@code-yeongyu/senpi 2026.9.28-4 → 2026.9.28-5

package/node_modules/@earendil-works/pi-ai/dist/providers/github-copilot-limits.d.ts

@@ -0,0 +1,11 @@
+import type { Credential } from "../auth/types.ts";
+import type { Api, Model } from "../types.ts";
+export interface GitHubCopilotModelLimit {
+    readonly id: string;
+    readonly maxContextWindowTokens?: number;
+    readonly maxPromptTokens?: number;
+    readonly maxOutputTokens?: number;
+}
+export declare function parseGitHubCopilotModelLimit(id: string, rawLimits: unknown): GitHubCopilotModelLimit | undefined;
+export declare function applyGitHubCopilotModelLimits<TApi extends Api>(models: readonly Model<TApi>[], credential: Credential | undefined): readonly Model<TApi>[];
+//# sourceMappingURL=github-copilot-limits.d.ts.map

package/node_modules/@earendil-works/pi-ai/dist/providers/github-copilot-limits.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"github-copilot-limits.d.ts","sourceRoot":"","sources":["../../src/providers/github-copilot-limits.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AACnD,OAAO,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AAE9C,MAAM,WAAW,uBAAuB;IACvC,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,sBAAsB,CAAC,EAAE,MAAM,CAAC;IACzC,QAAQ,CAAC,eAAe,CAAC,EAAE,MAAM,CAAC;IAClC,QAAQ,CAAC,eAAe,CAAC,EAAE,MAAM,CAAC;CAClC;AAUD,wBAAgB,4BAA4B,CAAC,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,OAAO,GAAG,uBAAuB,GAAG,SAAS,CAehH;AAmBD,wBAAgB,6BAA6B,CAAC,IAAI,SAAS,GAAG,EAC7D,MAAM,EAAE,SAAS,KAAK,CAAC,IAAI,CAAC,EAAE,EAC9B,UAAU,EAAE,UAAU,GAAG,SAAS,GAChC,SAAS,KAAK,CAAC,IAAI,CAAC,EAAE,CAuBxB"}

package/node_modules/@earendil-works/pi-ai/dist/providers/github-copilot-limits.js

@@ -0,0 +1,65 @@
+function asRecord(value) {
+    return value && typeof value === "object" ? value : undefined;
+}
+function positiveInteger(value) {
+    return typeof value === "number" && Number.isInteger(value) && value > 0 ? value : undefined;
+}
+export function parseGitHubCopilotModelLimit(id, rawLimits) {
+    const limits = asRecord(rawLimits);
+    if (!limits)
+        return undefined;
+    const maxContextWindowTokens = positiveInteger(limits.max_context_window_tokens);
+    const maxPromptTokens = positiveInteger(limits.max_prompt_tokens);
+    const maxOutputTokens = positiveInteger(limits.max_output_tokens);
+    if (maxContextWindowTokens === undefined && maxPromptTokens === undefined && maxOutputTokens === undefined) {
+        return undefined;
+    }
+    return {
+        id,
+        ...(maxContextWindowTokens !== undefined ? { maxContextWindowTokens } : {}),
+        ...(maxPromptTokens !== undefined ? { maxPromptTokens } : {}),
+        ...(maxOutputTokens !== undefined ? { maxOutputTokens } : {}),
+    };
+}
+function storedModelLimit(value) {
+    const entry = asRecord(value);
+    if (!entry || typeof entry.id !== "string")
+        return undefined;
+    const maxContextWindowTokens = positiveInteger(entry.maxContextWindowTokens);
+    const maxPromptTokens = positiveInteger(entry.maxPromptTokens);
+    const maxOutputTokens = positiveInteger(entry.maxOutputTokens);
+    if (maxContextWindowTokens === undefined && maxPromptTokens === undefined && maxOutputTokens === undefined) {
+        return undefined;
+    }
+    return {
+        id: entry.id,
+        ...(maxContextWindowTokens !== undefined ? { maxContextWindowTokens } : {}),
+        ...(maxPromptTokens !== undefined ? { maxPromptTokens } : {}),
+        ...(maxOutputTokens !== undefined ? { maxOutputTokens } : {}),
+    };
+}
+export function applyGitHubCopilotModelLimits(models, credential) {
+    if (credential?.type !== "oauth" || !Array.isArray(credential.copilotModelLimits))
+        return models;
+    const limitsById = new Map(credential.copilotModelLimits
+        .map(storedModelLimit)
+        .filter((entry) => entry !== undefined)
+        .map((entry) => [entry.id, entry]));
+    if (limitsById.size === 0)
+        return models;
+    return models.map((model) => {
+        const limits = limitsById.get(model.id);
+        if (!limits)
+            return model;
+        const reportedPromptLimit = limits.maxPromptTokens ?? limits.maxContextWindowTokens;
+        const contextWindow = reportedPromptLimit !== undefined && limits.maxContextWindowTokens !== undefined
+            ? Math.min(reportedPromptLimit, limits.maxContextWindowTokens)
+            : reportedPromptLimit;
+        return {
+            ...model,
+            contextWindow: contextWindow ?? model.contextWindow,
+            maxTokens: limits.maxOutputTokens ?? model.maxTokens,
+        };
+    });
+}
+//# sourceMappingURL=github-copilot-limits.js.map

package/node_modules/@earendil-works/pi-ai/dist/providers/github-copilot-limits.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"github-copilot-limits.js","sourceRoot":"","sources":["../../src/providers/github-copilot-limits.ts"],"names":[],"mappings":"AAUA,SAAS,QAAQ,CAAC,KAAc;IAC/B,OAAO,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAE,KAAiC,CAAC,CAAC,CAAC,SAAS,CAAC;AAC5F,CAAC;AAED,SAAS,eAAe,CAAC,KAAc;IACtC,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AAC9F,CAAC;AAED,MAAM,UAAU,4BAA4B,CAAC,EAAU,EAAE,SAAkB;IAC1E,MAAM,MAAM,GAAG,QAAQ,CAAC,SAAS,CAAC,CAAC;IACnC,IAAI,CAAC,MAAM;QAAE,OAAO,SAAS,CAAC;IAC9B,MAAM,sBAAsB,GAAG,eAAe,CAAC,MAAM,CAAC,yBAAyB,CAAC,CAAC;IACjF,MAAM,eAAe,GAAG,eAAe,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC;IAClE,MAAM,eAAe,GAAG,eAAe,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC;IAClE,IAAI,sBAAsB,KAAK,SAAS,IAAI,eAAe,KAAK,SAAS,IAAI,eAAe,KAAK,SAAS,EAAE,CAAC;QAC5G,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,OAAO;QACN,EAAE;QACF,GAAG,CAAC,sBAAsB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,sBAAsB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC3E,GAAG,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7D,GAAG,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC7D,CAAC;AACH,CAAC;AAED,SAAS,gBAAgB,CAAC,KAAc;IACvC,MAAM,KAAK,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC;IAC9B,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,CAAC,EAAE,KAAK,QAAQ;QAAE,OAAO,SAAS,CAAC;IAC7D,MAAM,sBAAsB,GAAG,eAAe,CAAC,KAAK,CAAC,sBAAsB,CAAC,CAAC;IAC7E,MAAM,eAAe,GAAG,eAAe,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC;IAC/D,MAAM,eAAe,GAAG,eAAe,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC;IAC/D,IAAI,sBAAsB,KAAK,SAAS,IAAI,eAAe,KAAK,SAAS,IAAI,eAAe,KAAK,SAAS,EAAE,CAAC;QAC5G,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,OAAO;QACN,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,GAAG,CAAC,sBAAsB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,sBAAsB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC3E,GAAG,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7D,GAAG,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC7D,CAAC;AACH,CAAC;AAED,MAAM,UAAU,6BAA6B,CAC5C,MAA8B,EAC9B,UAAkC;IAElC,IAAI,UAAU,EAAE,IAAI,KAAK,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,UAAU,CAAC,kBAAkB,CAAC;QAAE,OAAO,MAAM,CAAC;IACjG,MAAM,UAAU,GAAG,IAAI,GAAG,CACzB,UAAU,CAAC,kBAAkB;SAC3B,GAAG,CAAC,gBAAgB,CAAC;SACrB,MAAM,CAAC,CAAC,KAAK,EAAoC,EAAE,CAAC,KAAK,KAAK,SAAS,CAAC;SACxE,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAU,CAAC,CAC5C,CAAC;IACF,IAAI,UAAU,CAAC,IAAI,KAAK,CAAC;QAAE,OAAO,MAAM,CAAC;IACzC,OAAO,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;QAC3B,MAAM,MAAM,GAAG,UAAU,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACxC,IAAI,CAAC,MAAM;YAAE,OAAO,KAAK,CAAC;QAC1B,MAAM,mBAAmB,GAAG,MAAM,CAAC,eAAe,IAAI,MAAM,CAAC,sBAAsB,CAAC;QACpF,MAAM,aAAa,GAClB,mBAAmB,KAAK,SAAS,IAAI,MAAM,CAAC,sBAAsB,KAAK,SAAS;YAC/E,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,mBAAmB,EAAE,MAAM,CAAC,sBAAsB,CAAC;YAC9D,CAAC,CAAC,mBAAmB,CAAC;QACxB,OAAO;YACN,GAAG,KAAK;YACR,aAAa,EAAE,aAAa,IAAI,KAAK,CAAC,aAAa;YACnD,SAAS,EAAE,MAAM,CAAC,eAAe,IAAI,KAAK,CAAC,SAAS;SACpD,CAAC;IACH,CAAC,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import type { Credential } from \"../auth/types.ts\";\nimport type { Api, Model } from \"../types.ts\";\n\nexport interface GitHubCopilotModelLimit {\n\treadonly id: string;\n\treadonly maxContextWindowTokens?: number;\n\treadonly maxPromptTokens?: number;\n\treadonly maxOutputTokens?: number;\n}\n\nfunction asRecord(value: unknown): Record<string, unknown> | undefined {\n\treturn value && typeof value === \"object\" ? (value as Record<string, unknown>) : undefined;\n}\n\nfunction positiveInteger(value: unknown): number | undefined {\n\treturn typeof value === \"number\" && Number.isInteger(value) && value > 0 ? value : undefined;\n}\n\nexport function parseGitHubCopilotModelLimit(id: string, rawLimits: unknown): GitHubCopilotModelLimit | undefined {\n\tconst limits = asRecord(rawLimits);\n\tif (!limits) return undefined;\n\tconst maxContextWindowTokens = positiveInteger(limits.max_context_window_tokens);\n\tconst maxPromptTokens = positiveInteger(limits.max_prompt_tokens);\n\tconst maxOutputTokens = positiveInteger(limits.max_output_tokens);\n\tif (maxContextWindowTokens === undefined && maxPromptTokens === undefined && maxOutputTokens === undefined) {\n\t\treturn undefined;\n\t}\n\treturn {\n\t\tid,\n\t\t...(maxContextWindowTokens !== undefined ? { maxContextWindowTokens } : {}),\n\t\t...(maxPromptTokens !== undefined ? { maxPromptTokens } : {}),\n\t\t...(maxOutputTokens !== undefined ? { maxOutputTokens } : {}),\n\t};\n}\n\nfunction storedModelLimit(value: unknown): GitHubCopilotModelLimit | undefined {\n\tconst entry = asRecord(value);\n\tif (!entry || typeof entry.id !== \"string\") return undefined;\n\tconst maxContextWindowTokens = positiveInteger(entry.maxContextWindowTokens);\n\tconst maxPromptTokens = positiveInteger(entry.maxPromptTokens);\n\tconst maxOutputTokens = positiveInteger(entry.maxOutputTokens);\n\tif (maxContextWindowTokens === undefined && maxPromptTokens === undefined && maxOutputTokens === undefined) {\n\t\treturn undefined;\n\t}\n\treturn {\n\t\tid: entry.id,\n\t\t...(maxContextWindowTokens !== undefined ? { maxContextWindowTokens } : {}),\n\t\t...(maxPromptTokens !== undefined ? { maxPromptTokens } : {}),\n\t\t...(maxOutputTokens !== undefined ? { maxOutputTokens } : {}),\n\t};\n}\n\nexport function applyGitHubCopilotModelLimits<TApi extends Api>(\n\tmodels: readonly Model<TApi>[],\n\tcredential: Credential | undefined,\n): readonly Model<TApi>[] {\n\tif (credential?.type !== \"oauth\" || !Array.isArray(credential.copilotModelLimits)) return models;\n\tconst limitsById = new Map(\n\t\tcredential.copilotModelLimits\n\t\t\t.map(storedModelLimit)\n\t\t\t.filter((entry): entry is GitHubCopilotModelLimit => entry !== undefined)\n\t\t\t.map((entry) => [entry.id, entry] as const),\n\t);\n\tif (limitsById.size === 0) return models;\n\treturn models.map((model) => {\n\t\tconst limits = limitsById.get(model.id);\n\t\tif (!limits) return model;\n\t\tconst reportedPromptLimit = limits.maxPromptTokens ?? limits.maxContextWindowTokens;\n\t\tconst contextWindow =\n\t\t\treportedPromptLimit !== undefined && limits.maxContextWindowTokens !== undefined\n\t\t\t\t? Math.min(reportedPromptLimit, limits.maxContextWindowTokens)\n\t\t\t\t: reportedPromptLimit;\n\t\treturn {\n\t\t\t...model,\n\t\t\tcontextWindow: contextWindow ?? model.contextWindow,\n\t\t\tmaxTokens: limits.maxOutputTokens ?? model.maxTokens,\n\t\t};\n\t});\n}\n"]}

