`pattern` values carrying a property escape, returning the original
reference when nothing changed so the caller's schema stays intact for a
retry against another provider
- Treat `properties` keys as property names, so a field literally called
`pattern` is never read as the schema keyword; skip escaped literals via
backslash-parity counting
- Apply it in normalizeCodexTools for both function and namespace sub-tool
parameters, and log the strip count via dbg
- Add three cases to tests/unit/codex-tool-normalization.test.js
- Coalesce Qoder's empty finish-in-delta frame with the later choices:[] usage
frame so OpenAI and Claude clients receive prompt_tokens, completion_tokens
and cache-hit tokens (the dashboard already saw them)
- Upload inlined images through /api/v2/image/upload like qodercli, and stub
oversized non-image files instead of stuffing 30MB+ data URIs into
agent_chat_generation
- Emit response.completed -> response.usage for chat-native upstreams so
/v1/responses clients (Codex CLI, sub2api) no longer log 0/0/0
- Keep Claude message_delta.usage working when usage arrives without choices[0]
- Escalate to the smallest advertised Qoder context tier (200K/400K/1M) when
the estimated prompt no longer fits max_input_tokens
- Pass apiKey for PAT connections and list hidden enable:false catalog keys
from /v1/models
- add claude-fable-5-1 to the Claude Code model catalog (1M context,
permanent adaptive thinking)
- centralize the spoofed Claude Code version and update both request
and billing identities to 2.1.257 (Fable 5.1 rejects < 2.1.251)
- send output_config.effort without the redundant thinking switch for
permanently adaptive models
- add regression coverage for capabilities, headers, billing identity
and adaptive-effort payload
# Conflicts:
# open-sse/providers/registry/claude.js
# open-sse/providers/shared.js
# open-sse/utils/claudeCloaking.js
# tests/__baseline__/providers-baseline.json
With the 1M-context beta enabled, Claude Code sends model: "claude-opus-5[1m]".
The marker is a client-side annotation — it matches no combo name, no alias and
no provider/model pair — so the request dies at model resolution with
"Invalid model format" and the client reports "There's an issue with the
selected model". Every request from that session fails until the beta is
switched off.
New open-sse/utils/modelMarkers.js exporting stripModelContextMarker(modelStr)
-> { model, contextMarker }. handleChat strips the marker before resolution and
normalizes body.model so downstream logging and translation see the real name.
Only a trailing marker is stripped, so a model whose name genuinely contains
brackets is left alone.
The capability itself travels in anthropic-beta: context-1m-2025-08-07, which
the default executor already forwards untouched — only the routing key needed
cleaning.
Fixes#3690.
Tests: tests/unit/model-context-marker.test.js (6 cases).
Anthropic gates newly released models (e.g. claude-fable-5-1) to Claude
Code >= 2.1.251; the spoofed 2.1.92 client got HTTP 400 on every request.
Bump User-Agent + billing-header version to 2.1.258 and refresh the
providers baseline snapshot.
Co-Authored-By: Claude Code <noreply@anthropic.com>
Claude Code carries the session in metadata.user_id, which the Responses
API translation drops before the executor resolves a cache session. The
request then fell through to the assistant-text hash and the per-connection
fallback, so one conversation was split across several prompt_cache_key
values and the upstream prefix cache kept missing.
Fall back to the x-claude-code-session-id header, which survives every
translation. The body stays authoritative when both are present.
buildUsage() only emits cache reads under prompt_tokens_details, so the
top-level-only read dropped the count for every Responses-format provider
(codex, grok-cli, ...), persisting cached_tokens: 0 and billing cache hits
at the full input rate. Mirror the cache_creation fallback already used
just above.
translateResponse() short-circuited untouched on claude->claude streaming,
so OAuth-cloaked tool names (CLAUDE_TOOL_SUFFIX) leaked to the client and
every tool call was rejected as unknown. Add decloakStreamChunk(), the
streaming counterpart of decloakToolNames(), and call it on the same-format
path using the already-plumbed state.toolNameMap.
