ghc-proxy 0.10.0 → 0.10.1

package/README.md

@@ -2,7 +2,7 @@
 
 [![npm](https://img.shields.io/npm/v/ghc-proxy)](https://www.npmjs.com/package/ghc-proxy)
 [![CI](https://github.com/wxxb789/ghc-proxy/actions/workflows/ci.yml/badge.svg)](https://github.com/wxxb789/ghc-proxy/actions/workflows/ci.yml)
-[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://github.com/wxxb789/ghc-proxy/blob/master/LICENSE)
+[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://github.com/wxxb789/ghc-proxy/blob/main/LICENSE)
 
 A proxy that turns your GitHub Copilot subscription into an OpenAI and Anthropic compatible API. Use it to power [Claude Code](https://docs.anthropic.com/en/docs/claude-code/overview), [Cursor](https://www.cursor.com/), or any tool that speaks the OpenAI Chat Completions, OpenAI Responses, or Anthropic Messages protocol.
 
@@ -107,13 +107,13 @@ bunx ghc-proxy@latest selfcheck      # Probe tokenizer chunks and Bun/Node runti
 
 | Option | Alias | Default | Description |
 |--------|-------|---------|-------------|
-| `--port` | `-p` | `4141` | Port to listen on |
+| `--port` | `-p` | `4141` | Port to listen on (`1..65535`; malformed or zero values fail before startup) |
 | `--verbose` | `-v` | `false` | Enable verbose logging |
 | `--account-type` | `-a` | `individual` | `individual`, `business`, or `enterprise` |
 | `--rate-limit` | `-r` | -- | Minimum seconds between requests |
 | `--wait` | `-w` | `false` | Queue requests instead of rejecting with 429 when `--rate-limit` cooldown has not elapsed (requires `--rate-limit`) |
 | `--manual` | -- | `false` | Manually approve each request |
-| `--github-token` | `-g` | -- | Pass a GitHub token directly (from `auth`) |
+| `--github-token` | `-g` | -- | Use a GitHub token for this process only (normally obtained with `auth`); this flag does not persist it to `config.json` |
 | `--claude-code` | `-c` | `false` | Generate a Claude Code launch command |
 | `--show-token` | -- | `false` | Display tokens on auth and refresh |
 | `--dump-failed-payloads` | `-D` | `false` | Dump failed `/responses` payloads on upstream 400 errors for debugging. Can also be enabled with `DUMP_FAILED_PAYLOADS=1`. |
@@ -189,30 +189,31 @@ All fields are optional. The full schema:
 
 | Field | Type | Default | Description |
 |-------|------|---------|-------------|
-| `githubToken` | `string` | -- | Persisted GitHub token. Normally written automatically by `auth` / `--github-token`; you rarely set this by hand |
-| `modelRewrites` | `{ from, to }[]` | -- | Glob-pattern model substitution rules (see [Model Rewrites](#model-rewrites)) |
-| `modelFallback` | `object` | -- | Override default model fallbacks (see [Customizing Fallbacks](#customizing-fallbacks)) |
+| `githubToken` | `string` | unset | Persisted GitHub token. The device-code flow (`auth` or first startup) writes it automatically; `start --github-token` is runtime-only and does not write this field |
+| `modelRewrites` | `{ from, to }[]` | `[]` | Glob-pattern model substitution rules (see [Model Rewrites](#model-rewrites)) |
+| `modelFallback` | `object` | built-in family defaults | Override default model fallbacks (see [Customizing Fallbacks](#customizing-fallbacks)) |
 | `modelFallback.claudeOpus` | `string` | `claude-opus-5` | Fallback for `claude-opus-*` models |
 | `modelFallback.claudeSonnet` | `string` | `claude-sonnet-5` | Fallback for `claude-sonnet-*` models |
 | `modelFallback.claudeHaiku` | `string` | `claude-haiku-4.5` | Fallback for `claude-haiku-*` models |
-| `smallModel` | `string` | -- | Target model for compact request routing (see [Small-Model Routing](#small-model-routing)) |
+| `smallModel` | `string` | unset | Target model for compact request routing (see [Small-Model Routing](#small-model-routing)) |
 | `compactUseSmallModel` | `boolean` | `false` | Route compact/summarization requests to `smallModel` |
 | `useFunctionApplyPatch` | `boolean` | `true` | Rewrite `apply_patch` custom tool as function tool on Responses path |
 | `responsesApiAutoCompactInput` | `boolean` | `false` | Automatically trim Responses `input` to the latest `compaction` item |
 | `responsesApiAutoContextManagement` | `boolean` | `false` | Automatically inject Responses `context_management` for selected models |
-| `responsesApiContextManagementModels` | `string[]` | -- | Models eligible for auto-injected Responses `context_management` |
-| `responsesApiParameterFilters` | `{ models, params }[]` | -- | Extra rules to strip request parameters on the Responses boundary (see [Responses Parameter Filters](#responses-parameter-filters)) |
-| `responsesApiParameterFiltersReplaceDefault` | `boolean` | `false` | Disable the built-in reasoning-model default rule so only your `responsesApiParameterFilters` apply |
-| `responsesOfficialEmulator` | `boolean` | `false` | Enable local OpenAI-style Responses state emulation for `previous_response_id`, `conversation`, retrieve, input_items, delete, and input_tokens |
-| `responsesOfficialEmulatorTtlSeconds` | `number` | `14400` | In-memory TTL for locally emulated Responses state |
-| `modelReasoningEfforts` | `Record<string, string>` | -- | Per-model reasoning effort defaults for Anthropic-to-Responses translation. Each value must be one of `none`, `minimal`, `low`, `medium`, `high`, `xhigh`, or `max` (ascending) |
+| `responsesApiContextManagementModels` | `string[]` | `[]` | Models eligible for auto-injected Responses `context_management` |
+| `responsesApiParameterFilters` | `{ models, params }[]` | `[]` | Extra rules to strip request parameters on the Responses boundary; the built-in reasoning-model rule remains active unless replaced (see [Responses Parameter Filters](#responses-parameter-filters)) |
+| `responsesApiParameterFiltersReplaceDefault` | `boolean` | `false` | Disable the built-in reasoning-model default rule so only your `responsesApiParameterFilters` apply |
+| `chatCompletionsUseMaxCompletionTokens` | `string[]` | `[]` | Extra model globs that rename Chat Completions `max_tokens` to `max_completion_tokens`; adds to the built-in `gpt-5.4` / `gpt-5.4-*` rules |
+| `responsesOfficialEmulator` | `boolean` | `false` | Enable local OpenAI-style Responses state emulation for `previous_response_id`, `conversation`, retrieve, input_items, delete, and input_tokens |
+| `responsesOfficialEmulatorTtlSeconds` | `number` | `14400` | In-memory TTL for locally emulated Responses state |
+| `modelReasoningEfforts` | `Record<string, string>` | `{}`; unlisted models use `high` | Per-model reasoning effort defaults for Anthropic-to-Responses translation. Each value must be one of `none`, `minimal`, `low`, `medium`, `high`, `xhigh`, or `max` (ascending) |
 | `upstreamQueueConcurrency` | `number` | `10` | Maximum concurrent Copilot upstream requests |
 | `upstreamQueueMaxRetries` | `number` | `1` | Maximum retries across capacity and approved pre-connection failures (`0..2`) |
 | `upstreamRecoveryBudgetSeconds` | `number` | `60` | Shared recovery deadline after the first retryable outcome or active-cooldown encounter (`1..120` seconds) |
-| `overloadFallbacks` | `Record<string, string>` | -- | Exact effective-model mappings for one opt-in fallback dispatch after terminal model `529`; absent means disabled |
+| `overloadFallbacks` | `Record<string, string>` | `{}` (disabled) | Exact effective-model mappings for one opt-in fallback dispatch after terminal model `529` |
 | `upstreamQueueBaseDelaySeconds` | `number` | `2` | Base delay (seconds) for upstream retry backoff when `Retry-After` is absent |
 | `upstreamQueueMaxDelaySeconds` | `number` | `60` | Maximum computed backoff (seconds); does not clamp `Retry-After` |
-| `gheDomain` | `string` | -- | GitHub Enterprise Cloud company domain (persisted automatically after GHE.com auth) |
+| `gheDomain` | `string` | unset | GitHub Enterprise Cloud company domain (persisted automatically after GHE.com auth) |
 
 Example:
 
@@ -229,9 +230,10 @@ Example:
   "compactUseSmallModel": true,
   "useFunctionApplyPatch": true,
   "responsesApiAutoCompactInput": false,
-  "responsesApiAutoContextManagement": false,
-  "responsesApiContextManagementModels": ["gpt-5", "gpt-5-mini"],
-  "responsesOfficialEmulator": false,
+  "responsesApiAutoContextManagement": false,
+  "responsesApiContextManagementModels": ["gpt-5", "gpt-5-mini"],
+  "chatCompletionsUseMaxCompletionTokens": [],
+  "responsesOfficialEmulator": false,
   "responsesOfficialEmulatorTtlSeconds": 14400,
   "modelReasoningEfforts": {
     "gpt-5": "high",
@@ -359,9 +361,11 @@ The proxy authenticates with GitHub using the [device code OAuth flow](https://d
 
 When the Copilot token response includes `endpoints.api`, `ghc-proxy` now prefers that runtime API base automatically instead of relying only on the configured account type. This keeps enterprise/business routing aligned with the endpoint GitHub actually returned for the current token.
 