package/node_modules/@earendil-works/pi-ai/dist/providers/github-copilot.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"github-copilot.d.ts","sourceRoot":"","sources":["../../src/providers/github-copilot.ts"],"names":[],"mappings":"AAKA,OAAO,EAAkB,KAAK,QAAQ,EAAE,MAAM,cAAc,CAAC;AAG7D,wBAAgB,qBAAqB,IAAI,QAAQ,CAAC,oBAAoB,GAAG,oBAAoB,GAAG,kBAAkB,CAAC,CAyBlH"}
+{"version":3,"file":"github-copilot.d.ts","sourceRoot":"","sources":["../../src/providers/github-copilot.ts"],"names":[],"mappings":"AAMA,OAAO,EAAkB,KAAK,QAAQ,EAAE,MAAM,cAAc,CAAC;AAI7D,wBAAgB,qBAAqB,IAAI,QAAQ,CAAC,oBAAoB,GAAG,oBAAoB,GAAG,kBAAkB,CAAC,CAiClH"}

package/node_modules/@earendil-works/pi-ai/dist/providers/github-copilot.js

@@ -1,10 +1,12 @@
 import { anthropicMessagesApi } from "../api/anthropic-messages.lazy.js";
+import { GITHUB_COPILOT_REJECTED_TOKEN_STATUSES } from "../api/github-copilot-headers.js";
 import { openAICompletionsApi } from "../api/openai-completions.lazy.js";
 import { openAIResponsesApi } from "../api/openai-responses.lazy.js";
 import { envApiKeyAuth, lazyOAuth } from "../auth/helpers.js";
 import { loadGitHubCopilotOAuth } from "../auth/oauth/load.js";
 import { createProvider } from "../models.js";
 import { GITHUB_COPILOT_MODELS } from "./github-copilot.models.js";
+import { applyGitHubCopilotModelLimits } from "./github-copilot-limits.js";
 export function githubCopilotProvider() {
     return createProvider({
         id: "github-copilot",
@@ -12,7 +14,12 @@ export function githubCopilotProvider() {
         baseUrl: "https://api.individual.githubcopilot.com",
         auth: {
             apiKey: envApiKeyAuth("GitHub Copilot token", ["COPILOT_GITHUB_TOKEN"]),
-            oauth: lazyOAuth({ name: "GitHub Copilot", isSubscription: true, load: loadGitHubCopilotOAuth }),
+            oauth: lazyOAuth({
+                name: "GitHub Copilot",
+                isSubscription: true,
+                rejectedTokenStatuses: GITHUB_COPILOT_REJECTED_TOKEN_STATUSES,
+                load: loadGitHubCopilotOAuth,
+            }),
         },
         models: Object.values(GITHUB_COPILOT_MODELS),
         filterModels: (models, credential) => {
@@ -23,7 +30,7 @@ export function githubCopilotProvider() {
                 return models;
             }
             const available = new Set(availableModelIds);
-            return models.filter((model) => available.has(model.id));
+            return applyGitHubCopilotModelLimits(models.filter((model) => available.has(model.id)), credential);
         },
         api: {
             "anthropic-messages": anthropicMessagesApi(),

package/node_modules/@earendil-works/pi-ai/dist/providers/github-copilot.js.map

@@ -1 +1 @@
-{"version":3,"file":"github-copilot.js","sourceRoot":"","sources":["../../src/providers/github-copilot.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,mCAAmC,CAAC;AACzE,OAAO,EAAE,oBAAoB,EAAE,MAAM,mCAAmC,CAAC;AACzE,OAAO,EAAE,kBAAkB,EAAE,MAAM,iCAAiC,CAAC;AACrE,OAAO,EAAE,aAAa,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAC;AAC9D,OAAO,EAAE,sBAAsB,EAAE,MAAM,uBAAuB,CAAC;AAC/D,OAAO,EAAE,cAAc,EAAiB,MAAM,cAAc,CAAC;AAC7D,OAAO,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AAEnE,MAAM,UAAU,qBAAqB;IACpC,OAAO,cAAc,CAAC;QACrB,EAAE,EAAE,gBAAgB;QACpB,IAAI,EAAE,gBAAgB;QACtB,OAAO,EAAE,0CAA0C;QACnD,IAAI,EAAE;YACL,MAAM,EAAE,aAAa,CAAC,sBAAsB,EAAE,CAAC,sBAAsB,CAAC,CAAC;YACvE,KAAK,EAAE,SAAS,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,cAAc,EAAE,IAAI,EAAE,IAAI,EAAE,sBAAsB,EAAE,CAAC;SAChG;QACD,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,qBAAqB,CAAC;QAC5C,YAAY,EAAE,CAAC,MAAM,EAAE,UAAU,EAAE,EAAE;YACpC,IAAI,UAAU,EAAE,IAAI,KAAK,OAAO;gBAAE,OAAO,MAAM,CAAC;YAChD,MAAM,iBAAiB,GAAG,UAAU,CAAC,iBAAiB,CAAC;YACvD,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,iBAAiB,CAAC,IAAI,CAAC,iBAAiB,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,OAAO,EAAE,KAAK,QAAQ,CAAC,EAAE,CAAC;gBACnG,OAAO,MAAM,CAAC;YACf,CAAC;YACD,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,iBAAiB,CAAC,CAAC;YAC7C,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC;QAC1D,CAAC;QACD,GAAG,EAAE;YACJ,oBAAoB,EAAE,oBAAoB,EAAE;YAC5C,oBAAoB,EAAE,oBAAoB,EAAE;YAC5C,kBAAkB,EAAE,kBAAkB,EAAE;SACxC;KACD,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import { anthropicMessagesApi } from \"../api/anthropic-messages.lazy.ts\";\nimport { openAICompletionsApi } from \"../api/openai-completions.lazy.ts\";\nimport { openAIResponsesApi } from \"../api/openai-responses.lazy.ts\";\nimport { envApiKeyAuth, lazyOAuth } from \"../auth/helpers.ts\";\nimport { loadGitHubCopilotOAuth } from \"../auth/oauth/load.ts\";\nimport { createProvider, type Provider } from \"../models.ts\";\nimport { GITHUB_COPILOT_MODELS } from \"./github-copilot.models.ts\";\n\nexport function githubCopilotProvider(): Provider<\"anthropic-messages\" | \"openai-completions\" | \"openai-responses\"> {\n\treturn createProvider({\n\t\tid: \"github-copilot\",\n\t\tname: \"GitHub Copilot\",\n\t\tbaseUrl: \"https://api.individual.githubcopilot.com\",\n\t\tauth: {\n\t\t\tapiKey: envApiKeyAuth(\"GitHub Copilot token\", [\"COPILOT_GITHUB_TOKEN\"]),\n\t\t\toauth: lazyOAuth({ name: \"GitHub Copilot\", isSubscription: true, load: loadGitHubCopilotOAuth }),\n\t\t},\n\t\tmodels: Object.values(GITHUB_COPILOT_MODELS),\n\t\tfilterModels: (models, credential) => {\n\t\t\tif (credential?.type !== \"oauth\") return models;\n\t\t\tconst availableModelIds = credential.availableModelIds;\n\t\t\tif (!Array.isArray(availableModelIds) || !availableModelIds.every((id) => typeof id === \"string\")) {\n\t\t\t\treturn models;\n\t\t\t}\n\t\t\tconst available = new Set(availableModelIds);\n\t\t\treturn models.filter((model) => available.has(model.id));\n\t\t},\n\t\tapi: {\n\t\t\t\"anthropic-messages\": anthropicMessagesApi(),\n\t\t\t\"openai-completions\": openAICompletionsApi(),\n\t\t\t\"openai-responses\": openAIResponsesApi(),\n\t\t},\n\t});\n}\n"]}
+{"version":3,"file":"github-copilot.js","sourceRoot":"","sources":["../../src/providers/github-copilot.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,mCAAmC,CAAC;AACzE,OAAO,EAAE,sCAAsC,EAAE,MAAM,kCAAkC,CAAC;AAC1F,OAAO,EAAE,oBAAoB,EAAE,MAAM,mCAAmC,CAAC;AACzE,OAAO,EAAE,kBAAkB,EAAE,MAAM,iCAAiC,CAAC;AACrE,OAAO,EAAE,aAAa,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAC;AAC9D,OAAO,EAAE,sBAAsB,EAAE,MAAM,uBAAuB,CAAC;AAC/D,OAAO,EAAE,cAAc,EAAiB,MAAM,cAAc,CAAC;AAC7D,OAAO,EAAE,qBAAqB,EAAE,MAAM,4BAA4B,CAAC;AACnE,OAAO,EAAE,6BAA6B,EAAE,MAAM,4BAA4B,CAAC;AAE3E,MAAM,UAAU,qBAAqB;IACpC,OAAO,cAAc,CAAC;QACrB,EAAE,EAAE,gBAAgB;QACpB,IAAI,EAAE,gBAAgB;QACtB,OAAO,EAAE,0CAA0C;QACnD,IAAI,EAAE;YACL,MAAM,EAAE,aAAa,CAAC,sBAAsB,EAAE,CAAC,sBAAsB,CAAC,CAAC;YACvE,KAAK,EAAE,SAAS,CAAC;gBAChB,IAAI,EAAE,gBAAgB;gBACtB,cAAc,EAAE,IAAI;gBACpB,qBAAqB,EAAE,sCAAsC;gBAC7D,IAAI,EAAE,sBAAsB;aAC5B,CAAC;SACF;QACD,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,qBAAqB,CAAC;QAC5C,YAAY,EAAE,CAAC,MAAM,EAAE,UAAU,EAAE,EAAE;YACpC,IAAI,UAAU,EAAE,IAAI,KAAK,OAAO;gBAAE,OAAO,MAAM,CAAC;YAChD,MAAM,iBAAiB,GAAG,UAAU,CAAC,iBAAiB,CAAC;YACvD,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,iBAAiB,CAAC,IAAI,CAAC,iBAAiB,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,OAAO,EAAE,KAAK,QAAQ,CAAC,EAAE,CAAC;gBACnG,OAAO,MAAM,CAAC;YACf,CAAC;YACD,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,iBAAiB,CAAC,CAAC;YAC7C,OAAO,6BAA6B,CACnC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,EACjD,UAAU,CACV,CAAC;QACH,CAAC;QACD,GAAG,EAAE;YACJ,oBAAoB,EAAE,oBAAoB,EAAE;YAC5C,oBAAoB,EAAE,oBAAoB,EAAE;YAC5C,kBAAkB,EAAE,kBAAkB,EAAE;SACxC;KACD,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import { anthropicMessagesApi } from \"../api/anthropic-messages.lazy.ts\";\nimport { GITHUB_COPILOT_REJECTED_TOKEN_STATUSES } from \"../api/github-copilot-headers.ts\";\nimport { openAICompletionsApi } from \"../api/openai-completions.lazy.ts\";\nimport { openAIResponsesApi } from \"../api/openai-responses.lazy.ts\";\nimport { envApiKeyAuth, lazyOAuth } from \"../auth/helpers.ts\";\nimport { loadGitHubCopilotOAuth } from \"../auth/oauth/load.ts\";\nimport { createProvider, type Provider } from \"../models.ts\";\nimport { GITHUB_COPILOT_MODELS } from \"./github-copilot.models.ts\";\nimport { applyGitHubCopilotModelLimits } from \"./github-copilot-limits.ts\";\n\nexport function githubCopilotProvider(): Provider<\"anthropic-messages\" | \"openai-completions\" | \"openai-responses\"> {\n\treturn createProvider({\n\t\tid: \"github-copilot\",\n\t\tname: \"GitHub Copilot\",\n\t\tbaseUrl: \"https://api.individual.githubcopilot.com\",\n\t\tauth: {\n\t\t\tapiKey: envApiKeyAuth(\"GitHub Copilot token\", [\"COPILOT_GITHUB_TOKEN\"]),\n\t\t\toauth: lazyOAuth({\n\t\t\t\tname: \"GitHub Copilot\",\n\t\t\t\tisSubscription: true,\n\t\t\t\trejectedTokenStatuses: GITHUB_COPILOT_REJECTED_TOKEN_STATUSES,\n\t\t\t\tload: loadGitHubCopilotOAuth,\n\t\t\t}),\n\t\t},\n\t\tmodels: Object.values(GITHUB_COPILOT_MODELS),\n\t\tfilterModels: (models, credential) => {\n\t\t\tif (credential?.type !== \"oauth\") return models;\n\t\t\tconst availableModelIds = credential.availableModelIds;\n\t\t\tif (!Array.isArray(availableModelIds) || !availableModelIds.every((id) => typeof id === \"string\")) {\n\t\t\t\treturn models;\n\t\t\t}\n\t\t\tconst available = new Set(availableModelIds);\n\t\t\treturn applyGitHubCopilotModelLimits(\n\t\t\t\tmodels.filter((model) => available.has(model.id)),\n\t\t\t\tcredential,\n\t\t\t);\n\t\t},\n\t\tapi: {\n\t\t\t\"anthropic-messages\": anthropicMessagesApi(),\n\t\t\t\"openai-completions\": openAICompletionsApi(),\n\t\t\t\"openai-responses\": openAIResponsesApi(),\n\t\t},\n\t});\n}\n"]}