Strict Anthropic-compatible gateways (e.g. MiniMax) reject Claude-format
requests with HTTP 400 when tools[].type is missing. Normalize each
missing/falsy tools[].type to "custom" before dispatch when the final
request format is Claude. Built-in tool types (computer_use, bash,
web_search_*) are passed through untouched.
createSSEStream splits on "\n" and keeps the remainder, which only flush()
parses. That call omitted targetFormat, so parseSSELine required a "data: "
prefix and dropped whatever an NDJSON provider left without a closing
newline. The !parsed.done guard compounded it: the SSE sentinel and an
Ollama final chunk both carry done:true, but the latter is the real last
chunk holding done_reason and the token counts.
Pass targetFormat and scope the sentinel check to formats that emit one, so
the tail reaches the translator. Accumulate its usage into state the same way
the transform loop does, so finalizeStream logs those tokens instead of null.
Responses-API clients (codex, droid) close the socket on
response.completed because the protocol has no [DONE] sentinel, so
every successful request printed "⚡ DISCONNECT: ResponseAborted"
after its own "📊 done" line. Keep the dbg("CTRL", …) trace and drop
the console line; ABORTED and ERROR still print.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The Responses API has no [DONE] sentinel, so codex closes the socket
as soon as response.completed arrives. That cancels the reader before
flush() runs — and flush() held every usage side effect, so a fully
successful request logged nothing: no 📊 done line, no token stats,
no request detail.
Extract that tail into a once-guarded finalizeStream() and also call
it right after the terminal event is forwarded, in both passthrough
and translate mode. flush() still calls it; the guard makes the
second call a no-op. Streams that end normally are unaffected, and a
terminal event carrying no usage falls through to the existing
estimate/null path rather than blocking.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
detectClientTool only matched the legacy "codex-cli" User-Agent, so the
current codex-tui CLI and Codex Desktop (UA "Codex Desktop", originator
"codex_work_desktop") fell through to null and lost native passthrough —
their requests got re-translated, stripping/overwriting fields like
reasoning.summary instead of forwarding the client body as-is.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Codex Responses Lite clients routed to a chat-native OpenAI-compatible
provider lost tool use in three places: non-streaming Chat responses
leaked the raw chat.completion envelope instead of Responses output
items, internal reasoning continuity fields leaked into the outbound
Chat body causing some upstreams to reject the request, and the
Responses to Chat request translator ignored additional_tools,
custom_tool_call, and custom_tool_call_output items entirely.
Also fixes apiType (chat vs responses) for openai-compatible nodes
being resolved from the immutable provider ID instead of the stored
node config, so editing a node's API Type had no runtime effect.
The global claudeHeaderCache singleton overlaid the last-seen Claude Code
client's identity headers onto every subsequent request, leaking one
client's headers (anthropic-beta, user-agent, x-stainless-*, etc.) onto
another client/account sharing the same server. Removed the singleton and
the claudeOverlay hook entirely, falling back to static per-provider
headers. anthropic-beta is now computed per-request from the requested
model, gating heavy-agent flags (advanced-tool-use, effort) to
opus/sonnet only.
Real Cursor IDE now uses AgentService at agent.api5.cursor.sh (HTTP/2-only)
while 9router still spoke the retired ChatService at api2.cursor.sh with
outdated headers, producing HTTP 429 "Update Required". Add an executeAgent
path that builds an agent.v1.RunRequest Connect RPC over a raw http2 stream
and fetches the account-specific usable model catalog via GetUsableModels.
Also implement MCP tool calling over AgentService: encode OpenAI tools as
AgentRunRequest.mcp_tools (McpToolDefinition with google.protobuf.Value
input_schema), decode McpArgs tool calls, and forward them to the client as
OpenAI tool_calls so the client runs the tool and resumes in the next turn.
Reply to request_context_args with a non-empty RequestContext, to server
heartbeats with client_heartbeat, and to KV blob get/set with empty results,
so action queries no longer stall the stream. Fold the client system prompt
into the user message (custom_system_prompt makes the server return an empty
turn). Bump clientVersion to 3.12.17 and add the x-cursor-client-commit
header so the gateway identifies as a current Cursor IDE release.
Reshape Kiro direct requests so resumed client sessions reuse Kiro's
cache-affinity fields instead of starting unrelated CodeWhisperer
conversations.