-Incoming requests hit an [Elysia](https://elysiajs.com/) server. `chat/completions` requests are validated, normalized into the shared planning pipeline, and then forwarded to Copilot. `responses` requests use a native Responses path with explicit compatibility policies. `messages` requests are routed per-model and can use native Anthropic passthrough, the Responses translation path, or the existing chat-completions fallback. The translator tracks exact vs lossy vs unsupported behavior explicitly; see the [Messages Routing and Translation Guide](./docs/messages-routing-and-translation.md) and the [Anthropic Translation Matrix](./docs/anthropic-translation-matrix.md) for the current support surface.
-
-For Anthropic `search_result` blocks, current live probes show Copilot native `/v1/messages` accepts top-level search results and pure search-result tool outputs, but rejects top-level `citations` and mixed text/search-result tool output arrays. The native path sanitizes those known rejection cases, while translated paths flatten search results to text.
+Incoming requests hit an [Elysia](https://elysiajs.com/) server. `chat/completions` requests are validated, normalized into the shared planning pipeline, and then forwarded to Copilot. `responses` requests use a native Responses path with explicit compatibility policies. `messages` requests are routed per-model and can use native Anthropic passthrough, the Responses translation path, or the existing chat-completions fallback. The translator tracks exact vs lossy vs unsupported behavior explicitly; see the [Messages Routing and Translation Guide](./docs/messages-routing-and-translation.md) and the [Anthropic Translation Matrix](./docs/anthropic-translation-matrix.md) for the current support surface.
+
+The built-in, read-only Dashboard projects process health, model routing, behavior, and recent request lifecycle metadata without storing request or response content. See [Dashboard Observability](./docs/design/dashboard-observability.md).
+
+For Anthropic `search_result` blocks, an April 17, 2026 probe against `claude-opus-4.6` on Copilot native `/v1/messages` accepted top-level search results and pure search-result tool outputs, but rejected top-level `citations` and mixed text/search-result tool output arrays. The native path sanitizes those observed rejection cases, while translated paths flatten search results to text; re-run the probe before treating that dated upstream result as universal.
 
 ### Request Routing
 
@@ -402,22 +406,37 @@ This keeps the existing chat pipeline stable while allowing newer Copilot models
 | `POST` | `/v1/messages` | Messages API with per-model routing across native Messages, Responses translation, or chat-completions fallback |
 | `POST` | `/v1/messages/count_tokens` | Token counting |
 
-**Utility:**
+**Utility:**
 
 | Method | Path | Description |
 |--------|------|-------------|
 | `GET`  | `/health` | Liveness/readiness probe — returns `{ status, copilotToken, modelsLoaded, version }` |
 | `GET`  | `/usage` | Copilot quota / usage monitoring |
-| `GET`  | `/token` | Inspect the current Copilot token |
-
-> **Note:** The `/v1/` prefix is optional for OpenAI-compatible endpoints (`/chat/completions`, `/responses`, `/models`, `/embeddings`). Anthropic endpoints (`/v1/messages`, `/v1/messages/count_tokens`) require the `/v1` prefix. The utility endpoints (`/health`, `/usage`, `/token`) are root-only and not exposed under `/v1`.
+| `GET`  | `/token` | Inspect the current Copilot token |
+
+**Local Dashboard (read-only):**
+
+| Method | Path | Description |
+|--------|------|-------------|
+| `GET` | `/dashboard` | Dashboard application |
+| `GET` | `/dashboard/styles.css` | Dashboard stylesheet |
+| `GET` | `/dashboard/app.js` | Dashboard client script |
+| `GET` | `/dashboard/api/overview` | Process, authentication, quota, request, and queue summary |
+| `GET` | `/dashboard/api/models` | Upstream model metadata and effective proxy capabilities |
+| `GET` | `/dashboard/api/behavior` | Active routing, compatibility policies, strategies, and effect counters |
+| `GET` | `/dashboard/api/requests` | Active requests and the most recent 256 completed request summaries |
+
+Dashboard routes are restricted to local access and return `403` when the peer, request host, or supplied `Origin` fails the loopback/same-origin checks. They are excluded from request history and access logging. See [Dashboard Observability](./docs/design/dashboard-observability.md) for the projection and security contract.
+
+> **Note:** The `/v1/` prefix is optional for OpenAI-compatible endpoints (`/chat/completions`, `/responses`, `/models`, `/embeddings`). Anthropic endpoints (`/v1/messages`, `/v1/messages/count_tokens`) require the `/v1` prefix. The utility and Dashboard endpoints are root-only and not exposed under `/v1`.
 
 ## Responses Compatibility
 
 `/v1/responses` is designed to stay close to the OpenAI wire format while making Copilot limitations explicit:
 
-- requests are validated before any mutation
-- common official request fields such as `conversation`, `previous_response_id`, `max_tool_calls`, `truncation`, `user`, `prompt`, and `text` are now modeled explicitly instead of relying on loose passthrough alone
+- requests are validated before any mutation
+- client-supplied `top_k` is rejected with `400` on the OpenAI Chat Completions and Responses boundaries because neither official OpenAI schema defines it; clients that send it by mistake receive an explicit error instead of a silent drop. Anthropic Messages `top_k` remains supported and is preserved when the proxy translates that request internally for Copilot
+- common official request fields such as `conversation`, `previous_response_id`, `max_tool_calls`, `truncation`, `user`, `prompt`, and `text` are now modeled explicitly instead of relying on loose passthrough alone
 - official `text.format` options are modeled explicitly, including `text`, `json_object`, and `json_schema`
 - an opt-in `responsesOfficialEmulator` mode adds in-memory OpenAI-style state for `previous_response_id`, `conversation`, `GET /responses/{id}`, `GET /responses/{id}/input_items`, `DELETE /responses/{id}`, and `POST /responses/input_tokens`
 - emulator state is memory-only and expires after `responsesOfficialEmulatorTtlSeconds` (default `14400`, or 4 hours)
@@ -427,9 +446,9 @@ This keeps the existing chat pipeline stable while allowing newer Copilot models
 - automatic trimming of Responses `input` to the latest `compaction` item is disabled by default and only applies when `responsesApiAutoCompactInput` is `true`
 - reasoning defaults for Anthropic -> Responses translation can be tuned with `modelReasoningEfforts`
 - request parameters that a model rejects (e.g. `temperature`/`top_p` on reasoning models) are stripped on the Responses boundary rather than leaked upstream as a `400`; see [Responses Parameter Filters](#responses-parameter-filters)
-- built-in web search (`web_search`, `web_search_preview`, and their dated variants) is forwarded to Copilot rather than blocked; every `/responses` model probed accepts it and runs a real search, see [docs/research/responses-web-search.md](docs/research/responses-web-search.md)
-- external image URLs on the Responses path fail explicitly with `400`; use `file_id` or data URL image input instead
-- official `input_file` and `item_reference` input items are modeled explicitly and validated before forwarding
+- built-in web search (`web_search`, `web_search_preview`, and their dated variants) is forwarded to Copilot rather than blocked; every `/responses` model reached by the August 4, 2026 acceptance sweep accepted the tool, while functional search execution was verified on `gpt-5.6-sol` and `gpt-5.6-terra`, see [docs/research/responses-web-search.md](docs/research/responses-web-search.md)
+- external image URLs on the Responses path fail explicitly with `400`; use `file_id` or data URL image input instead
+- official `input_file` and `item_reference` input items are modeled explicitly and validated, but the verified Copilot GPT Responses boundary is stateless: it rejects `store: true` and cannot resolve returned item IDs on later requests. The proxy deliberately applies a proxy-wide `store: false` policy, removes all `item_reference` items before dispatch, and removes `function_call_output` items whose `call_id` has no matching `function_call` in the same input array. Without the optional emulator, a caller that requested storage still receives a successful stateless response; retrieve/delete/continuation semantics are available only from the local emulator
 
 Example opt-in configuration for these two Responses-specific policies:
 
@@ -451,7 +470,7 @@ Some Copilot models reject request parameters that the OpenAI wire format allows
 
 This is expressed as a small rule engine that runs on both the native `/v1/responses` path and the `/v1/messages` → Responses translation path:
 
-- **Built-in default rule:** any model that advertises `reasoning_effort` has `temperature` and `top_p` stripped. This covers the whole reasoning family (including future point releases like `gpt-5.4-mini`) with no configuration.
+- **Built-in default rule:** any model that advertises `reasoning_effort` has `temperature` stripped. It also has `top_p` stripped except for `*-codex` / `*-codex-*` models, which are exempt because the July 26, 2026 probe found the tested Codex model accepted `top_p` while its reasoning-model siblings rejected it. This exemption narrows only the built-in rule; an operator rule can still strip `top_p`.
 - **`responsesApiParameterFilters`:** add your own rules. Each rule is `{ "models": [glob, ...], "params": [name, ...] }`; every rule whose `models` glob matches the resolved model contributes its `params`. Rules are **added** to the default (the union of parameters is stripped). Model globs use the same `*` wildcard as `modelRewrites`.
 - **`responsesApiParameterFiltersReplaceDefault`:** set to `true` to disable the built-in reasoning-model rule, so only your `responsesApiParameterFilters` apply — use this to fully **overwrite** the default behavior.
 
@@ -461,7 +480,7 @@ Stripped parameters are removed entirely (never sent as `null`), because upstrea
 {
   "responsesApiParameterFilters": [
     { "models": ["gpt-5*", "o1*"], "params": ["temperature", "top_p"] },
-    { "models": ["some-model"], "params": ["top_k"] }
+    { "models": ["some-model"], "params": ["service_tier"] }
   ],
   "responsesApiParameterFiltersReplaceDefault": false
 }
@@ -516,16 +535,19 @@ services:
 ```bash
 git clone https://github.com/wxxb789/ghc-proxy.git
 cd ghc-proxy
-bun install
-bun run dev
+bun install
+bun run dev              # Start with --watch
+# Or use the production-style source command:
+bun run start
 ```
 