package/node_modules/@earendil-works/pi-ai/dist/utils/github-copilot-tool-limit.d.ts

@@ -0,0 +1,10 @@
+import type { AssistantMessage, ProviderId } from "../types.ts";
+export declare const GITHUB_COPILOT_TOOL_LIMIT = 128;
+export declare const GITHUB_COPILOT_TOOL_LIMIT_DIAGNOSTIC = "github_copilot_tool_limit";
+export declare function limitGitHubCopilotTools<T>(provider: ProviderId, tools: T[] | undefined, toolChoice?: unknown): {
+    tools: T[] | undefined;
+    omittedCount: number;
+};
+export declare function recordGitHubCopilotToolLimit(message: AssistantMessage, omittedCount: number): void;
+export declare function formatGitHubCopilotToolLimitError(message: AssistantMessage, errorMessage: string): string;
+//# sourceMappingURL=github-copilot-tool-limit.d.ts.map

package/node_modules/@earendil-works/pi-ai/dist/utils/github-copilot-tool-limit.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"github-copilot-tool-limit.d.ts","sourceRoot":"","sources":["../../src/utils/github-copilot-tool-limit.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAGhE,eAAO,MAAM,yBAAyB,MAAM,CAAC;AAC7C,eAAO,MAAM,oCAAoC,8BAA8B,CAAC;AA2BhF,wBAAgB,uBAAuB,CAAC,CAAC,EACxC,QAAQ,EAAE,UAAU,EACpB,KAAK,EAAE,CAAC,EAAE,GAAG,SAAS,EACtB,UAAU,CAAC,EAAE,OAAO,GAClB;IAAE,KAAK,EAAE,CAAC,EAAE,GAAG,SAAS,CAAC;IAAC,YAAY,EAAE,MAAM,CAAA;CAAE,CAmBlD;AAED,wBAAgB,4BAA4B,CAAC,OAAO,EAAE,gBAAgB,EAAE,YAAY,EAAE,MAAM,GAAG,IAAI,CAgBlG;AAED,wBAAgB,iCAAiC,CAAC,OAAO,EAAE,gBAAgB,EAAE,YAAY,EAAE,MAAM,GAAG,MAAM,CAazG"}

package/node_modules/@earendil-works/pi-ai/dist/utils/github-copilot-tool-limit.js

@@ -0,0 +1,75 @@
+import { appendAssistantMessageDiagnostic } from "./diagnostics.js";
+export const GITHUB_COPILOT_TOOL_LIMIT = 128;
+export const GITHUB_COPILOT_TOOL_LIMIT_DIAGNOSTIC = "github_copilot_tool_limit";
+function isRecord(value) {
+    return typeof value === "object" && value !== null;
+}
+function selectedToolName(toolChoice) {
+    if (!isRecord(toolChoice))
+        return undefined;
+    if (toolChoice.type === "function") {
+        if (isRecord(toolChoice.function) && typeof toolChoice.function.name === "string") {
+            return toolChoice.function.name;
+        }
+        return typeof toolChoice.name === "string" ? toolChoice.name : undefined;
+    }
+    if (toolChoice.type !== "custom" && toolChoice.type !== "tool")
+        return undefined;
+    return typeof toolChoice.name === "string" ? toolChoice.name : undefined;
+}
+function toolName(tool) {
+    if (!isRecord(tool))
+        return undefined;
+    if (tool.type === "function") {
+        if (isRecord(tool.function) && typeof tool.function.name === "string")
+            return tool.function.name;
+        return typeof tool.name === "string" ? tool.name : undefined;
+    }
+    return typeof tool.name === "string" ? tool.name : undefined;
+}
+export function limitGitHubCopilotTools(provider, tools, toolChoice) {
+    if (provider !== "github-copilot" || tools === undefined || tools.length <= GITHUB_COPILOT_TOOL_LIMIT) {
+        return { tools, omittedCount: 0 };
+    }
+    const forcedName = selectedToolName(toolChoice);
+    const forcedIndex = forcedName === undefined ? -1 : tools.findIndex((tool) => toolName(tool) === forcedName);
+    if (forcedIndex >= GITHUB_COPILOT_TOOL_LIMIT) {
+        const forcedTool = tools[forcedIndex];
+        if (forcedTool !== undefined) {
+            return {
+                tools: [...tools.slice(0, GITHUB_COPILOT_TOOL_LIMIT - 1), forcedTool],
+                omittedCount: tools.length - GITHUB_COPILOT_TOOL_LIMIT,
+            };
+        }
+    }
+    return {
+        tools: tools.slice(0, GITHUB_COPILOT_TOOL_LIMIT),
+        omittedCount: tools.length - GITHUB_COPILOT_TOOL_LIMIT,
+    };
+}
+export function recordGitHubCopilotToolLimit(message, omittedCount) {
+    if (omittedCount === 0 ||
+        message.diagnostics?.some((diagnostic) => diagnostic.type === GITHUB_COPILOT_TOOL_LIMIT_DIAGNOSTIC)) {
+        return;
+    }
+    appendAssistantMessageDiagnostic(message, {
+        type: GITHUB_COPILOT_TOOL_LIMIT_DIAGNOSTIC,
+        timestamp: Date.now(),
+        details: {
+            limit: GITHUB_COPILOT_TOOL_LIMIT,
+            omittedCount,
+            message: `GitHub Copilot accepts at most ${GITHUB_COPILOT_TOOL_LIMIT} tools on endpoints without tool search; senpi omitted ${omittedCount} excess tool definition${omittedCount === 1 ? "" : "s"}.`,
+        },
+    });
+}
+export function formatGitHubCopilotToolLimitError(message, errorMessage) {
+    const toolLimitApplied = message.diagnostics?.some((diagnostic) => diagnostic.type === GITHUB_COPILOT_TOOL_LIMIT_DIAGNOSTIC);
+    if (message.provider !== "github-copilot" ||
+        !toolLimitApplied ||
+        message.providerDiagnostic?.httpStatus !== 400 ||
+        !/400 Bad Request$/i.test(errorMessage)) {
+        return errorMessage;
+    }
+    return `GitHub Copilot rejected the request after senpi limited its tool list to ${GITHUB_COPILOT_TOOL_LIMIT}. The endpoint may enforce a lower limit or count additional hosted tools. (${errorMessage})`;
+}
+//# sourceMappingURL=github-copilot-tool-limit.js.map