- keep conversationState.conversationId stable when the client sends an
explicit session id (x-session-id, session_id, conversation_id, Claude
Code session metadata)
- add a stable conversationState.agentContinuationId per Kiro session
- send conversationState.agentTaskType: "vibe" and agentMode: "vibe",
matching the normal Kiro CLI/KAS chat path
- move Kiro thinking instructions into Kiro-compatible systemPrompt /
additionalModelRequestFields instead of generic top-level thinking
- keep volatile timestamp context out of the top-level systemPrompt; it
remains only in user content fallback
- suppress additionalModelRequestFields for legacy 4.5-era Claude/Kiro
models that reject it, while defaulting future Claude/Kiro model ids
to supported
- preserve Kiro meteringEvent credit usage internally for accounting
without leaking provider-specific fields into OpenAI-compatible usage
- prevent unrelated headerless Kiro requests from sharing one
connection-wide continuation
- cap/evict continuation sessions so long-running processes do not grow
the continuation map unbounded
- treat generated headerless Kiro sessions as one-shot so they do not
evict real explicit-session continuations
- keep credit-only Kiro metering valid for internal persistence when
token metrics are unavailable
Normalize every provider to one cache-inclusive convention via
canonicalizeUsage() before persist, and price cached + cache_creation as
subsets of prompt_tokens in calculateCostFromTokens() to stop
double-counting. usageRepo now delegates cost math to a single source.
Surface Cached tokens/cost across dashboard (overview, tokens, cost,
details). Merge Claude message_start cache with message_delta output so
cache counts survive. Compatible LLM nodes now allow multiple API-key
connections (key pool).
Co-authored-by: Cursor <cursoragent@cursor.com>
CodeBuddy CN includes "tool_calls": [] in every SSE streaming delta.
@ai-sdk/openai-compatible checks delta.tool_calls != null — an empty
array passes ([] != null is true in JS), triggering premature
reasoning-end on every reasoning chunk (0/1ms durations in OpenCode).
Strip empty tool_calls arrays in passthrough before hasValuableContent.
Zero side-effect: real tool_calls always have at least one element.
Closes#2176
Co-authored-by: Cursor <cursoragent@cursor.com>
Treat response.done as a terminal OpenAI Responses stream event so
passthrough streams ending with response.done are not flagged incomplete
and no synthetic response.failed is emitted. Restore the data: [DONE]
sentinel for same-format Responses passthrough streams.
Co-authored-by: Cursor <cursoragent@cursor.com>
Backfill DEFAULT_THINKING_AG_SIGNATURE onto functionCall parts missing it
(client history replay) and on Claude tool_use blocks, fixing 400
INVALID_ARGUMENT from Gemini-family APIs. Suppress the OpenAI-style
data: [DONE] sentinel for antigravity/gemini/vertex to avoid parser crashes.
Fixes#2193.
Co-authored-by: Cursor <cursoragent@cursor.com>
logUsage now only logs to console; DB write removed. Streaming usage is
recorded once via saveUsageStats (onStreamComplete), eliminating duplicate
usageHistory rows that inflated dashboard totals.
Co-authored-by: Cursor <cursoragent@cursor.com>
## Fixes
- **Kiro**: honor thinking effort budgets
- **AG/Kiro/Xiaomi**: provider fixes
- **Combo/Fusion**: flatten tool history in panel calls to prevent 503
- **LLM selector**: show custom vision models in selector and model list
- **Image**: prevent compatible nodes from shadowing provider aliases
Gom block SSE headers + CORS lặp ở streamingHandler + responsesHandler
vào sseConstants.SSE_HEADERS_CORS. Codex format-routing giữ nguyên
(logic-driven theo kim chỉ nam DATA/LOGIC docs 07).
Co-authored-by: Cursor <cursoragent@cursor.com>
convertOpenAIToolChoice mapped {type:"function"} verbatim and cloakClaudeTools
left tool_choice.name unsuffixed, both rejected by Claude on the cc/ path.
Map forced-function to {type:"tool",name}, allowlist Claude-valid types, and
suffix tool_choice.name when it targets a renamed client tool.
Fixes#1592
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>
## 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
## 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)
* fix: enhance stall detection in stream handling for improved disconnect management
* fix: improve stall detection handling in pipeWithDisconnect to prevent stale aborts