 ## Development
 
 ```bash
-bun install              # Install dependencies
-bun run dev              # Start with --watch
-bun run build            # Build with tsdown
+bun install              # Install dependencies
+bun run dev              # Start with --watch
+bun run start            # Start without --watch
+bun run build            # Build with tsdown
 bun run lint             # ESLint
 bun run typecheck        # tsc --noEmit
 bun test                 # Run tests

package/dist/main.mjs

@@ -6735,14 +6735,27 @@ var FixedRing = class {
 		}
 		return values;
 	}
+	updateFirst(predicate, update) {
+		for (let offset = 0; offset < this.size; offset++) {
+			const index = (this.next - 1 - offset + this.entries.length) % this.entries.length;
+			const value = this.entries[index];
+			if (value !== void 0 && predicate(value)) {
+				this.entries[index] = update(value);
+				return true;
+			}
+		}
+		return false;
+	}
 };
 var RequestActivityStore = class {
 	active = /* @__PURE__ */ new Map();
 	recent = new FixedRing(256);
 	effectCounts = {};
+	effectBuffers = /* @__PURE__ */ new Map();
 	started = 0;
 	completed = 0;
 	failed = 0;
+	aborted = 0;
 	now;
 	constructor(now = Date.now) {
 		this.now = now;
@@ -6787,11 +6800,28 @@ var RequestActivityStore = class {
 		const request = this.active.get(requestId);
 		if (request) request.selectedStrategy = sanitizeTag(strategy);
 	}
+	beginEffectBuffer(requestId) {
+		if (this.active.has(requestId) && !this.effectBuffers.has(requestId)) this.effectBuffers.set(requestId, {});
+	}
+	commitEffectBuffer(requestId) {
+		const buffer = this.effectBuffers.get(requestId);
+		if (!buffer) return;
+		this.effectBuffers.delete(requestId);
+		for (const [effect, count] of Object.entries(buffer)) if (count) this.recordEffect(requestId, effect, count);
+	}
+	discardEffectBuffer(requestId) {
+		this.effectBuffers.delete(requestId);
+	}
 	recordEffect(requestId, effect, count = 1) {
 		if (!Number.isFinite(count) || count <= 0) return;
 		const request = this.active.get(requestId);
 		if (!request) return;
 		const normalizedCount = Math.floor(count);
+		const buffer = this.effectBuffers.get(requestId);
+		if (buffer) {
+			buffer[effect] = (buffer[effect] ?? 0) + normalizedCount;
+			return;
+		}
 		this.effectCounts[effect] = (this.effectCounts[effect] ?? 0) + normalizedCount;
 		request.effectCounts[effect] = (request.effectCounts[effect] ?? 0) + normalizedCount;
 	}
@@ -6801,16 +6831,33 @@ var RequestActivityStore = class {
 	}
 	markStreaming(requestId) {
 		const request = this.active.get(requestId);
-		if (request) request.state = "streaming";
+		if (request && request.state !== "aborted") request.state = "streaming";
+	}
+	markAborted(requestId) {
+		const request = this.active.get(requestId);
+		if (request) {
+			request.state = "aborted";
+			return;
+		}
+		if (this.recent.updateFirst((request) => request.requestId === requestId && request.state === "completed", (request) => ({
+			...request,
+			state: "aborted"
+		}))) {
+			this.completed--;
+			this.aborted++;
+		}
 	}
 	complete(requestId, status) {
 		const request = this.active.get(requestId);
 		if (!request) return false;
 		this.active.delete(requestId);
-		const failed = status >= 400 || request.errorSummary !== void 0;
-		const completedRequest = projectRequest(request, Math.max(0, this.now() - request.startedAtMs), failed ? "failed" : "completed", status);
+		this.effectBuffers.delete(requestId);
+		const aborted = request.state === "aborted" && request.errorSummary === void 0;
+		const failed = !aborted && (status >= 400 || request.errorSummary !== void 0);
+		const completedRequest = projectRequest(request, Math.max(0, this.now() - request.startedAtMs), aborted ? "aborted" : failed ? "failed" : "completed", status);
 		this.recent.push(completedRequest);
-		this.completed++;
+		if (aborted) this.aborted++;
+		else this.completed++;
 		if (failed) this.failed++;
 		return true;
 	}
@@ -6822,7 +6869,8 @@ var RequestActivityStore = class {
 			totals: {
 				started: this.started,
 				completed: this.completed,
-				failed: this.failed
+				failed: this.failed,
+				aborted: this.aborted
 			},
 			effectCounts: { ...this.effectCounts }
 		};
@@ -6834,18 +6882,21 @@ var RequestActivityStore = class {
 			totals: {
 				started: this.started,
 				completed: this.completed,
-				failed: this.failed
+				failed: this.failed,
+				aborted: this.aborted
 			},
 			effectCounts: { ...this.effectCounts }
 		};
 	}
 	reset() {
 		this.active.clear();
+		this.effectBuffers.clear();
 		this.recent = new FixedRing(256);
 		for (const key of Object.keys(this.effectCounts)) delete this.effectCounts[key];
 		this.started = 0;
 		this.completed = 0;
 		this.failed = 0;
+		this.aborted = 0;
 	}
 };
 function classifyObservedEndpoint(rawUrl) {
@@ -6920,14 +6971,21 @@ function readSafeErrorCategory(error) {
 		const body = "body" in error ? error.body : void 0;
 		if (typeof body !== "object" || body === null) return void 0;
 		const nested = "error" in body ? body.error : void 0;
-		let candidate;
-		if (typeof nested === "object" && nested !== null) candidate = "code" in nested ? nested.code : "type" in nested ? nested.type : void 0;
-		else if ("type" in body) candidate = body.type;
-		return typeof candidate === "string" && OBSERVED_ERROR_CATEGORIES.has(candidate) ? candidate : void 0;
+		if (typeof nested === "object" && nested !== null) {
+			const code = "code" in nested ? nested.code : void 0;
+			if (isObservedErrorCategory(code)) return code;
+			const type = "type" in nested ? nested.type : void 0;
+			return isObservedErrorCategory(type) ? type : void 0;
+		}
+		const type = "type" in body ? body.type : void 0;
+		return isObservedErrorCategory(type) ? type : void 0;
 	} catch {
 		return;
 	}
 }
+function isObservedErrorCategory(value) {
+	return typeof value === "string" && OBSERVED_ERROR_CATEGORIES.has(value);
+}
 //#endregion
 //#region src/state/runtime.ts
 var RuntimeStore = class {
@@ -7496,7 +7554,10 @@ var UpstreamRequestQueue = class {
 			let response;
 			try {
 				this.throwIfRecoveryExpired(recovery, lastConnectionError);
-				if (context.fallbackAttempt) recovery.fallbackFetchStarted = true;
+				if (context.fallbackAttempt) {
+					recovery.fallbackFetchStarted = true;
+					recovery.onFallbackFetchStart?.();
+				}
 				response = await this.fetchBeforeDeadline(fetcher, signal, recovery, lastConnectionError);
 				lastConnectionError = void 0;
 			} catch (error) {
@@ -8745,7 +8806,7 @@ const checkUsage = defineCommand({
 });
 //#endregion
 //#region src/util/version.ts
-const VERSION = "0.10.0";
+const VERSION = "0.10.1";
 //#endregion
 //#region src/debug.ts
 function getRuntimeInfo() {
@@ -48660,8 +48721,8 @@ async function getTokenCount(payload, model) {
 		output: outputTokens
 	};
 }
-async function estimateResponsesInputTokens(inputItems, model) {
-	return (await getEncoder(getTokenizerFromModel(model))).encode(JSON.stringify(inputItems)).length;
+async function estimateSerializedTokens(value, model) {
+	return (await getEncoder(getTokenizerFromModel(model))).encode(JSON.stringify(value)).length;
 }
 //#endregion
 //#region \0@oxc-project+runtime@0.134.0/helpers/esm/taggedTemplateLiteral.js
@@ -48671,7 +48732,7 @@ function _taggedTemplateLiteral(e, t) {
 //#endregion
 //#region src/routes/dashboard/assets.ts
 var _templateObject;
-const DASHBOARD_HTML = String.raw(_templateObject || (_templateObject = _taggedTemplateLiteral(["<!doctype html>\n<html lang=\"en\">\n<head>\n  <meta charset=\"utf-8\">\n  <meta name=\"viewport\" content=\"width=device-width, initial-scale=1\">\n  <title>ghc-proxy Dashboard</title>\n  <link rel=\"icon\" href=\"data:,\">\n  <link rel=\"stylesheet\" href=\"/dashboard/styles.css\">\n</head>\n<body>\n  <header class=\"app-header\">\n    <div class=\"brand-block\">\n      <div class=\"brand\">ghc-proxy</div>\n      <div class=\"status-line\">\n        <span id=\"health-dot\" class=\"status-dot unknown\"></span>\n        <span id=\"health-label\">Connecting</span>\n        <span id=\"version-label\"></span>\n      </div>\n    </div>\n    <nav class=\"tabs\" aria-label=\"Dashboard views\">\n      <button type=\"button\" class=\"tab active\" data-tab=\"overview\">Overview</button>\n      <button type=\"button\" class=\"tab\" data-tab=\"models\">Models</button>\n      <button type=\"button\" class=\"tab\" data-tab=\"behavior\">Behavior</button>\n      <button type=\"button\" class=\"tab\" data-tab=\"requests\">Requests</button>\n    </nav>\n    <div class=\"header-actions\">\n      <label class=\"toggle\"><input id=\"live-refresh\" type=\"checkbox\" checked> Live</label>\n      <label class=\"theme-toggle\">\n        <input id=\"theme-toggle\" type=\"checkbox\" role=\"switch\" aria-label=\"Use dark theme\">\n        <span id=\"theme-label\">Light</span>\n      </label>\n      <button id=\"refresh-button\" type=\"button\" class=\"command\">Refresh</button>\n    </div>\n  </header>\n\n  <main>\n    <section id=\"view-overview\" class=\"view active\" data-view=\"overview\">\n      <div class=\"metric-strip\" aria-label=\"Runtime summary\">\n        <div><span class=\"metric-label\">Uptime</span><strong id=\"metric-uptime\">-</strong></div>\n        <div><span class=\"metric-label\">Active</span><strong id=\"metric-active\">0</strong></div>\n        <div><span class=\"metric-label\">Completed</span><strong id=\"metric-completed\">0</strong></div>\n        <div><span class=\"metric-label\">Failed</span><strong id=\"metric-failed\">0</strong></div>\n        <div><span class=\"metric-label\">Queue</span><strong id=\"metric-queue\">0 / 0</strong></div>\n      </div>\n\n      <div