package/node_modules/@earendil-works/pi-ai/dist/utils/github-copilot-tool-limit.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"github-copilot-tool-limit.js","sourceRoot":"","sources":["../../src/utils/github-copilot-tool-limit.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,gCAAgC,EAAE,MAAM,kBAAkB,CAAC;AAEpE,MAAM,CAAC,MAAM,yBAAyB,GAAG,GAAG,CAAC;AAC7C,MAAM,CAAC,MAAM,oCAAoC,GAAG,2BAA2B,CAAC;AAEhF,SAAS,QAAQ,CAAC,KAAc;IAC/B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,CAAC;AACpD,CAAC;AAED,SAAS,gBAAgB,CAAC,UAAmB;IAC5C,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC;QAAE,OAAO,SAAS,CAAC;IAC5C,IAAI,UAAU,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;QACpC,IAAI,QAAQ,CAAC,UAAU,CAAC,QAAQ,CAAC,IAAI,OAAO,UAAU,CAAC,QAAQ,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YACnF,OAAO,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC;QACjC,CAAC;QACD,OAAO,OAAO,UAAU,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;IAC1E,CAAC;IACD,IAAI,UAAU,CAAC,IAAI,KAAK,QAAQ,IAAI,UAAU,CAAC,IAAI,KAAK,MAAM;QAAE,OAAO,SAAS,CAAC;IACjF,OAAO,OAAO,UAAU,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;AAC1E,CAAC;AAED,SAAS,QAAQ,CAAC,IAAa;IAC9B,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC;QAAE,OAAO,SAAS,CAAC;IACtC,IAAI,IAAI,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;QAC9B,IAAI,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,OAAO,IAAI,CAAC,QAAQ,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC;QACjG,OAAO,OAAO,IAAI,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;IAC9D,CAAC;IACD,OAAO,OAAO,IAAI,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;AAC9D,CAAC;AAED,MAAM,UAAU,uBAAuB,CACtC,QAAoB,EACpB,KAAsB,EACtB,UAAoB;IAEpB,IAAI,QAAQ,KAAK,gBAAgB,IAAI,KAAK,KAAK,SAAS,IAAI,KAAK,CAAC,MAAM,IAAI,yBAAyB,EAAE,CAAC;QACvG,OAAO,EAAE,KAAK,EAAE,YAAY,EAAE,CAAC,EAAE,CAAC;IACnC,CAAC;IACD,MAAM,UAAU,GAAG,gBAAgB,CAAC,UAAU,CAAC,CAAC;IAChD,MAAM,WAAW,GAAG,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,UAAU,CAAC,CAAC;IAC7G,IAAI,WAAW,IAAI,yBAAyB,EAAE,CAAC;QAC9C,MAAM,UAAU,GAAG,KAAK,CAAC,WAAW,CAAC,CAAC;QACtC,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;YAC9B,OAAO;gBACN,KAAK,EAAE,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,yBAAyB,GAAG,CAAC,CAAC,EAAE,UAAU,CAAC;gBACrE,YAAY,EAAE,KAAK,CAAC,MAAM,GAAG,yBAAyB;aACtD,CAAC;QACH,CAAC;IACF,CAAC;IACD,OAAO;QACN,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,yBAAyB,CAAC;QAChD,YAAY,EAAE,KAAK,CAAC,MAAM,GAAG,yBAAyB;KACtD,CAAC;AACH,CAAC;AAED,MAAM,UAAU,4BAA4B,CAAC,OAAyB,EAAE,YAAoB;IAC3F,IACC,YAAY,KAAK,CAAC;QAClB,OAAO,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC,UAAU,EAAE,EAAE,CAAC,UAAU,CAAC,IAAI,KAAK,oCAAoC,CAAC,EAClG,CAAC;QACF,OAAO;IACR,CAAC;IACD,gCAAgC,CAAC,OAAO,EAAE;QACzC,IAAI,EAAE,oCAAoC;QAC1C,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;QACrB,OAAO,EAAE;YACR,KAAK,EAAE,yBAAyB;YAChC,YAAY;YACZ,OAAO,EAAE,kCAAkC,yBAAyB,0DAA0D,YAAY,0BAA0B,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,GAAG;SACpM;KACD,CAAC,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,iCAAiC,CAAC,OAAyB,EAAE,YAAoB;IAChG,MAAM,gBAAgB,GAAG,OAAO,CAAC,WAAW,EAAE,IAAI,CACjD,CAAC,UAAU,EAAE,EAAE,CAAC,UAAU,CAAC,IAAI,KAAK,oCAAoC,CACxE,CAAC;IACF,IACC,OAAO,CAAC,QAAQ,KAAK,gBAAgB;QACrC,CAAC,gBAAgB;QACjB,OAAO,CAAC,kBAAkB,EAAE,UAAU,KAAK,GAAG;QAC9C,CAAC,mBAAmB,CAAC,IAAI,CAAC,YAAY,CAAC,EACtC,CAAC;QACF,OAAO,YAAY,CAAC;IACrB,CAAC;IACD,OAAO,4EAA4E,yBAAyB,+EAA+E,YAAY,GAAG,CAAC;AAC5M,CAAC","sourcesContent":["import type { AssistantMessage, ProviderId } from \"../types.ts\";\nimport { appendAssistantMessageDiagnostic } from \"./diagnostics.ts\";\n\nexport const GITHUB_COPILOT_TOOL_LIMIT = 128;\nexport const GITHUB_COPILOT_TOOL_LIMIT_DIAGNOSTIC = \"github_copilot_tool_limit\";\n\nfunction isRecord(value: unknown): value is Record<string, unknown> {\n\treturn typeof value === \"object\" && value !== null;\n}\n\nfunction selectedToolName(toolChoice: unknown): string | undefined {\n\tif (!isRecord(toolChoice)) return undefined;\n\tif (toolChoice.type === \"function\") {\n\t\tif (isRecord(toolChoice.function) && typeof toolChoice.function.name === \"string\") {\n\t\t\treturn toolChoice.function.name;\n\t\t}\n\t\treturn typeof toolChoice.name === \"string\" ? toolChoice.name : undefined;\n\t}\n\tif (toolChoice.type !== \"custom\" && toolChoice.type !== \"tool\") return undefined;\n\treturn typeof toolChoice.name === \"string\" ? toolChoice.name : undefined;\n}\n\nfunction toolName(tool: unknown): string | undefined {\n\tif (!isRecord(tool)) return undefined;\n\tif (tool.type === \"function\") {\n\t\tif (isRecord(tool.function) && typeof tool.function.name === \"string\") return tool.function.name;\n\t\treturn typeof tool.name === \"string\" ? tool.name : undefined;\n\t}\n\treturn typeof tool.name === \"string\" ? tool.name : undefined;\n}\n\nexport function limitGitHubCopilotTools<T>(\n\tprovider: ProviderId,\n\ttools: T[] | undefined,\n\ttoolChoice?: unknown,\n): { tools: T[] | undefined; omittedCount: number } {\n\tif (provider !== \"github-copilot\" || tools === undefined || tools.length <= GITHUB_COPILOT_TOOL_LIMIT) {\n\t\treturn { tools, omittedCount: 0 };\n\t}\n\tconst forcedName = selectedToolName(toolChoice);\n\tconst forcedIndex = forcedName === undefined ? -1 : tools.findIndex((tool) => toolName(tool) === forcedName);\n\tif (forcedIndex >= GITHUB_COPILOT_TOOL_LIMIT) {\n\t\tconst forcedTool = tools[forcedIndex];\n\t\tif (forcedTool !== undefined) {\n\t\t\treturn {\n\t\t\t\ttools: [...tools.slice(0, GITHUB_COPILOT_TOOL_LIMIT - 1), forcedTool],\n\t\t\t\tomittedCount: tools.length - GITHUB_COPILOT_TOOL_LIMIT,\n\t\t\t};\n\t\t}\n\t}\n\treturn {\n\t\ttools: tools.slice(0, GITHUB_COPILOT_TOOL_LIMIT),\n\t\tomittedCount: tools.length - GITHUB_COPILOT_TOOL_LIMIT,\n\t};\n}\n\nexport function recordGitHubCopilotToolLimit(message: AssistantMessage, omittedCount: number): void {\n\tif (\n\t\tomittedCount === 0 ||\n\t\tmessage.diagnostics?.some((diagnostic) => diagnostic.type === GITHUB_COPILOT_TOOL_LIMIT_DIAGNOSTIC)\n\t) {\n\t\treturn;\n\t}\n\tappendAssistantMessageDiagnostic(message, {\n\t\ttype: GITHUB_COPILOT_TOOL_LIMIT_DIAGNOSTIC,\n\t\ttimestamp: Date.now(),\n\t\tdetails: {\n\t\t\tlimit: GITHUB_COPILOT_TOOL_LIMIT,\n\t\t\tomittedCount,\n\t\t\tmessage: `GitHub Copilot accepts at most ${GITHUB_COPILOT_TOOL_LIMIT} tools on endpoints without tool search; senpi omitted ${omittedCount} excess tool definition${omittedCount === 1 ? \"\" : \"s\"}.`,\n\t\t},\n\t});\n}\n\nexport function formatGitHubCopilotToolLimitError(message: AssistantMessage, errorMessage: string): string {\n\tconst toolLimitApplied = message.diagnostics?.some(\n\t\t(diagnostic) => diagnostic.type === GITHUB_COPILOT_TOOL_LIMIT_DIAGNOSTIC,\n\t);\n\tif (\n\t\tmessage.provider !== \"github-copilot\" ||\n\t\t!toolLimitApplied ||\n\t\tmessage.providerDiagnostic?.httpStatus !== 400 ||\n\t\t!/400 Bad Request$/i.test(errorMessage)\n\t) {\n\t\treturn errorMessage;\n\t}\n\treturn `GitHub Copilot rejected the request after senpi limited its tool list to ${GITHUB_COPILOT_TOOL_LIMIT}. The endpoint may enforce a lower limit or count additional hosted tools. (${errorMessage})`;\n}\n"]}

package/node_modules/@earendil-works/pi-ai/dist/utils/overflow.js

@@ -50,7 +50,7 @@ const OVERFLOW_PATTERNS = [
     /maximum context length is \d+ tokens/i, // OpenRouter (most backends)
     /exceeds (?:the )?maximum allowed input length of [\d,]+ tokens?/i, // OpenRouter/Poolside
     /input \(\d+ tokens\) is longer than the model'?s context length \(\d+ tokens\)/i, // Together AI
-    /exceeds the limit of \d+/i, // GitHub Copilot
+    /model_max_prompt_tokens_exceeded|exceeds the limit of \d+/i, // GitHub Copilot
     /exceeds the available context size/i, // llama.cpp server
     /greater than the context length/i, // LM Studio
     /context window exceeds limit/i, // MiniMax

package/node_modules/@earendil-works/pi-ai/dist/utils/overflow.js.map

@@ -1 +1 @@
-{"version":3,"file":"overflow.js","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCG;AACH,MAAM,iBAAiB,GAAG;IACzB,6BAA6B,EAAE,qEAAqE;IACpG,qBAAqB,EAAE,2BAA2B;IAClD,oBAAoB,EAAE,kDAAkD;IACxE,wCAAwC,EAAE,iBAAiB;IAC3D,yDAAyD,EAAE,uCAAuC;IAClG,4FAA4F,EAAE,sCAAsC;IACpI,yCAAyC,EAAE,kBAAkB;IAC7D,+BAA+B,EAAE,aAAa;IAC9C,oCAAoC,EAAE,OAAO;IAC7C,uCAAuC,EAAE,6BAA6B;IACtE,kEAAkE,EAAE,sBAAsB;IAC1F,iFAAiF,EAAE,cAAc;IACjG,2BAA2B,EAAE,iBAAiB;IAC9C,qCAAqC,EAAE,mBAAmB;IAC1D,kCAAkC,EAAE,YAAY;IAChD,+BAA+B,EAAE,UAAU;IAC3C,6BAA6B,EAAE,kBAAkB;IACjD,sDAAsD,EAAE,UAAU;IAClE,+EAA+E,EAAE,aAAa;IAC9F,gCAAgC,EAAE,yDAAyD;IAC3F,oDAAoD,EAAE,iCAAiC;IACvF,kCAAkC,EAAE,8BAA8B;IAClE,gCAAgC,EAAE,mBAAmB;IACrD,kBAAkB,EAAE,mBAAmB;IACvC,uBAAuB,EAAE,mBAAmB;IAC5C,+CAA+C,EAAE,iCAAiC;IAClF,0DAA0D,EAAE,4UAA4U;IACxY,qGAAqG,EAAE,+GAA+G;IACtN,0EAA0E,EAAE,+DAA+D;CAC3I,CAAC;AAEF;;;;;;;;;;GAUG;AACH,MAAM,qBAAqB,GAAG;IAC7B,2CAA2C,EAAE,oFAAoF;IACjI,aAAa,EAAE,wBAAwB;IACvC,oBAAoB,EAAE,yBAAyB;IAC/C,qCAAqC,EAAE,sDAAsD;IAC7F,UAAU,EAAE,iCAAiC;IAC7C,UAAU,EAAE,mCAAmC;IAC/C,iBAAiB,EAAE,qCAAqC;IACxD,wBAAwB,EAAE,iCAAiC;IAC3D,QAAQ,EAAE,kBAAkB;IAC5B,kBAAkB,EAAE,iCAAiC;IACrD,aAAa,EAAE,sCAAsC;CACrD,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,0BAA0B,GAAG,sBAAsB,CAAC;AAE1D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAqDG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAyB,EAAE,aAAsB;IAClF,uCAAuC;IACvC,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;QAC5D,oFAAoF;QACpF,MAAM,aAAa,GAAG,qBAAqB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,CAAC;QACvF,IAAI,CAAC,aAAa,IAAI,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,EAAE,CAAC;YACpF,OAAO,IAAI,CAAC;QACb,CAAC;QACD,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;QAC5B,MAAM,gBAAgB,GACrB,CAAC,KAAK,CAAC,WAAW,IAAI,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC;QAC5F,IACC,CAAC,aAAa;YACd,gBAAgB;YAChB,0BAA0B,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC;YACrD,CAAC,aAAa,KAAK,SAAS,IAAI,aAAa,IAAI,CAAC,IAAI,oBAAoB,CAAC,OAAO,CAAC,IAAI,aAAa,GAAG,GAAG,CAAC,EAC1G,CAAC;YACF,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,8EAA8E;IAC9E,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,MAAM,EAAE,CAAC;QACpD,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,GAAG,aAAa,EAAE,CAAC;YACjC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,sFAAsF;IACtF,qFAAqF;IACrF,gEAAgE;IAChE,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,QAAQ,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpF,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,IAAI,aAAa,GAAG,IAAI,EAAE,CAAC;YACzC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,OAAO,KAAK,CAAC;AACd,CAAC;AACD;;;;;GAKG;AACH,MAAM,UAAU,mBAAmB,CAAC,OAAyB,EAAE,gBAAwB;IACtF,OAAO,OAAO,CAAC,UAAU,KAAK,QAAQ,IAAI,gBAAgB,GAAG,CAAC,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,gBAAgB,CAAC;AAC3G,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB;IAClC,OAAO,CAAC,GAAG,iBAAiB,CAAC,CAAC;AAC/B,CAAC;AAED,SAAS,oBAAoB,CAAC,OAE7B;IACA,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;IAC5B,IAAI,CAAC,KAAK;QAAE,OAAO,CAAC,CAAC;IACrB,OAAO,CACN,KAAK,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,IAAI,CAAC,CAAC,CAChH,CAAC;AACH,CAAC;AAED,MAAM,UAAU,gCAAgC,CAC/C,OAIC,EACD,eAAuB;IAEvB,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,sBAAsB,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC,EAAE,CAAC;QAChG,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,CAAC,CAAC,oBAAoB,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC;AAC7C,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,8BAA8B,CAC7C,OAIC,EACD,aAAqB;IAErB,MAAM,MAAM,GAAG,oBAAoB,CAAC,OAAO,CAAC,CAAC;IAC7C,OAAO,CACN,OAAO,CAAC,UAAU,KAAK,OAAO;QAC9B,0BAA0B,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC;QAC3D,aAAa,GAAG,CAAC;QACjB,MAAM,GAAG,CAAC;QACV,MAAM,GAAG,aAAa,GAAG,GAAG,CAC5B,CAAC;AACH,CAAC;AAED,MAAM,UAAU,kCAAkC,CAAC,OAIlD;IACA,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,sBAAsB,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC,EAAE,CAAC;QAChG,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,CAAC,CAAC,oBAAoB,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC;AAC7C,CAAC;AAED,MAAM,UAAU,yCAAyC,CAAC,OAAkD;IAC3G,OAAO,OAAO,EAAE,eAAe,KAAK,IAAI,CAAC;AAC1C,CAAC;AAED,MAAM,UAAU,iCAAiC,CAAC,SAAkB,EAAE,YAAoB;IACzF,IAAI,SAAS;QAAE,OAAO,KAAK,CAAC;IAC5B,OAAO,qBAAqB,CAAC,IAAI,CAAC,YAAY,IAAI,EAAE,CAAC,CAAC;AACvD,CAAC;AAED,MAAM,UAAU,gCAAgC,CAC/C,QAAW,EACX,QAA4B,EAC5B,MAA0B;IAE1B,IAAI,MAAM,KAAK,UAAU;QAAE,OAAO,QAAQ,CAAC;IAC3C,IAAI,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,kBAAkB;QAAE,OAAO,QAAQ,CAAC;IAC9E,OAAO,EAAE,GAAG,QAAQ,EAAE,gBAAgB,EAAE,CAAC,EAAE,kBAAkB,EAAE,KAAK,EAAE,CAAC;AACxE,CAAC","sourcesContent":["import type { AssistantMessage } from \"../types.ts\";\n\n/**\n * Regex patterns to detect context overflow errors from different providers.\n *\n * These patterns match error messages returned when the input exceeds\n * the model's context window.\n *\n * Provider-specific patterns (with example error messages):\n *\n * - Anthropic: \"prompt is too long: 213462 tokens > 200000 maximum\"\n * - Anthropic: \"413 {\\\"error\\\":{\\\"type\\\":\\\"request_too_large\\\",\\\"message\\\":\\\"Request exceeds the maximum size\\\"}}\"\n * - OpenAI: \"Your input exceeds the context window of this model\"\n * - OpenAI: \"Your input exceeds the model's context window\" / \"exceeds this model's context window\"\n * - OpenAI/LiteLLM: \"Requested token count exceeds the model's maximum context length of 131072 tokens\"\n * - OpenAI-compatible: \"Input length (265330) exceeds model's maximum context length (262144).\"\n * - Google: \"The input token count (1196265) exceeds the maximum number of tokens allowed (1048575)\"\n * - xAI: \"This model's maximum prompt length is 131072 but the request contains 537812 tokens\"\n * - Groq: \"Please reduce the length of the messages or completion\"\n * - OpenRouter: \"This endpoint's maximum context length is X tokens. However, you requested about Y tokens\"\n * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (Y tokens).\"\n * - llama.cpp: \"the request exceeds the available context size, try increasing it\"\n * - LM Studio: \"tokens to keep from the initial prompt is greater than the context length\"\n * - GitHub Copilot: \"prompt token count of X exceeds the limit of Y\"\n * - MiniMax: \"invalid params, context window exceeds limit\"\n * - Kimi For Coding: \"Your request exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - Cerebras: \"400/413 status code (no body)\"\n * - Gateways: \"413 Request body too large\" / \"Request Entity Too Large\" / \"Payload Too Large\" (byte-size overflow)\n * - kiro-lb gateways: \"Request payload is 1095225 bytes, over the 1085435 byte limit Kiro accepts.\" / \"Request payload is N tokens, over the M token limit Kiro accepts.\" (HTTP 400 local payload guard)\n * - Kiro upstream via kiro-lb: \"Model context limit reached. Conversation size exceeds model capacity.\" (CONTENT_LENGTH_EXCEEDS_THRESHOLD token overflow)\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow\n * - Xiaomi MiMo: Truncates input to fill contextWindow exactly, then returns finish_reason \"length\"\n *   with output=0 (no room left to generate). Detected via stopReason \"length\" + zero output +\n *   input filling the context window.\n * - DashScope/Qwen: \"Range of input length should be [1, X]\" (HTTP 400 invalid_parameter_error)\n * - Ollama: Some deployments truncate silently, others return errors like \"prompt too long; exceeded max context length by X tokens\"\n * - pi-ai pre-flight guard (api/context-room.ts): \"Context window exhausted: the conversation is estimated at X of Y tokens, ...\" - raised before any provider call\n */\nconst OVERFLOW_PATTERNS = [\n\t/^Context window exhausted: /, // pi-ai pre-flight guard: no answer room left, provider never called\n\t/prompt is too long/i, // Anthropic token overflow\n\t/request_too_large/i, // Anthropic request byte-size overflow (HTTP 413)\n\t/input is too long for requested model/i, // Amazon Bedrock\n\t/exceeds (?:(?:the|this) )?(?:model'?s )?context window/i, // OpenAI (Completions & Responses API)\n\t/exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\\d,]+ tokens?|\\s*\\([\\d,]+\\))/i, // OpenAI-compatible proxies (LiteLLM)\n\t/input token count.*exceeds the maximum/i, // Google (Gemini)\n\t/maximum prompt length is \\d+/i, // xAI (Grok)\n\t/reduce the length of the messages/i, // Groq\n\t/maximum context length is \\d+ tokens/i, // OpenRouter (most backends)\n\t/exceeds (?:the )?maximum allowed input length of [\\d,]+ tokens?/i, // OpenRouter/Poolside\n\t/input \\(\\d+ tokens\\) is longer than the model'?s context length \\(\\d+ tokens\\)/i, // Together AI\n\t/exceeds the limit of \\d+/i, // GitHub Copilot\n\t/exceeds the available context size/i, // llama.cpp server\n\t/greater than the context length/i, // LM Studio\n\t/context window exceeds limit/i, // MiniMax\n\t/exceeded model token limit/i, // Kimi For Coding\n\t/too large for model with \\d+ maximum context length/i, // Mistral\n\t/prompt has [\\d,]+ tokens?, but the configured context size is [\\d,]+ tokens?/i, // DS4 server\n\t/model_context_window_exceeded/i, // z.ai non-standard finish_reason surfaced as error text\n\t/prompt too long; exceeded (?:max )?context length/i, // Ollama explicit overflow error\n\t/range of input length should be/i, // DashScope / Qwen Token Plan\n\t/context[_ ]length[_ ]exceeded/i, // Generic fallback\n\t/too many tokens/i, // Generic fallback\n\t/token limit exceeded/i, // Generic fallback\n\t/^4(?:00|13)\\s*(?:status code)?\\s*\\(no body\\)/i, // Cerebras: 400/413 with no body\n\t/(?:request[ _])?(?:body|entity|payload)[_ ]too[_ ]large/i, // Gateway HTTP 413 byte-size rejections (\"Request body too large\", \"Request Entity Too Large\", \"body_too_large\", \"Payload Too Large\"). Substring-anchored by design (JSON bodies lack an adjacent status code); a non-context size rejection (e.g. an oversized image) can over-match, which costs one bounded shrink-retry, never a wedge.\n\t/Request payload is \\d+ (?:bytes, over the \\d+ byte|tokens, over the \\d+ token) limit Kiro accepts\\./, // kiro-lb local byte/token payload guard (HTTP 400; Anthropic uses invalid_request_error, OpenAI uses detail).\n\t/Model context limit reached\\. Conversation size exceeds model capacity\\./, // kiro-lb enhancement of Kiro CONTENT_LENGTH_EXCEEDS_THRESHOLD\n];\n\n/**\n * Patterns that indicate non-overflow errors (e.g. rate limiting, server errors).\n * Error messages matching any of these are excluded from overflow detection\n * even if they also match an OVERFLOW_PATTERN.\n *\n * Example: Bedrock formats throttling errors as \"ThrottlingException: Too many tokens,\n * please wait before trying again.\" which would match the /too many tokens/i overflow\n * pattern without this exclusion. Token-quota / TPM messages such as \"Too many tokens\n * per minute\" or \"This request exceeds the limit of 30000 tokens per minute\" similarly\n * match generic overflow fallbacks and must stay on the rate-limit path.\n */\nconst NON_OVERFLOW_PATTERNS = [\n\t/^(Throttling error|Service unavailable):/i, // AWS Bedrock non-overflow errors (human-readable prefixes from formatBedrockError)\n\t/rate limit/i, // Generic rate limiting\n\t/too many requests/i, // Generic HTTP 429 style\n\t/tokens per (?:min|minute|hour|day)/i, // Token-quota windows (TPM/TPH/TPD), not context size\n\t/\\bTPM\\b/i, // Tokens-per-minute abbreviation\n\t/\\bRPM\\b/i, // Requests-per-minute abbreviation\n\t/quota exceeded/i, // Provider quota, not context window\n\t/retry (?:after|in) \\d/i, // Retry-after rate-limit wording\n\t/^429\\b/, // HTTP 429 prefix\n\t/status code 429/i, // HTTP 429 mentioned mid-message\n\t/overloaded/i, // Provider capacity, not context size\n];\n\n/**\n * Cursor's api2.cursor.sh surfaces a context overflow as a bare gRPC\n * `resource_exhausted` end-stream — the same wording its backend uses for\n * quota and poisoned-conversation rejections. Token evidence disambiguates:\n * a request that already streamed or billed tokens overflowed mid-flight,\n * while a zero-token rejection is a quota/conversation failure that must stay\n * on the rate-limit path (the cursor client rotates the conversation id for\n * those and retry handling supplies the backoff).\n */\nconst RESOURCE_EXHAUSTED_PATTERN = /resource.?exhausted/i;\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles three cases:\n * 1. Error-based overflow: Most providers return stopReason \"error\" with a\n *    specific error message pattern.\n * 2. Silent overflow: Some providers accept overflow requests and return\n *    successfully. For these, we check if usage.input exceeds the context window.\n * 3. Length-stop overflow: Xiaomi MiMo can return \"length\" with zero output when\n *    the input fills the context window.\n *\n * ## Reliability by Provider\n *\n * **Reliable detection (returns error with detectable message):**\n * - Anthropic: \"prompt is too long: X tokens > Y maximum\" or \"request_too_large\"\n * - OpenAI (Completions & Responses): \"exceeds the context window\", \"exceeds the model's context window\", \"exceeds this model's context window\", \"exceeds the model's maximum context length of X tokens\", or \"exceeds model's maximum context length (X)\"\n * - Google Gemini: \"input token count exceeds the maximum\"\n * - xAI (Grok): \"maximum prompt length is X but request contains Y\"\n * - Groq: \"reduce the length of the messages\"\n * - Cerebras: 400/413 status code (no body)\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - OpenRouter (most backends): \"maximum context length is X tokens\"\n * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (Y tokens).