class=\"overview-grid\">\n        <section class=\"panel\">\n          <div class=\"section-heading\"><h1>Runtime</h1><span id=\"overview-updated\" class=\"muted\"></span></div>\n          <dl id=\"runtime-details\" class=\"kv-list\"></dl>\n        </section>\n        <section class=\"panel\">\n          <div class=\"section-heading\"><h1>Authentication</h1></div>\n          <div class=\"table-scroll\"><table><thead><tr><th>Service</th><th>Status</th><th>Identity</th><th>Last check</th></tr></thead><tbody id=\"auth-body\"></tbody></table></div>\n        </section>\n      </div>\n\n      <section class=\"panel full-width\">\n        <div class=\"section-heading\"><h1>Quota</h1><span id=\"quota-status\" class=\"muted\"></span></div>\n        <div class=\"table-scroll\"><table><thead><tr><th>Pool</th><th>Remaining</th><th>Entitlement</th><th>Percent</th><th>Overage</th></tr></thead><tbody id=\"quota-body\"></tbody></table></div>\n      </section>\n    </section>\n\n    <section id=\"view-models\" class=\"view\" data-view=\"models\" hidden>\n      <div class=\"toolbar model-toolbar\">\n        <div><h1>Models</h1><span id=\"model-count\" class=\"muted\"></span></div>\n        <div class=\"model-controls\">\n          <label class=\"toggle\"><input id=\"model-group-vendor\" type=\"checkbox\" checked> Group vendor</label>\n          <select id=\"model-sort\" aria-label=\"Order models by name\">\n            <option value=\"asc\">Name A-Z</option>\n            <option value=\"desc\">Name Z-A</option>\n          </select>\n          <input id=\"model-filter\" type=\"search\" placeholder=\"Filter models\" autocomplete=\"off\">\n          <button id=\"copy-models\" type=\"button\" class=\"command\" disabled>Copy selected (0)</button>\n        </div>\n      </div>\n      <div id=\"model-copy-status\" class=\"muted model-copy-status\" role=\"status\" aria-live=\"polite\"></div>\n      <div class=\"table-scroll model-table\"><table><thead><tr><th class=\"model-select-header\" aria-label=\"Select model\"></th><th>Model</th><th>Vendor</th><th>Messages route</th><th>Upstream endpoints</th><th>Limits</th><th>Capabilities</th><th>Proxy compatibility</th></tr></thead><tbody id=\"models-body\"></tbody></table></div>\n    </section>\n\n    <section id=\"view-behavior\" class=\"view\" data-view=\"behavior\" hidden>\n      <div class=\"toolbar\"><div><h1>Behavior</h1><span class=\"muted\">Current process configuration</span></div></div>\n      <div class=\"behavior-grid\">\n        <section class=\"panel\"><div class=\"section-heading\"><h2>Model routing</h2></div><dl id=\"behavior-routing\" class=\"kv-list\"></dl></section>\n        <section class=\"panel\"><div class=\"section-heading\"><h2>Parameters and context</h2></div><dl id=\"behavior-parameters\" class=\"kv-list\"></dl></section>\n      </div>\n      <section class=\"panel full-width\">\n        <div class=\"section-heading\"><h2>Effects since startup</h2></div>\n        <div class=\"table-scroll\"><table><thead><tr><th>Category</th><th>Effect</th><th>Count</th></tr></thead><tbody id=\"effects-body\"></tbody></table></div>\n      </section>\n    </section>\n\n    <section id=\"view-requests\" class=\"view requests-view\" data-view=\"requests\" hidden>\n      <div class=\"toolbar\"><div><h1>Requests</h1><span id=\"request-count\" class=\"muted\"></span></div><span class=\"muted\">256 completed max</span></div>\n      <div class=\"request-layout\">\n        <div class=\"request-list table-scroll\"><table><thead><tr><th>State</th><th>Endpoint</th><th>Model</th><th>Strategy</th><th>Status</th><th>Duration</th><th>Started</th></tr></thead><tbody id=\"requests-body\"></tbody></table></div>\n        <aside class=\"request-detail\" aria-label=\"Selected request details\">\n          <div class=\"section-heading\"><h2>Request detail</h2></div>\n          <pre id=\"request-detail\">No request selected</pre>\n        </aside>\n      </div>\n    </section>\n  </main>\n\n  <div id=\"error-banner\" class=\"error-banner\" role=\"status\" hidden></div>\n  <script src=\"/dashboard/app.js\" defer><\/script>\n</body>\n</html>"])));
+const DASHBOARD_HTML = String.raw(_templateObject || (_templateObject = _taggedTemplateLiteral(["<!doctype html>\n<html lang=\"en\">\n<head>\n  <meta charset=\"utf-8\">\n  <meta name=\"viewport\" content=\"width=device-width, initial-scale=1\">\n  <title>ghc-proxy Dashboard</title>\n  <link rel=\"icon\" href=\"data:,\">\n  <link rel=\"stylesheet\" href=\"/dashboard/styles.css\">\n</head>\n<body>\n  <header class=\"app-header\">\n    <div class=\"brand-block\">\n      <div class=\"brand\">ghc-proxy</div>\n      <div class=\"status-line\">\n        <span id=\"health-dot\" class=\"status-dot unknown\"></span>\n        <span id=\"health-label\">Connecting</span>\n        <span id=\"version-label\"></span>\n      </div>\n    </div>\n    <nav class=\"tabs\" aria-label=\"Dashboard views\">\n      <button type=\"button\" class=\"tab active\" data-tab=\"overview\">Overview</button>\n      <button type=\"button\" class=\"tab\" data-tab=\"models\">Models</button>\n      <button type=\"button\" class=\"tab\" data-tab=\"behavior\">Behavior</button>\n      <button type=\"button\" class=\"tab\" data-tab=\"requests\">Requests</button>\n    </nav>\n    <div class=\"header-actions\">\n      <label class=\"toggle\"><input id=\"live-refresh\" type=\"checkbox\" checked> Live</label>\n      <label class=\"theme-toggle\">\n        <input id=\"theme-toggle\" type=\"checkbox\" role=\"switch\" aria-label=\"Use dark theme\">\n        <span id=\"theme-label\">Light</span>\n      </label>\n      <button id=\"refresh-button\" type=\"button\" class=\"command\">Refresh</button>\n    </div>\n  </header>\n\n  <main>\n    <section id=\"view-overview\" class=\"view active\" data-view=\"overview\">\n      <div class=\"metric-strip\" aria-label=\"Runtime summary\">\n        <div><span class=\"metric-label\">Uptime</span><strong id=\"metric-uptime\">-</strong></div>\n        <div><span class=\"metric-label\">Active</span><strong id=\"metric-active\">0</strong></div>\n        <div><span class=\"metric-label\">Completed</span><strong id=\"metric-completed\">0</strong></div>\n        <div><span class=\"metric-label\">Failed</span><strong id=\"metric-failed\">0</strong></div>\n        <div><span class=\"metric-label\">Aborted</span><strong id=\"metric-aborted\">0</strong></div>\n        <div><span class=\"metric-label\">Queue</span><strong id=\"metric-queue\">0 / 0</strong></div>\n      </div>\n\n      <div class=\"overview-grid\">\n        <section class=\"panel\">\n          <div class=\"section-heading\"><h1>Runtime</h1><span id=\"overview-updated\" class=\"muted\"></span></div>\n          <dl id=\"runtime-details\" class=\"kv-list\"></dl>\n        </section>\n        <section class=\"panel\">\n          <div class=\"section-heading\"><h1>Authentication</h1></div>\n          <div class=\"table-scroll\"><table><thead><tr><th>Service</th><th>Status</th><th>Identity</th><th>Last check</th></tr></thead><tbody id=\"auth-body\"></tbody></table></div>\n        </section>\n      </div>\n\n      <section class=\"panel full-width\">\n        <div class=\"section-heading\"><h1>Quota</h1><span id=\"quota-status\" class=\"muted\"></span></div>\n        <div class=\"table-scroll\"><table><thead><tr><th>Pool</th><th>Remaining</th><th>Entitlement</th><th>Percent</th><th>Overage</th></tr></thead><tbody id=\"quota-body\"></tbody></table></div>\n      </section>\n    </section>\n\n    <section id=\"view-models\" class=\"view\" data-view=\"models\" hidden>\n      <div class=\"toolbar model-toolbar\">\n        <div><h1>Models</h1><span id=\"model-count\" class=\"muted\"></span></div>\n        <div class=\"model-controls\">\n          <label class=\"toggle\"><input id=\"model-group-vendor\" type=\"checkbox\" checked> Group vendor</label>\n          <select id=\"model-sort\" aria-label=\"Order models by name\">\n            <option value=\"asc\">Name A-Z</option>\n            <option value=\"desc\">Name Z-A</option>\n          </select>\n          <input id=\"model-filter\" type=\"search\" placeholder=\"Filter models\" autocomplete=\"off\">\n          <button id=\"copy-models\" type=\"button\" class=\"command\" disabled>Copy selected (0)</button>\n        </div>\n      </div>\n      <div id=\"model-copy-status\" class=\"muted model-copy-status\" role=\"status\" aria-live=\"polite\"></div>\n      <div class=\"table-scroll model-table\"><table><thead><tr><th class=\"model-select-header\" aria-label=\"Select model\"></th><th>Model</th><th>Vendor</th><th>Messages route</th><th>Upstream endpoints</th><th>Limits</th><th>Capabilities</th><th>Proxy compatibility</th></tr></thead><tbody id=\"models-body\"></tbody></table></div>\n    </section>\n\n    <section id=\"view-behavior\" class=\"view\" data-view=\"behavior\" hidden>\n      <div class=\"toolbar\"><div><h1>Behavior</h1><span class=\"muted\">Current process configuration</span></div></div>\n      <div class=\"behavior-grid\">\n        <section class=\"panel\"><div class=\"section-heading\"><h2>Model routing</h2></div><dl id=\"behavior-routing\" class=\"kv-list\"></dl></section>\n        <section class=\"panel\"><div class=\"section-heading\"><h2>Parameters and context</h2></div><dl id=\"behavior-parameters\" class=\"kv-list\"></dl></section>\n      </div>\n      <section class=\"panel full-width\">\n        <div class=\"section-heading\"><h2>Effects since startup</h2></div>\n        <div class=\"table-scroll\"><table><thead><tr><th>Category</th><th>Effect</th><th>Count</th></tr></thead><tbody id=\"effects-body\"></tbody></table></div>\n      </section>\n    </section>\n\n    <section id=\"view-requests\" class=\"view requests-view\" data-view=\"requests\" hidden>\n      <div class=\"toolbar\"><div><h1>Requests</h1><span id=\"request-count\" class=\"muted\"></span></div><span class=\"muted\">256 finished max</span></div>\n      <div class=\"request-layout\">\n        <div class=\"request-list table-scroll\"><table><thead><tr><th>State</th><th>Endpoint</th><th>Model</th><th>Strategy</th><th>Status</th><th>Duration</th><th>Started</th></tr></thead><tbody id=\"requests-body\"></tbody></table></div>\n        <aside class=\"request-detail\" aria-label=\"Selected request details\">\n          <div class=\"section-heading\"><h2>Request detail</h2></div>\n          <pre id=\"request-detail\">No request selected</pre>\n        </aside>\n      </div>\n    </section>\n  </main>\n\n  <div id=\"error-banner\" class=\"error-banner\" role=\"status\" hidden></div>\n  <script src=\"/dashboard/app.js\" defer><\/script>\n</body>\n</html>"])));