\"\n * - llama.cpp: \"exceeds the available context size\"\n * - LM Studio: \"greater than the context length\"\n * - Kimi For Coding: \"exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - DashScope/Qwen: \"Range of input length should be [1, X]\"\n *\n * **Unreliable detection:**\n * - z.ai: Sometimes accepts overflow silently (detectable via usage.input > contextWindow),\n *   sometimes returns rate limit errors. Pass contextWindow param to detect silent overflow.\n * - Xiaomi MiMo: Truncates input to fit contextWindow then returns stopReason \"length\" with\n *   output=0. Pass contextWindow param to detect via the \"filled context + zero output\" signal.\n * - Ollama: May truncate input silently for some setups, but may also return explicit\n *   overflow errors that match the patterns above. Silent truncation still cannot be\n *   detected here because we do not know the expected token count.\n *\n * ## Custom Providers\n *\n * If you've added custom models via settings.json, this function may not detect\n * overflow errors from those providers. To add support:\n *\n * 1. Send a request that exceeds the model's context window\n * 2. Check the errorMessage in the response\n * 3. Create a regex pattern that matches the error\n * 4. The pattern should be added to OVERFLOW_PATTERNS in this file, or\n *    check the errorMessage yourself before calling this function\n *\n * @param message - The assistant message to check\n * @param contextWindow - Optional context window size for detecting silent overflow (z.ai)\n * @returns true if the message indicates a context overflow\n */\nexport function isContextOverflow(message: AssistantMessage, contextWindow?: number): boolean {\n\t// Case 1: Check error message patterns\n\tif (message.stopReason === \"error\" && message.errorMessage) {\n\t\t// Skip messages matching known non-overflow patterns (e.g. throttling / rate-limit)\n\t\tconst isNonOverflow = NON_OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!));\n\t\tif (!isNonOverflow && OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!))) {\n\t\t\treturn true;\n\t\t}\n\t\tconst usage = message.usage;\n\t\tconst hasTokenEvidence =\n\t\t\t(usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite) > 0;\n\t\tif (\n\t\t\t!isNonOverflow &&\n\t\t\thasTokenEvidence &&\n\t\t\tRESOURCE_EXHAUSTED_PATTERN.test(message.errorMessage) &&\n\t\t\t(contextWindow === undefined || contextWindow <= 0 || cursorZeroTokenCount(message) >= contextWindow * 0.5)\n\t\t) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 2: Silent overflow (z.ai style) - successful but usage exceeds context\n\tif (contextWindow && message.stopReason === \"stop\") {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens > contextWindow) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 3: Length-stop overflow (Xiaomi MiMo style) - server truncates oversized input\n\t// to fit the context window, leaving no room for output. Returns stopReason \"length\"\n\t// with output=0 and input+cacheRead filling the context window.\n\tif (contextWindow && message.stopReason === \"length\" && message.usage.output === 0) {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens >= contextWindow * 0.99) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n/**\n * Check whether a length stop ended below the caller or model's intended output limit.\n * Such responses may be caused by context pressure or provider-side truncation, so callers\n * can make one bounded compact-and-retry attempt. `desiredMaxOutput` must be the original\n * limit before any context-based clamping.\n */\nexport function isRecoverableLength(message: AssistantMessage, desiredMaxOutput: number): boolean {\n\treturn message.stopReason === \"length\" && desiredMaxOutput > 0 && message.usage.output < desiredMaxOutput;\n}\n\n/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n\nfunction cursorZeroTokenCount(message: {\n\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n}): number {\n\tconst usage = message.usage;\n\tif (!usage) return 0;\n\treturn (\n\t\tusage.totalTokens || (usage.input ?? 0) + (usage.output ?? 0) + (usage.cacheRead ?? 0) + (usage.cacheWrite ?? 0)\n\t);\n}\n\nexport function isCursorPayloadResourceExhausted(\n\tmessage: {\n\t\tstopReason?: string;\n\t\terrorMessage?: string;\n\t\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n\t},\n\t_estimateTokens: number,\n): boolean {\n\tif (message.stopReason !== \"error\" || !/resource.?exhausted/i.test(message.errorMessage || \"\")) {\n\t\treturn false;\n\t}\n\treturn !(cursorZeroTokenCount(message) > 0);\n}\n\n/**\n * Detects Cursor's verified usage-pool exhaustion signature: a token-bearing\n * `resource_exhausted` error while the conversation is well below the model\n * context window.\n */\nexport function isCursorQuotaResourceExhausted(\n\tmessage: {\n\t\tstopReason?: string;\n\t\terrorMessage?: string;\n\t\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n\t},\n\tcontextWindow: number,\n): boolean {\n\tconst tokens = cursorZeroTokenCount(message);\n\treturn (\n\t\tmessage.stopReason === \"error\" &&\n\t\tRESOURCE_EXHAUSTED_PATTERN.test(message.errorMessage || \"\") &&\n\t\tcontextWindow > 0 &&\n\t\ttokens > 0 &&\n\t\ttokens < contextWindow * 0.5\n\t);\n}\n\nexport function isCursorZeroTokenResourceExhausted(message: {\n\tstopReason?: string;\n\terrorMessage?: string;\n\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n}): boolean {\n\tif (message.stopReason !== \"error\" || !/resource.?exhausted/i.test(message.errorMessage || \"\")) {\n\t\treturn false;\n\t}\n\treturn !(cursorZeroTokenCount(message) > 0);\n}\n\nexport function shouldSkipProviderFallbackForCursorZeroRe(options: { sameModelRemint?: boolean } | undefined): boolean {\n\treturn options?.sameModelRemint === true;\n}\n\nexport function shouldRetryOverflowWithoutCompact(compacted: boolean, errorMessage: string): boolean {\n\tif (compacted) return false;\n\treturn /Nothing to compact/i.test(errorMessage || \"\");\n}\n\nexport function cursorOverflowCompactionSettings<T extends { keepRecentTokens?: number; restorationEnabled?: boolean }>(\n\tsettings: T,\n\tprovider: string | undefined,\n\treason: string | undefined,\n): T {\n\tif (reason !== \"overflow\") return settings;\n\tif (provider !== \"cursor\" && provider !== \"cursor-cli-oauth\") return settings;\n\treturn { ...settings, keepRecentTokens: 0, restorationEnabled: false };\n}\n"]}
+{"version":3,"file":"overflow.js","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCG;AACH,MAAM,iBAAiB,GAAG;IACzB,6BAA6B,EAAE,qEAAqE;IACpG,qBAAqB,EAAE,2BAA2B;IAClD,oBAAoB,EAAE,kDAAkD;IACxE,wCAAwC,EAAE,iBAAiB;IAC3D,yDAAyD,EAAE,uCAAuC;IAClG,4FAA4F,EAAE,sCAAsC;IACpI,yCAAyC,EAAE,kBAAkB;IAC7D,+BAA+B,EAAE,aAAa;IAC9C,oCAAoC,EAAE,OAAO;IAC7C,uCAAuC,EAAE,6BAA6B;IACtE,kEAAkE,EAAE,sBAAsB;IAC1F,iFAAiF,EAAE,cAAc;IACjG,4DAA4D,EAAE,iBAAiB;IAC/E,qCAAqC,EAAE,mBAAmB;IAC1D,kCAAkC,EAAE,YAAY;IAChD,+BAA+B,EAAE,UAAU;IAC3C,6BAA6B,EAAE,kBAAkB;IACjD,sDAAsD,EAAE,UAAU;IAClE,+EAA+E,EAAE,aAAa;IAC9F,gCAAgC,EAAE,yDAAyD;IAC3F,oDAAoD,EAAE,iCAAiC;IACvF,kCAAkC,EAAE,8BAA8B;IAClE,gCAAgC,EAAE,mBAAmB;IACrD,kBAAkB,EAAE,mBAAmB;IACvC,uBAAuB,EAAE,mBAAmB;IAC5C,+CAA+C,EAAE,iCAAiC;IAClF,0DAA0D,EAAE,4UAA4U;IACxY,qGAAqG,EAAE,+GAA+G;IACtN,0EAA0E,EAAE,+DAA+D;CAC3I,CAAC;AAEF;;;;;;;;;;GAUG;AACH,MAAM,qBAAqB,GAAG;IAC7B,2CAA2C,EAAE,oFAAoF;IACjI,aAAa,EAAE,wBAAwB;IACvC,oBAAoB,EAAE,yBAAyB;IAC/C,qCAAqC,EAAE,sDAAsD;IAC7F,UAAU,EAAE,iCAAiC;IAC7C,UAAU,EAAE,mCAAmC;IAC/C,iBAAiB,EAAE,qCAAqC;IACxD,wBAAwB,EAAE,iCAAiC;IAC3D,QAAQ,EAAE,kBAAkB;IAC5B,kBAAkB,EAAE,iCAAiC;IACrD,aAAa,EAAE,sCAAsC;CACrD,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,0BAA0B,GAAG,sBAAsB,CAAC;AAE1D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAqDG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAyB,EAAE,aAAsB;IAClF,uCAAuC;IACvC,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,OAAO,CAAC,YAAY,EAAE,CAAC;QAC5D,oFAAoF;QACpF,MAAM,aAAa,GAAG,qBAAqB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,CAAC;QACvF,IAAI,CAAC,aAAa,IAAI,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,YAAa,CAAC,CAAC,EAAE,CAAC;YACpF,OAAO,IAAI,CAAC;QACb,CAAC;QACD,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;QAC5B,MAAM,gBAAgB,GACrB,CAAC,KAAK,CAAC,WAAW,IAAI,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC;QAC5F,IACC,CAAC,aAAa;YACd,gBAAgB;YAChB,0BAA0B,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC;YACrD,CAAC,aAAa,KAAK,SAAS,IAAI,aAAa,IAAI,CAAC,IAAI,oBAAoB,CAAC,OAAO,CAAC,IAAI,aAAa,GAAG,GAAG,CAAC,EAC1G,CAAC;YACF,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,8EAA8E;IAC9E,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,MAAM,EAAE,CAAC;QACpD,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,GAAG,aAAa,EAAE,CAAC;YACjC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,sFAAsF;IACtF,qFAAqF;IACrF,gEAAgE;IAChE,IAAI,aAAa,IAAI,OAAO,CAAC,UAAU,KAAK,QAAQ,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpF,MAAM,WAAW,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC;QAClE,IAAI,WAAW,IAAI,aAAa,GAAG,IAAI,EAAE,CAAC;YACzC,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,OAAO,KAAK,CAAC;AACd,CAAC;AACD;;;;;GAKG;AACH,MAAM,UAAU,mBAAmB,CAAC,OAAyB,EAAE,gBAAwB;IACtF,OAAO,OAAO,CAAC,UAAU,KAAK,QAAQ,IAAI,gBAAgB,GAAG,CAAC,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,gBAAgB,CAAC;AAC3G,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB;IAClC,OAAO,CAAC,GAAG,iBAAiB,CAAC,CAAC;AAC/B,CAAC;AAED,SAAS,oBAAoB,CAAC,OAE7B;IACA,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;IAC5B,IAAI,CAAC,KAAK;QAAE,OAAO,CAAC,CAAC;IACrB,OAAO,CACN,KAAK,CAAC,WAAW,IAAI,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,IAAI,CAAC,CAAC,CAChH,CAAC;AACH,CAAC;AAED,MAAM,UAAU,gCAAgC,CAC/C,OAIC,EACD,eAAuB;IAEvB,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,sBAAsB,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC,EAAE,CAAC;QAChG,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,CAAC,CAAC,oBAAoB,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC;AAC7C,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,8BAA8B,CAC7C,OAIC,EACD,aAAqB;IAErB,MAAM,MAAM,GAAG,oBAAoB,CAAC,OAAO,CAAC,CAAC;IAC7C,OAAO,CACN,OAAO,CAAC,UAAU,KAAK,OAAO;QAC9B,0BAA0B,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC;QAC3D,aAAa,GAAG,CAAC;QACjB,MAAM,GAAG,CAAC;QACV,MAAM,GAAG,aAAa,GAAG,GAAG,CAC5B,CAAC;AACH,CAAC;AAED,MAAM,UAAU,kCAAkC,CAAC,OAIlD;IACA,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,sBAAsB,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,CAAC,EAAE,CAAC;QAChG,OAAO,KAAK,CAAC;IACd,CAAC;IACD,OAAO,CAAC,CAAC,oBAAoB,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC;AAC7C,CAAC;AAED,MAAM,UAAU,yCAAyC,CAAC,OAAkD;IAC3G,OAAO,OAAO,EAAE,eAAe,KAAK,IAAI,CAAC;AAC1C,CAAC;AAED,MAAM,UAAU,iCAAiC,CAAC,SAAkB,EAAE,YAAoB;IACzF,IAAI,SAAS;QAAE,OAAO,KAAK,CAAC;IAC5B,OAAO,qBAAqB,CAAC,IAAI,CAAC,YAAY,IAAI,EAAE,CAAC,CAAC;AACvD,CAAC;AAED,MAAM,UAAU,gCAAgC,CAC/C,QAAW,EACX,QAA4B,EAC5B,MAA0B;IAE1B,IAAI,MAAM,KAAK,UAAU;QAAE,OAAO,QAAQ,CAAC;IAC3C,IAAI,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,kBAAkB;QAAE,OAAO,QAAQ,CAAC;IAC9E,OAAO,EAAE,GAAG,QAAQ,EAAE,gBAAgB,EAAE,CAAC,EAAE,kBAAkB,EAAE,KAAK,EAAE,CAAC;AACxE,CAAC","sourcesContent":["import type { AssistantMessage } from \"../types.ts\";\n\n/**\n * Regex patterns to detect context overflow errors from different providers.\n *\n * These patterns match error messages returned when the input exceeds\n * the model's context window.\n *\n * Provider-specific patterns (with example error messages):\n *\n * - Anthropic: \"prompt is too long: 213462 tokens > 200000 maximum\"\n * - Anthropic: \"413 {\\\"error\\\":{\\\"type\\\":\\\"request_too_large\\\",\\\"message\\\":\\\"Request exceeds the maximum size\\\"}}\"\n * - OpenAI: \"Your input exceeds the context window of this model\"\n * - OpenAI: \"Your input exceeds the model's context window\" / \"exceeds this model's context window\"\n * - OpenAI/LiteLLM: \"Requested token count exceeds the model's maximum context length of 131072 tokens\"\n * - OpenAI-compatible: \"Input length (265330) exceeds model's maximum context length (262144).\"\n * - Google: \"The input token count (1196265) exceeds the maximum number of tokens allowed (1048575)\"\n * - xAI: \"This model's maximum prompt length is 131072 but the request contains 537812 tokens\"\n * - Groq: \"Please reduce the length of the messages or completion\"\n * - OpenRouter: \"This endpoint's maximum context length is X tokens. However, you requested about Y tokens\"\n * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (Y tokens).\"\n * - llama.cpp: \"the request exceeds the available context size, try increasing it\"\n * - LM Studio: \"tokens to keep from the initial prompt is greater than the context length\"\n * - GitHub Copilot: \"prompt token count of X exceeds the limit of Y\"\n * - MiniMax: \"invalid params, context window exceeds limit\"\n * - Kimi For Coding: \"Your request exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - Cerebras: \"400/413 status code (no body)\"\n * - Gateways: \"413 Request body too large\" / \"Request Entity Too Large\" / \"Payload Too Large\" (byte-size overflow)\n * - kiro-lb gateways: \"Request payload is 1095225 bytes, over the 1085435 byte limit Kiro accepts.\" / \"Request payload is N tokens, over the M token limit Kiro accepts.\" (HTTP 400 local payload guard)\n * - Kiro upstream via kiro-lb: \"Model context limit reached. Conversation size exceeds model capacity.