 const DASHBOARD_CSS = String.raw`:root {
   color-scheme: light;
   font-family: Inter, ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
@@ -48804,7 +48865,7 @@ h2 { font-size: 15px; }
 
 .metric-strip {
   display: grid;
-  grid-template-columns: repeat(5, minmax(110px, 1fr));
+  grid-template-columns: repeat(6, minmax(100px, 1fr));
   margin-bottom: 18px;
   border: 1px solid var(--border);
   background: var(--surface);
@@ -48848,7 +48909,7 @@ tbody tr:hover td { background: var(--surface-hover); }
 .mono { font-family: ui-monospace, SFMono-Regular, Consolas, "Liberation Mono", monospace; font-size: 12px; overflow-wrap: anywhere; }
 .badge { display: inline-flex; align-items: center; min-height: 22px; padding: 2px 7px; border: 1px solid var(--border); border-radius: 4px; background: var(--surface-subtle); color: var(--text-secondary); font-size: 11px; white-space: nowrap; }
 .badge.ok, .badge.completed { color: var(--ok); border-color: var(--ok-border); background: var(--ok-bg); }
-.badge.degraded, .badge.streaming, .badge.in_flight { color: var(--warn); border-color: var(--warn-border); background: var(--warn-bg); }
+.badge.degraded, .badge.streaming, .badge.in_flight, .badge.aborted { color: var(--warn); border-color: var(--warn-border); background: var(--warn-bg); }
 .badge.failed, .badge.missing { color: var(--bad); border-color: var(--bad-border); background: var(--bad-bg); }
 
 .behavior-grid { margin-bottom: 18px; }
@@ -48870,7 +48931,7 @@ tbody tr:hover td { background: var(--surface-hover); }
   .metric-strip { grid-template-columns: repeat(2, minmax(0, 1fr)); }
   .metric-strip > div { border-bottom: 1px solid var(--border); }
   .metric-strip > div:nth-child(2n) { border-right: 0; }
-  .metric-strip > div:last-child { grid-column: 1 / -1; border-bottom: 0; }
+  .metric-strip > div:nth-last-child(-n + 2) { border-bottom: 0; }
   .overview-grid, .behavior-grid, .request-layout { grid-template-columns: 1fr; }
   .request-list, .request-detail pre { max-height: none; }
 }
@@ -49017,6 +49078,7 @@ function renderOverview(data) {
   byId('metric-active').textContent = formatNumber(data.activity.activeRequests);
   byId('metric-completed').textContent = formatNumber(data.activity.completed);
   byId('metric-failed').textContent = formatNumber(data.activity.failed);
+  byId('metric-aborted').textContent = formatNumber(data.activity.aborted);
   const queue = data.activity.upstreamQueue || {};
   byId('metric-queue').textContent = formatNumber(queue.active) + ' / ' + formatNumber(queue.pending);
   byId('overview-updated').textContent = 'Updated ' + new Date().toLocaleTimeString();
@@ -49024,7 +49086,7 @@ function renderOverview(data) {
   const runtime = byId('runtime-details');
   clearNode(runtime);
   appendKv(runtime, 'Started', formatDate(data.startedAt));
-  appendKv(runtime, 'Recent completed', data.activity.recentRequests);
+  appendKv(runtime, 'Recent finished', data.activity.recentRequests);
   appendKv(runtime, 'Upstream slots', formatNumber(queue.active) + ' / ' + formatNumber(queue.concurrency));
   appendKv(runtime, 'Pending queue', formatNumber(queue.pending) + ' / ' + formatNumber(queue.maxPending));
   appendKv(runtime, 'Cooldowns', (queue.accountCooldown ? 'account ' : '') + formatNumber(queue.modelCooldowns) + ' model');
@@ -49298,7 +49360,7 @@ function renderRequests(data) {
       ? dashboardState.requests[0].requestId
       : null;
   }
-  byId('request-count').textContent = active.length + ' active / ' + recent.length + ' completed';
+  byId('request-count').textContent = active.length + ' active / ' + recent.length + ' finished';
   const body = byId('requests-body');
   clearNode(body);
   dashboardState.requests.forEach(function (request) {
@@ -50140,6 +50202,11 @@ function formatSearchResultBlock(block) {
 	].filter((part) => Boolean(part)).join("\n");
 }
 //#endregion
+//#region src/translator/anthropic/types.ts
+function isAnthropicBuiltinTool(tool) {
+	return typeof tool.type === "string" && tool.type !== "custom";
+}
+//#endregion
 //#region src/translator/anthropic/anthropic-normalizer.ts
 function textBlock(text) {
 	return {
@@ -50280,11 +50347,11 @@ function normalizeAnthropicRequest(payload) {
 		topP: payload.top_p,
 		topK: payload.top_k,
 		userId: payload.metadata?.user_id,
-		tools: payload.tools?.map((tool) => ({
+		tools: payload.tools?.flatMap((tool) => !isAnthropicBuiltinTool(tool) && tool.name !== void 0 && tool.input_schema !== void 0 ? [{
 			name: tool.name,
 			description: tool.description,
 			inputSchema: tool.input_schema
-		})),
+		}] : []),
 		toolChoice: normalizeToolChoice(payload.tool_choice),
 		thinking: normalizeThinking(payload.thinking),
 		outputEffort: payload.output_config?.effort ?? void 0,
@@ -50757,11 +50824,6 @@ function toConversationTurn(turn) {
 	};
 }
 function recordAnthropicRequestIssues(request, context) {
-	if (request.serviceTier !== void 0) context.record({
-		kind: "unsupported_service_tier",
-		severity: "warning",
-		message: "Anthropic service_tier is not supported by the upstream Copilot CAPI payload and was dropped."
-	}, { fatalInStrict: true });
 	for (const turn of request.turns) {
 		if (turn.role === "user") {
 			for (const block of turn.blocks) if (block.kind === "tool_result" && block.isError) context.record({
@@ -50811,6 +50873,15 @@ function applyThinkingBudgetOverride(plan, request, options) {
 }
 function normalizeAnthropicConversation(payload, policy) {
 	const context = new TranslationContext(policy);
+	if (payload.service_tier !== void 0) throw new TranslationFailure("Anthropic service_tier cannot be translated to Chat Completions.", {
+		status: 400,
+		kind: "unsupported_service_tier"
+	});
+	const builtinTool = payload.tools?.find(isAnthropicBuiltinTool);
+	if (builtinTool) throw new TranslationFailure(`Anthropic built-in/toolset "${builtinTool.type ?? builtinTool.name ?? "unknown"}" cannot be translated to Chat Completions.`, {
+		status: 400,
+		kind: "unsupported_server_tool"
+	});
 	const normalized = normalizeAnthropicRequest(payload);
 	recordAnthropicRequestIssues(normalized, context);
 	return {
@@ -52249,12 +52320,22 @@ function translateSystemPrompt(system) {
 }
 function convertAnthropicTools(tools) {
 	if (!tools || tools.length === 0) return null;
-	return tools.map((tool) => ({
-		type: "function",
-		name: tool.name,
-		parameters: normalizeFunctionParametersSchemaForCopilot(tool.input_schema),
-		...tool.description ? { description: tool.description } : {}
-	}));
+	return tools.map((tool) => {
+		if (isAnthropicBuiltinTool(tool)) throw new TranslationFailure(`Anthropic built-in/toolset "${tool.type ?? tool.name ?? "unknown"}" cannot be translated to the Responses API.`, {
+			status: 400,
+			kind: "unsupported_server_tool"
+		});
+		if (tool.name === void 0 || tool.input_schema === void 0) throw new TranslationFailure("Anthropic function tools require both name and input_schema.", {
+			status: 400,
+			kind: "invalid_tool_schema"
+		});
+		return {
+			type: "function",
+			name: tool.name,
+			parameters: normalizeFunctionParametersSchemaForCopilot(tool.input_schema),
+			...tool.description ? { description: tool.description } : {}
+		};
+	});
 }
 function convertAnthropicToolChoice(choice) {
 	if (!choice) return "auto";
@@ -54356,6 +54437,69 @@ var cors = (config) => {
 	});
 };
 //#endregion
+//#region src/lib/upstream-signal.ts
+const DEFAULT_TIMEOUT_MS = 18e5;
+const CLIENT_ABORT_MARKER = "ghcProxyClientAbort";
+var ClientAbortError = class extends DOMException {
+	[CLIENT_ABORT_MARKER] = true;
+	constructor() {
+		super("The client aborted the request.", "AbortError");
+	}
+};
+function isClientAbortError(error) {
+	return errorCauseChainSome(error, (candidate) => candidate[CLIENT_ABORT_MARKER] === true);
+}
+function createUpstreamDeadlineFromConfig(now = performance.now()) {
+	const timeoutMs = authStore.upstreamTimeoutSeconds !== void 0 ? authStore.upstreamTimeoutSeconds * 1e3 : DEFAULT_TIMEOUT_MS;
+	return timeoutMs > 0 ? now + timeoutMs : null;
+}
+function createUpstreamSignal(clientSignal, timeoutMs = DEFAULT_TIMEOUT_MS, onClientAbort) {
+	const controller = new AbortController();
+	const timeout = timeoutMs > 0 ? setTimeout(() => controller.abort(), timeoutMs) : void 0;
+	const onAbort = () => {
+		onClientAbort?.();
+		controller.abort(new ClientAbortError());
+	};
+	if (clientSignal && !clientSignal.aborted) clientSignal.addEventListener("abort", onAbort);
+	return {
+		signal: controller.signal,
+		clientSignal,
+		cleanup: () => {
+			if (timeout) clearTimeout(timeout);
+			clientSignal?.removeEventListener("abort", onAbort);
+		}
+	};
+}
+/**
+* Convenience wrapper that reads the upstream timeout from runtime config.
+*
+* This signal is a *total-duration* limit. Both runtimes separately apply an
+* ~300s **idle** timeout to `fetch` — Bun's is built in, Node's is undici's
+* `headersTimeout` / `bodyTimeout` default of `300e3` — which resets on every
+* byte received. A response that keeps streaming therefore runs past 300s and
+* is bounded only by this signal; a stalled one is rejected at ~300s by the
+* runtime instead. `isTimeoutLikeError` recognizes both runtimes' shapes so
+* every path maps to a 504.
+*/
+function createUpstreamSignalFromConfig(clientSignal, deadlineMonotonicMs = createUpstreamDeadlineFromConfig(), onClientAbort) {
+	const remainingMs = deadlineMonotonicMs === null ? void 0 : deadlineMonotonicMs - performance.now();
+	return {
+		...remainingMs !== void 0 && remainingMs <= 0 ? createExpiredUpstreamSignal(clientSignal, onClientAbort) : createUpstreamSignal(clientSignal, remainingMs ?? 0, onClientAbort),
+		deadlineMonotonicMs
+	};
+}
+function createExpiredUpstreamSignal(clientSignal, onClientAbort) {
+	const controller = new AbortController();
+	controller.abort();
+	const onAbort = () => onClientAbort?.();
+	if (!clientSignal.aborted) clientSignal.addEventListener("abort", onAbort);
+	return {
+		signal: controller.signal,
+		clientSignal,
+		cleanup: () => clientSignal.removeEventListener("abort", onAbort)
+	};
+}
+//#endregion
 //#region src/deliver/index.ts
 function deliverResult(request, result, modelMapping) {
 	setRequestModelMapping(request, modelMapping);
@@ -54621,10 +54765,23 @@ const anthropicMessageSchema = union([
 	}).loose()
 ]);
 const anthropicToolSchema = object({
-	name: string().min(1),
+	type: string().min(1).nullable().optional(),
+	name: string().min(1).optional(),
 	description: string().optional(),
-	input_schema: createObjectSchemaDefinitionSchema("tool input_schema must describe an object")
-}).loose();
+	input_schema: createObjectSchemaDefinitionSchema("tool input_schema must describe an object").optional()
+}).loose().superRefine((tool, ctx) => {
+	const isBuiltinTool = isAnthropicBuiltinTool(tool);
+	if (!isBuiltinTool && tool.input_schema === void 0) ctx.addIssue({
+		code: "custom",
+		message: "tool input_schema must describe an object",
+		path: ["input_schema"]
+	});
+	if (!isBuiltinTool && tool.name === void 0) ctx.addIssue({
+		code: "custom",
+		message: "function tools require a name",
+		path: ["name"]
+	});
+});
 const anthropicToolChoiceSchema = union([
 	object({ type: literal("auto") }).loose(),
 	object({ type: literal("any") }).loose(),
@@ -54831,6 +54988,11 @@ const openAIChatPayloadSchema = object({
 	reasoning_effort: _enum(REASONING_EFFORT_VALUES).nullable().optional(),
 	thinking_budget: number().int().positive().nullable().optional()
 }).loose().superRefine((payload, ctx) => {
+	if (Object.hasOwn(payload, "top_k")) ctx.addIssue({
+		code: "custom",
+		message: "top_k is not supported by the OpenAI Chat Completions API",
+		path: ["top_k"]
+	});
 	const toolChoice = payload.tool_choice;
 	if (toolChoice && typeof toolChoice === "object" && "function" in toolChoice && !payload.tools?.some((tool) => tool.function.name === toolChoice.function.name)) ctx.addIssue({
 		code: "custom",
@@ -55135,6 +55297,11 @@ function createResponsesPayloadSchema(options) {
 			"priority"
 		]).nullable().optional()
 	}).loose().superRefine((payload, ctx) => {
+		if (Object.hasOwn(payload, "top_k")) ctx.addIssue({
+			code: "custom",
+			message: "top_k is not supported by the OpenAI Responses API",
+			path: ["top_k"]
+		});
 		if (payload.previous_response_id && payload.conversation) ctx.addIssue({
 			code: "custom",
 			message: "previous_response_id cannot be used together with conversation",
@@ -55266,54 +55433,6 @@ protocolRegistry.register("responses", responsesProtocol);
 protocolRegistry.register("responses-input-tokens", responsesInputTokensProtocol);
 protocolRegistry.register("embeddings", embeddingsProtocol);
 //#endregion
-//#region src/lib/upstream-signal.ts
-const DEFAULT_TIMEOUT_MS = 18e5;
-function createUpstreamDeadlineFromConfig(now = performance.now()) {
-	const timeoutMs = authStore.upstreamTimeoutSeconds !== void 0 ? authStore.upstreamTimeoutSeconds * 1e3 : DEFAULT_TIMEOUT_MS;
-	return timeoutMs > 0 ? now + timeoutMs : null;
-}
-function createUpstreamSignal(clientSignal, timeoutMs = DEFAULT_TIMEOUT_MS) {
-	const controller = new AbortController();
-	const timeout = timeoutMs > 0 ? setTimeout(() => controller.abort(), timeoutMs) : void 0;
-	const onAbort = () => controller.abort();
-	if (clientSignal && !clientSignal.aborted) clientSignal.addEventListener("abort", onAbort);
-	return {
-		signal: controller.signal,
-		clientSignal,
-		cleanup: () => {
-			if (timeout) clearTimeout(timeout);
-			clientSignal?.removeEventListener("abort", onAbort);
-		}
-	};
-}
-/**
-* Convenience wrapper that reads the upstream timeout from runtime config.
-*
-* This signal is a *total-duration* limit. Both runtimes separately apply an
-* ~300s **idle** timeout to `fetch` — Bun's is built in, Node's is undici's
-* `headersTimeout` / `bodyTimeout` default of `300e3` — which resets on every
-* byte received. A response that keeps streaming therefore runs past 300s and
-* is bounded only by this signal; a stalled one is rejected at ~300s by the
-* runtime instead. `isTimeoutLikeError` recognizes both runtimes' shapes so
-* every path maps to a 504.
-*/
-function createUpstreamSignalFromConfig(clientSignal, deadlineMonotonicMs = createUpstreamDeadlineFromConfig()) {
-	const remainingMs = deadlineMonotonicMs === null ? void 0 : deadlineMonotonicMs - performance.now();
-	return {
-		...remainingMs !== void 0 && remainingMs <= 0 ? createExpiredUpstreamSignal(clientSignal) : createUpstreamSignal(clientSignal, remainingMs ?? 0),
-		deadlineMonotonicMs
-	};
-}
-function createExpiredUpstreamSignal(clientSignal) {
-	const controller = new AbortController();
-	controller.abort();
-	return {
-		signal: controller.signal,
-		clientSignal,
-		cleanup: () => {}
-	};
-}
-//#endregion
 //#region src/pipeline/runner.ts
 async function runPipeline(params, config) {
 	const upstreamDeadlineMonotonicMs = createUpstreamDeadlineFromConfig();
@@ -55329,6 +55448,7 @@ async function runPipeline(params, config) {
 	const fallbackPossible = shouldPreservePristinePayload(config.protocol, baseSourceModel);
 	const pristinePayload = fallbackPossible ? structuredClone(payload) : payload;
 	const sourceAttempt = await prepareAttempt(payload, meta, params, config, recovery, upstreamDeadlineMonotonicMs, { offerLocalModelCooldown: (sourceModel) => fallbackPossible && validateFallback(config.protocol, pristinePayload, sourceModel).ok });
+	sourceAttempt.commitObservability();
 	try {
 		return {
 			result: await sourceAttempt.execute(),
@@ -55375,14 +55495,18 @@ async function runPipeline(params, config) {
 		}
 		error.recovery.retryLimit = error.recovery.retryCount;
 		emitFallbackEvent(error.recovery, candidate.target.id, "selected", 529);
+		error.recovery.onFallbackFetchStart = fallbackAttempt.commitObservability;
 		try {
 			const result = await fallbackAttempt.execute();
+			fallbackAttempt.commitObservability();
 			emitFallbackEvent(error.recovery, candidate.target.id, "succeeded");
 			return {
 				result: discloseActualModel(result, candidate.target.id),
 				modelMapping: fallbackMapping
 			};
 		} catch (fallbackError) {
+			if (error.recovery.fallbackFetchStarted) fallbackAttempt.commitObservability();
+			else fallbackAttempt.discardObservability();
 			if (params.signal.aborted) throw params.signal.reason;
 			if (fallbackError instanceof FallbackCooldownError) {
 				emitFallbackEvent(error.recovery, candidate.target.id, "target-cooldown", 529);
@@ -55394,6 +55518,8 @@ async function runPipeline(params, config) {
 			}
 			emitFallbackEvent(error.recovery, candidate.target.id, "target-failed", fallbackError instanceof HTTPError ? fallbackError.status : void 0, isRetryableConnectionEstablishmentError(fallbackError));
 			throw fallbackError;
+		} finally {
+			delete error.recovery.onFallbackFetchStart;
 		}
 	}
 }
@@ -55406,17 +55532,24 @@ async function prepareAttempt(payload, meta, params, config, recovery, upstreamD
 	const baseModel = resolved.model;
 	const selectedModel = options.target ?? resolved.resolvedModel;
 	const modelMapping = options.modelMapping ?? resolved.modelMapping;
-	runtimeStore.requests.recordModelMapping(params.requestId, modelMapping);
+	const deferObservability = options.fallbackAttempt === true;
+	if (deferObservability) runtimeStore.requests.beginEffectBuffer(params.requestId);
+	else runtimeStore.requests.recordModelMapping(params.requestId, modelMapping);
 	const recordedModelStepCount = modelMapping.steps.length;
-	if (options.target) payload.model = options.target.id;
-	if (config.afterTransform) await config.afterTransform({
-		payload,
-		meta,
-		headers: params.headers,
-		selectedModel
-	});
+	try {
+		if (options.target) payload.model = options.target.id;
+		if (config.afterTransform) await config.afterTransform({
+			payload,
+			meta,
+			headers: params.headers,
+			selectedModel
+		});
+	} catch (error) {
+		if (deferObservability) runtimeStore.requests.discardEffectBuffer(params.requestId);
+		throw error;
+	}
 	params.signal.throwIfAborted();
-	const upstreamSignal = createUpstreamSignalFromConfig(params.signal, upstreamDeadlineMonotonicMs);
+	const upstreamSignal = createUpstreamSignalFromConfig(params.signal, upstreamDeadlineMonotonicMs, () => runtimeStore.requests.markAborted(params.requestId));
 	const copilotClient = createCopilotClient(recovery, {
 		offerLocalModelCooldown: options.offerLocalModelCooldown,
 		fallbackAttempt: options.fallbackAttempt