\" (CONTENT_LENGTH_EXCEEDS_THRESHOLD token overflow)\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow\n * - Xiaomi MiMo: Truncates input to fill contextWindow exactly, then returns finish_reason \"length\"\n *   with output=0 (no room left to generate). Detected via stopReason \"length\" + zero output +\n *   input filling the context window.\n * - DashScope/Qwen: \"Range of input length should be [1, X]\" (HTTP 400 invalid_parameter_error)\n * - Ollama: Some deployments truncate silently, others return errors like \"prompt too long; exceeded max context length by X tokens\"\n * - pi-ai pre-flight guard (api/context-room.ts): \"Context window exhausted: the conversation is estimated at X of Y tokens, ...\" - raised before any provider call\n */\nconst OVERFLOW_PATTERNS = [\n\t/^Context window exhausted: /, // pi-ai pre-flight guard: no answer room left, provider never called\n\t/prompt is too long/i, // Anthropic token overflow\n\t/request_too_large/i, // Anthropic request byte-size overflow (HTTP 413)\n\t/input is too long for requested model/i, // Amazon Bedrock\n\t/exceeds (?:(?:the|this) )?(?:model'?s )?context window/i, // OpenAI (Completions & Responses API)\n\t/exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\\d,]+ tokens?|\\s*\\([\\d,]+\\))/i, // OpenAI-compatible proxies (LiteLLM)\n\t/input token count.*exceeds the maximum/i, // Google (Gemini)\n\t/maximum prompt length is \\d+/i, // xAI (Grok)\n\t/reduce the length of the messages/i, // Groq\n\t/maximum context length is \\d+ tokens/i, // OpenRouter (most backends)\n\t/exceeds (?:the )?maximum allowed input length of [\\d,]+ tokens?/i, // OpenRouter/Poolside\n\t/input \\(\\d+ tokens\\) is longer than the model'?s context length \\(\\d+ tokens\\)/i, // Together AI\n\t/model_max_prompt_tokens_exceeded|exceeds the limit of \\d+/i, // GitHub Copilot\n\t/exceeds the available context size/i, // llama.cpp server\n\t/greater than the context length/i, // LM Studio\n\t/context window exceeds limit/i, // MiniMax\n\t/exceeded model token limit/i, // Kimi For Coding\n\t/too large for model with \\d+ maximum context length/i, // Mistral\n\t/prompt has [\\d,]+ tokens?, but the configured context size is [\\d,]+ tokens?/i, // DS4 server\n\t/model_context_window_exceeded/i, // z.ai non-standard finish_reason surfaced as error text\n\t/prompt too long; exceeded (?:max )?context length/i, // Ollama explicit overflow error\n\t/range of input length should be/i, // DashScope / Qwen Token Plan\n\t/context[_ ]length[_ ]exceeded/i, // Generic fallback\n\t/too many tokens/i, // Generic fallback\n\t/token limit exceeded/i, // Generic fallback\n\t/^4(?:00|13)\\s*(?:status code)?\\s*\\(no body\\)/i, // Cerebras: 400/413 with no body\n\t/(?:request[ _])?(?:body|entity|payload)[_ ]too[_ ]large/i, // Gateway HTTP 413 byte-size rejections (\"Request body too large\", \"Request Entity Too Large\", \"body_too_large\", \"Payload Too Large\"). Substring-anchored by design (JSON bodies lack an adjacent status code); a non-context size rejection (e.g. an oversized image) can over-match, which costs one bounded shrink-retry, never a wedge.\n\t/Request payload is \\d+ (?:bytes, over the \\d+ byte|tokens, over the \\d+ token) limit Kiro accepts\\./, // kiro-lb local byte/token payload guard (HTTP 400; Anthropic uses invalid_request_error, OpenAI uses detail).\n\t/Model context limit reached\\. Conversation size exceeds model capacity\\./, // kiro-lb enhancement of Kiro CONTENT_LENGTH_EXCEEDS_THRESHOLD\n];\n\n/**\n * Patterns that indicate non-overflow errors (e.g. rate limiting, server errors).\n * Error messages matching any of these are excluded from overflow detection\n * even if they also match an OVERFLOW_PATTERN.\n *\n * Example: Bedrock formats throttling errors as \"ThrottlingException: Too many tokens,\n * please wait before trying again.\" which would match the /too many tokens/i overflow\n * pattern without this exclusion. Token-quota / TPM messages such as \"Too many tokens\n * per minute\" or \"This request exceeds the limit of 30000 tokens per minute\" similarly\n * match generic overflow fallbacks and must stay on the rate-limit path.\n */\nconst NON_OVERFLOW_PATTERNS = [\n\t/^(Throttling error|Service unavailable):/i, // AWS Bedrock non-overflow errors (human-readable prefixes from formatBedrockError)\n\t/rate limit/i, // Generic rate limiting\n\t/too many requests/i, // Generic HTTP 429 style\n\t/tokens per (?:min|minute|hour|day)/i, // Token-quota windows (TPM/TPH/TPD), not context size\n\t/\\bTPM\\b/i, // Tokens-per-minute abbreviation\n\t/\\bRPM\\b/i, // Requests-per-minute abbreviation\n\t/quota exceeded/i, // Provider quota, not context window\n\t/retry (?:after|in) \\d/i, // Retry-after rate-limit wording\n\t/^429\\b/, // HTTP 429 prefix\n\t/status code 429/i, // HTTP 429 mentioned mid-message\n\t/overloaded/i, // Provider capacity, not context size\n];\n\n/**\n * Cursor's api2.cursor.sh surfaces a context overflow as a bare gRPC\n * `resource_exhausted` end-stream — the same wording its backend uses for\n * quota and poisoned-conversation rejections. Token evidence disambiguates:\n * a request that already streamed or billed tokens overflowed mid-flight,\n * while a zero-token rejection is a quota/conversation failure that must stay\n * on the rate-limit path (the cursor client rotates the conversation id for\n * those and retry handling supplies the backoff).\n */\nconst RESOURCE_EXHAUSTED_PATTERN = /resource.?exhausted/i;\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles three cases:\n * 1. Error-based overflow: Most providers return stopReason \"error\" with a\n *    specific error message pattern.\n * 2. Silent overflow: Some providers accept overflow requests and return\n *    successfully. For these, we check if usage.input exceeds the context window.\n * 3. Length-stop overflow: Xiaomi MiMo can return \"length\" with zero output when\n *    the input fills the context window.\n *\n * ## Reliability by Provider\n *\n * **Reliable detection (returns error with detectable message):**\n * - Anthropic: \"prompt is too long: X tokens > Y maximum\" or \"request_too_large\"\n * - OpenAI (Completions & Responses): \"exceeds the context window\", \"exceeds the model's context window\", \"exceeds this model's context window\", \"exceeds the model's maximum context length of X tokens\", or \"exceeds model's maximum context length (X)\"\n * - Google Gemini: \"input token count exceeds the maximum\"\n * - xAI (Grok): \"maximum prompt length is X but request contains Y\"\n * - Groq: \"reduce the length of the messages\"\n * - Cerebras: 400/413 status code (no body)\n * - Mistral: \"Prompt contains X tokens ... too large for model with Y maximum context length\"\n * - OpenRouter (most backends): \"maximum context length is X tokens\"\n * - OpenRouter/Poolside: \"Input length X exceeds the maximum allowed input length of Y tokens.\"\n * - Together AI: \"The input (X tokens) is longer than the model's context length (Y tokens).\"\n * - llama.cpp: \"exceeds the available context size\"\n * - LM Studio: \"greater than the context length\"\n * - Kimi For Coding: \"exceeded model token limit: X (requested: Y)\"\n * - DS4: \"Prompt has X tokens, but the configured context size is Y tokens\"\n * - DashScope/Qwen: \"Range of input length should be [1, X]\"\n *\n * **Unreliable detection:**\n * - z.ai: Sometimes accepts overflow silently (detectable via usage.input > contextWindow),\n *   sometimes returns rate limit errors. Pass contextWindow param to detect silent overflow.\n * - Xiaomi MiMo: Truncates input to fit contextWindow then returns stopReason \"length\" with\n *   output=0. Pass contextWindow param to detect via the \"filled context + zero output\" signal.\n * - Ollama: May truncate input silently for some setups, but may also return explicit\n *   overflow errors that match the patterns above. Silent truncation still cannot be\n *   detected here because we do not know the expected token count.\n *\n * ## Custom Providers\n *\n * If you've added custom models via settings.json, this function may not detect\n * overflow errors from those providers. To add support:\n *\n * 1. Send a request that exceeds the model's context window\n * 2. Check the errorMessage in the response\n * 3. Create a regex pattern that matches the error\n * 4. The pattern should be added to OVERFLOW_PATTERNS in this file, or\n *    check the errorMessage yourself before calling this function\n *\n * @param message - The assistant message to check\n * @param contextWindow - Optional context window size for detecting silent overflow (z.ai)\n * @returns true if the message indicates a context overflow\n */\nexport function isContextOverflow(message: AssistantMessage, contextWindow?: number): boolean {\n\t// Case 1: Check error message patterns\n\tif (message.stopReason === \"error\" && message.errorMessage) {\n\t\t// Skip messages matching known non-overflow patterns (e.g. throttling / rate-limit)\n\t\tconst isNonOverflow = NON_OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!));\n\t\tif (!isNonOverflow && OVERFLOW_PATTERNS.some((p) => p.test(message.errorMessage!))) {\n\t\t\treturn true;\n\t\t}\n\t\tconst usage = message.usage;\n\t\tconst hasTokenEvidence =\n\t\t\t(usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite) > 0;\n\t\tif (\n\t\t\t!isNonOverflow &&\n\t\t\thasTokenEvidence &&\n\t\t\tRESOURCE_EXHAUSTED_PATTERN.test(message.errorMessage) &&\n\t\t\t(contextWindow === undefined || contextWindow <= 0 || cursorZeroTokenCount(message) >= contextWindow * 0.5)\n\t\t) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 2: Silent overflow (z.ai style) - successful but usage exceeds context\n\tif (contextWindow && message.stopReason === \"stop\") {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens > contextWindow) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\t// Case 3: Length-stop overflow (Xiaomi MiMo style) - server truncates oversized input\n\t// to fit the context window, leaving no room for output. Returns stopReason \"length\"\n\t// with output=0 and input+cacheRead filling the context window.\n\tif (contextWindow && message.stopReason === \"length\" && message.usage.output === 0) {\n\t\tconst inputTokens = message.usage.input + message.usage.cacheRead;\n\t\tif (inputTokens >= contextWindow * 0.99) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n/**\n * Check whether a length stop ended below the caller or model's intended output limit.\n * Such responses may be caused by context pressure or provider-side truncation, so callers\n * can make one bounded compact-and-retry attempt. `desiredMaxOutput` must be the original\n * limit before any context-based clamping.\n */\nexport function isRecoverableLength(message: AssistantMessage, desiredMaxOutput: number): boolean {\n\treturn message.stopReason === \"length\" && desiredMaxOutput > 0 && message.usage.output < desiredMaxOutput;\n}\n\n/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n\nfunction cursorZeroTokenCount(message: {\n\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n}): number {\n\tconst usage = message.usage;\n\tif (!usage) return 0;\n\treturn (\n\t\tusage.totalTokens || (usage.input ?? 0) + (usage.output ?? 0) + (usage.cacheRead ?? 0) + (usage.cacheWrite ?? 0)\n\t);\n}\n\nexport function isCursorPayloadResourceExhausted(\n\tmessage: {\n\t\tstopReason?: string;\n\t\terrorMessage?: string;\n\t\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n\t},\n\t_estimateTokens: number,\n): boolean {\n\tif (message.stopReason !== \"error\" || !/resource.?exhausted/i.test(message.errorMessage || \"\")) {\n\t\treturn false;\n\t}\n\treturn !(cursorZeroTokenCount(message) > 0);\n}\n\n/**\n * Detects Cursor's verified usage-pool exhaustion signature: a token-bearing\n * `resource_exhausted` error while the conversation is well below the model\n * context window.\n */\nexport function isCursorQuotaResourceExhausted(\n\tmessage: {\n\t\tstopReason?: string;\n\t\terrorMessage?: string;\n\t\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n\t},\n\tcontextWindow: number,\n): boolean {\n\tconst tokens = cursorZeroTokenCount(message);\n\treturn (\n\t\tmessage.stopReason === \"error\" &&\n\t\tRESOURCE_EXHAUSTED_PATTERN.test(message.errorMessage || \"\") &&\n\t\tcontextWindow > 0 &&\n\t\ttokens > 0 &&\n\t\ttokens < contextWindow * 0.5\n\t);\n}\n\nexport function isCursorZeroTokenResourceExhausted(message: {\n\tstopReason?: string;\n\terrorMessage?: string;\n\tusage?: { input?: number; output?: number; cacheRead?: number; cacheWrite?: number; totalTokens?: number };\n}): boolean {\n\tif (message.stopReason !== \"error\" || !/resource.?exhausted/i.test(message.errorMessage || \"\")) {\n\t\treturn false;\n\t}\n\treturn !(cursorZeroTokenCount(message) > 0);\n}\n\nexport function shouldSkipProviderFallbackForCursorZeroRe(options: { sameModelRemint?: boolean } | undefined): boolean {\n\treturn options?.sameModelRemint === true;\n}\n\nexport function shouldRetryOverflowWithoutCompact(compacted: boolean, errorMessage: string): boolean {\n\tif (compacted) return false;\n\treturn /Nothing to compact/i.test(errorMessage || \"\");\n}\n\nexport function cursorOverflowCompactionSettings<T extends { keepRecentTokens?: number; restorationEnabled?: boolean }>(\n\tsettings: T,\n\tprovider: string | undefined,\n\treason: string | undefined,\n): T {\n\tif (reason !== \"overflow\") return settings;\n\tif (provider !== \"cursor\" && provider !== \"cursor-cli-oauth\") return settings;\n\treturn { ...settings, keepRecentTokens: 0, restorationEnabled: false };\n}\n"]}