@@ -55433,12 +55566,31 @@ async function prepareAttempt(payload, meta, params, config, recovery, upstreamD
 			recovery
 		});
 		const entry = config.strategyRegistry.select(selectedModel, ctx);
-		runtimeStore.requests.recordStrategy(params.requestId, entry.name);
 		const strategyEffect = effectForStrategy(config.protocol, entry.name);
-		if (strategyEffect) runtimeStore.requests.recordEffect(params.requestId, strategyEffect);
+		let observabilityCommitted = !deferObservability;
+		let observabilityDiscarded = false;
+		if (!deferObservability) {
+			runtimeStore.requests.recordStrategy(params.requestId, entry.name);
+			if (strategyEffect) runtimeStore.requests.recordEffect(params.requestId, strategyEffect);
+		}
+		const commitObservability = () => {
+			if (observabilityCommitted || observabilityDiscarded) return;
+			observabilityCommitted = true;
+			runtimeStore.requests.recordModelMapping(params.requestId, modelMapping);
+			runtimeStore.requests.recordStrategy(params.requestId, entry.name);
+			if (strategyEffect) runtimeStore.requests.recordEffect(params.requestId, strategyEffect);
+			runtimeStore.requests.commitEffectBuffer(params.requestId);
+		};
+		const discardObservability = () => {
+			if (observabilityCommitted || observabilityDiscarded) return;
+			observabilityDiscarded = true;
+			runtimeStore.requests.discardEffectBuffer(params.requestId);
+		};
 		return {
 			baseModel,
 			modelMapping,
+			commitObservability,
+			discardObservability,
 			execute: async () => {
 				try {
 					params.signal.throwIfAborted();
@@ -55447,12 +55599,13 @@ async function prepareAttempt(payload, meta, params, config, recovery, upstreamD
 					upstreamSignal.cleanup();
 					throw error;
 				} finally {
-					if (modelMapping.steps.length !== recordedModelStepCount) runtimeStore.requests.recordModelMapping(params.requestId, modelMapping);
+					if (observabilityCommitted && modelMapping.steps.length !== recordedModelStepCount) runtimeStore.requests.recordModelMapping(params.requestId, modelMapping);
 				}
 			}
 		};
 	} catch (error) {
 		upstreamSignal.cleanup();
+		if (deferObservability) runtimeStore.requests.discardEffectBuffer(params.requestId);
 		throw error;
 	}
 }
@@ -55725,10 +55878,10 @@ function normalizeEmbeddingRequest(payload) {
 * derives its own upstream signal — otherwise a client disconnect would leave
 * the upstream request running and the configured timeout unenforced.
 */
-async function handleEmbeddingsCore(body, headers, client, signal, recovery) {
+async function handleEmbeddingsCore(body, headers, client, signal, recovery, onClientAbort) {
 	const { payload } = protocolRegistry.ingest("embeddings", body, headers);
 	const copilotClient = client ?? createCopilotClient(recovery);
-	const upstreamSignal = signal ? createUpstreamSignalFromConfig(signal) : void 0;
+	const upstreamSignal = signal ? createUpstreamSignalFromConfig(signal, void 0, onClientAbort) : void 0;
 	try {
 		return await copilotClient.createEmbeddings(normalizeEmbeddingRequest(payload), { signal: upstreamSignal?.signal });
 	} finally {
@@ -55739,7 +55892,8 @@ async function handleEmbeddingsCore(body, headers, client, signal, recovery) {
 //#region src/routes/embeddings/route.ts
 function createEmbeddingRoutes() {
 	return new Elysia().use(requestGuardPlugin).post("/embeddings", async ({ body, request }) => {
-		return handleEmbeddingsCore(body, request.headers, void 0, request.signal, createRequestRecoveryRecord(request));
+		const { requestId } = getOrCreateRequestCorrelation(request);
+		return handleEmbeddingsCore(body, request.headers, void 0, request.signal, createRequestRecoveryRecord(request), () => runtimeStore.requests.markAborted(requestId));
 	}, { guarded: true });
 }
 //#endregion
@@ -55754,18 +55908,42 @@ const ESTIMATION_FACTOR = {
 	grok: 1.03,
 	gpt: 1.1
 };
+const BUILTIN_TOOLSET_TOKENS = {
+	browser_toolset_20260801: 7550,
+	computer_toolset_20260801: 4590
+};
 /**
 * Core handler for counting tokens.
 */
 async function handleCountTokensCore({ body, headers }) {
 	const anthropicBeta = headers.get("anthropic-beta") ?? void 0;
 	const { payload: anthropicPayload } = protocolRegistry.ingest("anthropic-count-tokens", body, headers);
+	const functionTools = [];
+	const uncalibratedBuiltinTools = [];
+	let calibratedBuiltinTokens = 0;
+	let hasBuiltinTools = false;
+	for (const tool of anthropicPayload.tools ?? []) {
+		if (!isAnthropicBuiltinTool(tool)) {
+			functionTools.push(tool);
+			continue;
+		}
+		hasBuiltinTools = true;
+		const builtinTokens = tool.type ? BUILTIN_TOOLSET_TOKENS[tool.type] : void 0;
+		if (builtinTokens === void 0) uncalibratedBuiltinTools.push(tool);
+		else calibratedBuiltinTokens += builtinTokens;
+	}
+	const countPayload = hasBuiltinTools ? {
+		...anthropicPayload,
+		tools: functionTools
+	} : anthropicPayload;
 	const adapter = createAnthropicAdapter();
-	const openAIPayload = withTranslationErrors(() => adapter.toTokenCountPayload(anthropicPayload));
-	const tokenCount = await getTokenCount(openAIPayload, resolveModelOrThrow(openAIPayload.model));
-	if (anthropicPayload.tools && anthropicPayload.tools.length > 0) {
+	const openAIPayload = withTranslationErrors(() => adapter.toTokenCountPayload(countPayload));
+	const selectedModel = resolveModelOrThrow(openAIPayload.model);
+	const tokenCount = await getTokenCount(openAIPayload, selectedModel);
+	if (uncalibratedBuiltinTools.length > 0) tokenCount.input += await estimateSerializedTokens(uncalibratedBuiltinTools, selectedModel);
+	if (functionTools.length > 0) {
 		let mcpToolExist = false;
-		if (anthropicBeta?.startsWith("claude-code")) mcpToolExist = anthropicPayload.tools.some((tool) => tool.name.startsWith("mcp__"));
+		if (anthropicBeta?.startsWith("claude-code")) mcpToolExist = functionTools.some((tool) => tool.name?.startsWith("mcp__") ?? false);
 		if (!mcpToolExist) {
 			const overhead = TOOL_OVERHEAD_TOKENS[inferModelFamily(anthropicPayload.model)];
 			if (overhead) tokenCount.input = tokenCount.input + overhead;
@@ -55774,6 +55952,7 @@ async function handleCountTokensCore({ body, headers }) {
 	let finalTokenCount = tokenCount.input + tokenCount.output;
 	const factor = ESTIMATION_FACTOR[inferModelFamily(anthropicPayload.model)];
 	if (factor) finalTokenCount = Math.round(finalTokenCount * factor);
+	finalTokenCount += calibratedBuiltinTokens;
 	consola.info("Token count:", finalTokenCount);
 	return { input_tokens: finalTokenCount };
 }
@@ -55958,7 +56137,7 @@ function deleteStoredResponseOrThrow(responseId) {
 async function estimateEmulatorInputTokens(payload, selectedModel) {
 	return {
 		object: "response.input_tokens",
-		input_tokens: await estimateResponsesInputTokens(resolveEffectiveInputForInputTokens(payload), selectedModel)
+		input_tokens: await estimateSerializedTokens(resolveEffectiveInputForInputTokens(payload), selectedModel)
 	};
 }
 function resolveEffectiveInputForInputTokens(payload) {
@@ -56155,6 +56334,8 @@ async function handleResponsesCore({ body, signal, headers, requestId, callerReq
 			const { vision, initiator } = getResponsesRequestOptions(payload);
 			const prepared = emulatorPrepared;
 			const requestPayload = originalPayload ?? payload;
+			let effectiveInputItems;
+			if (prepared?.shouldStore) effectiveInputItems = Array.isArray(payload.input) ? structuredClone(payload.input) : [];
 			const strategyRequestId = recovery.requestId;
 			return {
 				requestId: strategyRequestId,
@@ -56170,7 +56351,7 @@ async function handleResponsesCore({ body, signal, headers, requestId, callerReq
 				}, requestPayload, prepared) : void 0,
 				onTerminalResponse(terminalResponse) {
 					if (terminalResponse.status === "failed" || terminalResponse.error) runtimeStore.requests.recordError(strategyRequestId, "Upstream HTTP 200 (response_failed)");
-					if (prepared?.shouldStore) persistEmulatorResponse(terminalResponse, prepared.effectiveInputItems);
+					if (effectiveInputItems) persistEmulatorResponse(terminalResponse, effectiveInputItems);
 				},
 				onStreamEndWithoutTerminal() {
 					if (upstreamSignal.clientSignal?.aborted) return;
@@ -56250,7 +56431,7 @@ function applyResponsesInputPolicies(payload) {
 }
 /**
 * Remove input items that Copilot cannot resolve and would trigger 404:
-* - `item_reference` items (opaque IDs from store=true sessions)
+* - `item_reference` items (Copilot does not resolve returned item IDs later)
 * - `function_call_output` items whose `call_id` has no matching prior
 *   `function_call` in the same input array (orphaned outputs)
 */
@@ -56316,7 +56497,10 @@ var UpstreamResourceDispatcher = class {
 		return this.client.getResponseInputItems(responseId, params, options);
 	}
 	createInputTokens(payload, options) {
-		return this.client.createResponseInputTokens(payload, options);
+		return this.client.createResponseInputTokens({
+			...payload,
+			store: false
+		}, options);
 	}
 	delete(responseId, options) {
 		return this.client.deleteResponse(responseId, options);
@@ -56334,33 +56518,45 @@ function createResourceDispatcher(client, recovery) {
 }
 //#endregion
 //#region src/routes/responses/resource-handler.ts
-async function handleRetrieveResponseCore({ params, url, headers, signal, client, recovery }) {
+async function handleRetrieveResponseCore({ params, url, headers, signal, client, recovery, onClientAbort }) {
 	const responseId = requireResponseId(params.responseId);
-	return await createResourceDispatcher(client, recovery).retrieve(responseId, getRetrieveParamsFromUrl(url), {
+	const dispatcher = createResourceDispatcher(client, recovery);
+	return withUpstreamSignal(signal, onClientAbort, (upstreamSignal) => dispatcher.retrieve(responseId, getRetrieveParamsFromUrl(url), {
 		requestContext: readCapiRequestContext(headers),
-		signal
-	});
+		signal: upstreamSignal
+	}));
 }
-async function handleListResponseInputItemsCore({ params, url, headers, signal, client, recovery }) {
+async function handleListResponseInputItemsCore({ params, url, headers, signal, client, recovery, onClientAbort }) {
 	const responseId = requireResponseId(params.responseId);
-	return await createResourceDispatcher(client, recovery).listInputItems(responseId, getInputItemsParamsFromUrl(url), {
+	const dispatcher = createResourceDispatcher(client, recovery);
+	return withUpstreamSignal(signal, onClientAbort, (upstreamSignal) => dispatcher.listInputItems(responseId, getInputItemsParamsFromUrl(url), {
 		requestContext: readCapiRequestContext(headers),
-		signal
-	});
+		signal: upstreamSignal
+	}));
 }
-async function handleCreateResponseInputTokensCore({ body, headers, signal, client, recovery }) {
+async function handleCreateResponseInputTokensCore({ body, headers, signal, client, recovery, onClientAbort }) {
 	const { payload, meta } = protocolRegistry.ingest("responses-input-tokens", body, headers);
-	return await createResourceDispatcher(client, recovery).createInputTokens(payload, {
+	const dispatcher = createResourceDispatcher(client, recovery);
+	return withUpstreamSignal(signal, onClientAbort, (upstreamSignal) => dispatcher.createInputTokens(payload, {
 		requestContext: meta.requestContext,
-		signal
-	});
+		signal: upstreamSignal
+	}));
 }
-async function handleDeleteResponseCore({ params, headers, signal, client, recovery }) {
+async function handleDeleteResponseCore({ params, headers, signal, client, recovery, onClientAbort }) {
 	const responseId = requireResponseId(params.responseId);
-	return await createResourceDispatcher(client, recovery).delete(responseId, {
+	const dispatcher = createResourceDispatcher(client, recovery);
+	return withUpstreamSignal(signal, onClientAbort, (upstreamSignal) => dispatcher.delete(responseId, {
 		requestContext: readCapiRequestContext(headers),
-		signal
-	});
+		signal: upstreamSignal
+	}));
+}
+async function withUpstreamSignal(clientSignal, onClientAbort, dispatch) {
+	const upstreamSignal = createUpstreamSignalFromConfig(clientSignal, void 0, onClientAbort);
+	try {
+		return await dispatch(upstreamSignal.signal);
+	} finally {
+		upstreamSignal.cleanup();
+	}
 }
 function requireResponseId(responseId) {
 	if (!responseId) throwInvalidRequestError("Response id is required.", "response_id");
@@ -56432,35 +56628,43 @@ function createResponsesRoutes() {
 		yield* delivery.stream;
 	}, { guarded: true }).post("/responses/input_tokens", async ({ body, request, server }) => {
 		disableIdleTimeout(server, request);
+		const { requestId } = getOrCreateRequestCorrelation(request);
 		return handleCreateResponseInputTokensCore({
 			body,
 			headers: request.headers,
 			signal: request.signal,
-			recovery: createRequestRecoveryRecord(request)
+			recovery: createRequestRecoveryRecord(request),
+			onClientAbort: () => runtimeStore.requests.markAborted(requestId)
 		});
 	}, { guarded: true }).get("/responses/:responseId/input_items", async ({ params, request }) => {
+		const { requestId } = getOrCreateRequestCorrelation(request);
 		return handleListResponseInputItemsCore({
 			params,
 			url: request.url,
 			headers: request.headers,
 			signal: request.signal,
-			recovery: createRequestRecoveryRecord(request)
+			recovery: createRequestRecoveryRecord(request),
+			onClientAbort: () => runtimeStore.requests.markAborted(requestId)
 		});
 	}, { guarded: true }).get("/responses/:responseId", async ({ params, request, server }) => {
 		if (hasStreamingResponsesQuery(request)) disableIdleTimeout(server, request);
+		const { requestId } = getOrCreateRequestCorrelation(request);
 		return handleRetrieveResponseCore({
 			params,
 			url: request.url,
 			headers: request.headers,
 			signal: request.signal,
-			recovery: createRequestRecoveryRecord(request)
+			recovery: createRequestRecoveryRecord(request),
+			onClientAbort: () => runtimeStore.requests.markAborted(requestId)
 		});
 	}, { guarded: true }).delete("/responses/:responseId", async ({ params, request }) => {
+		const { requestId } = getOrCreateRequestCorrelation(request);
 		return handleDeleteResponseCore({
 			params,
 			headers: request.headers,
 			signal: request.signal,
-			recovery: createRequestRecoveryRecord(request)
+			recovery: createRequestRecoveryRecord(request),
+			onClientAbort: () => runtimeStore.requests.markAborted(requestId)
 		});
 	}, { guarded: true });
 }
@@ -56619,7 +56823,7 @@ function createServer$1(options) {
 			set
 		});
 		const status = response?.status ?? (error instanceof HTTPError ? error.status : 500);
-		runtimeStore.requests.recordError(getOrCreateRequestCorrelation(request).requestId, sanitizeObservedError(error, code, status));
+		if (!isClientAbortError(error)) runtimeStore.requests.recordError(getOrCreateRequestCorrelation(request).requestId, sanitizeObservedError(error, code, status));
 		return response;
 	}).get("/", () => "Server running").get("/health", () => ({
 		status: "ok",
@@ -56632,6 +56836,7 @@ function createServer$1(options) {
 }
 //#endregion
 //#region src/start.ts
+const UNSIGNED_INTEGER_RE = /^\d+$/;
 async function maybeCopyClaudeCodeCommand(serverUrl) {
 	const models = modelCache.getModels();
 	if (!models) return;
@@ -56711,24 +56916,16 @@ async function runServer(options) {
 	process$1.on("SIGTERM", shutdown);
 	process$1.on("SIGINT", shutdown);
 }
-function parseIntArg(raw, name, fallbackMsg) {
+function parseIntArg(raw, name, fallbackMsg, min = 0, max = Number.MAX_SAFE_INTEGER) {
 	if (raw === void 0) return void 0;
-	const n = Number.parseInt(raw, 10);
-	if (Number.isNaN(n) || n < 0) {
+	const normalized = raw.trim();
+	const n = Number(normalized);
+	if (!UNSIGNED_INTEGER_RE.test(normalized) || !Number.isSafeInteger(n) || n < min || n > max) {
 		consola.warn(`Invalid --${name} value "${raw}". ${fallbackMsg}`);
 		return;
 	}
 	return n;
 }
-function parseBoundedIntArg(raw, name, fallbackMsg, min, max) {
-	if (raw === void 0) return void 0;
-	const value = Number(raw);
-	if (!raw.trim() || !Number.isInteger(value) || value < min || value > max) {
-		consola.warn(`Invalid --${name} value "${raw}". ${fallbackMsg}`);
-		return;
-	}
-	return value;
-}
 function secondsToMs(seconds) {
 	return seconds === void 0 ? void 0 : seconds * 1e3;
 }
@@ -56840,16 +57037,18 @@ const start = defineCommand({
 		}
 	},
 	run({ args }) {
+		const port = parseIntArg(args.port, "port", "Server not started.", 1, 65535);
+		if (port === void 0) throw new Error(`Invalid --port value "${args.port}".`);
 		const rateLimit = parseIntArg(args["rate-limit"], "rate-limit", "Rate limiting disabled.");
 		const idleTimeoutSeconds = parseIntArg(args["idle-timeout"], "idle-timeout", "Falling back to Bun default.");
-		const upstreamTimeoutSeconds = parseIntArg(args["upstream-timeout"], "upstream-timeout", "Falling back to default (300s).");
-		const upstreamQueueConcurrency = parseIntArg(args["upstream-queue-concurrency"], "upstream-queue-concurrency", "Using default upstream queue concurrency.");
-		const upstreamQueueMaxRetries = parseBoundedIntArg(args["upstream-queue-retries"], "upstream-queue-retries", "Using default upstream queue retry count.", 0, 2);
-		const upstreamRecoveryBudgetSeconds = parseBoundedIntArg(args["upstream-recovery-budget"], "upstream-recovery-budget", "Using default upstream recovery budget.", 1, 120);
+		const upstreamTimeoutSeconds = parseIntArg(args["upstream-timeout"], "upstream-timeout", "Falling back to default (1800s).");
+		const upstreamQueueConcurrency = parseIntArg(args["upstream-queue-concurrency"], "upstream-queue-concurrency", "Using default upstream queue concurrency.", 1);
+		const upstreamQueueMaxRetries = parseIntArg(args["upstream-queue-retries"], "upstream-queue-retries", "Using default upstream queue retry count.", 0, 2);
+		const upstreamRecoveryBudgetSeconds = parseIntArg(args["upstream-recovery-budget"], "upstream-recovery-budget", "Using default upstream recovery budget.", 1, 120);
 		const upstreamQueueBaseDelaySeconds = parseIntArg(args["upstream-queue-base-delay"], "upstream-queue-base-delay", "Using default upstream queue base delay.");
 		const upstreamQueueMaxDelaySeconds = parseIntArg(args["upstream-queue-max-delay"], "upstream-queue-max-delay", "Using default upstream queue max delay.");
 		return runServer({
-			port: Number.parseInt(args.port, 10),
+			port,
 			verbose: args.verbose,
 			accountType: args["account-type"],
 			manual: args.manual,

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "ghc-proxy",
   "type": "module",
-  "version": "0.10.0",
+  "version": "0.10.1",
   "description": "GitHub Copilot to OpenAI/Anthropic API proxy - Use Copilot with Claude Code, Cursor, and more",
   "author": "wxxb789 <wxxb789@outlook.com>",
   "homepage": "https://github.com/wxxb789/ghc-proxy",
@@ -27,7 +27,7 @@
   },
   "scripts": {
     "build": "tsdown",
-    "dev": "bun run --watch ./src/main.ts",
+    "dev": "bun run --watch ./src/main.ts start",
     "knip": "knip-bun",
     "lint": "eslint --cache",
     "lint:all": "eslint --cache .",
@@ -42,7 +42,7 @@
     "smoke:packaged": "bun run scripts/smoke/packaged-cli.ts",
     "smoke:translation": "bun run scripts/smoke/translation.ts",
     "serve": "NODE_ENV=production bun run ./src/main.ts start --wait --idle-timeout=255 --upstream-timeout=1800",
-    "start": "NODE_ENV=production bun run ./src/main.ts",
+    "start": "NODE_ENV=production bun run ./src/main.ts start",
     "typecheck": "npx tsc --project tsconfig.json"
   },
   "dependencies": {