package/node_modules/@earendil-works/pi-ai/dist/utils/retry.d.ts

@@ -1,4 +1,7 @@
 import type { AssistantMessage } from "../types.ts";
+export declare const PROVIDER_REQUEST_ID_MARKER = "request id:";
+export declare function formatProviderRequestId(label: string, id: string): string;
+export declare function stripProviderRequestIds(text: string): string;
 /**
  * OpenAI hard account-quota exhaustion (senpi#1969). The wire evidence is a 429
  * whose body is `{"type":"usage_limit_reached","message":"The usage limit has

package/node_modules/@earendil-works/pi-ai/dist/utils/retry.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"retry.d.ts","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAWpD;;;;;;;;;GASG;AACH,eAAO,MAAM,sBAAsB;IAClC,+EAA+E;aAC/E,KAAK,YAAG,qBAAqB,EAAE,oBAAoB;IACnD,+EAA+E;aAC/E,OAAO,YAAG,qBAAqB,EAAE,oBAAoB,EAAE,8BAA8B;CAC5E,CAAC;AA0CX,wBAAgB,wBAAwB,CAAC,YAAY,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,CAElF;AAoID;;;;;;GAMG;AACH,MAAM,WAAW,WAAW;IAC3B,OAAO,EAAE,OAAO,CAAC;IACjB,qFAAqF;IACrF,UAAU,EAAE,MAAM,CAAC;IACnB,0FAA0F;IAC1F,WAAW,EAAE,MAAM,CAAC;IACpB,+EAA+E;IAC/E,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,oEAAoE;IACpE,MAAM,CAAC,EAAE,MAAM,MAAM,CAAC;CACtB;AAED,eAAO,MAAM,gCAAgC,QAAS,CAAC;AAEvD;;;;;;GAMG;AACH,wBAAgB,YAAY,CAC3B,MAAM,EAAE,IAAI,CAAC,WAAW,EAAE,aAAa,GAAG,iBAAiB,GAAG,QAAQ,CAAC,EACvE,OAAO,EAAE,MAAM,GACb,MAAM,CAMR;AAED,kFAAkF;AAClF,MAAM,WAAW,cAAc;IAC9B,0EAA0E;IAC1E,gBAAgB,CAAC,EAAE,CAClB,OAAO,EAAE,MAAM,EACf,WAAW,EAAE,MAAM,EACnB,OAAO,EAAE,MAAM,EACf,YAAY,EAAE,MAAM,KAChB,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC1B,mFAAmF;IACnF,mBAAmB,CAAC,EAAE,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACjD,mFAAmF;IACnF,eAAe,CAAC,EAAE,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,MAAM,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CACnG;AA0BD;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,wBAAsB,kBAAkB,CAAC,CAAC,EACzC,OAAO,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,EACzB,WAAW,EAAE,CAAC,KAAK,EAAE,OAAO,KAAK,OAAO,EACxC,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,CAAC,CAAC,CAiCZ;AAOD;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAsB,kBAAkB,CACvC,OAAO,EAAE,MAAM,OAAO,CAAC,gBAAgB,CAAC,EACxC,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,gBAAgB,CAAC,CA6C3B;AAED;;;;;;;;GAQG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAU5E;AAgBD,wBAAgB,0BAA0B,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAE7E;AA8BD,MAAM,WAAW,+BAA+B;IAC/C,sDAAsD;IACtD,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,iFAAiF;IACjF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,QAAQ,CAAC,EAAE,wBAAwB,GAAG,iBAAiB,CAAC;CACxD;AAED,wBAAgB,mBAAmB,CAAC,SAAS,EAAE,MAAM,GAAG,MAAM,CAI7D;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,4BAA4B,CAC3C,YAAY,EAAE,MAAM,GAAG,SAAS,EAChC,OAAO,GAAE,+BAAoC,GAC3C,MAAM,GAAG,SAAS,CAuBpB;AAED;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAKzE;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CAAC,YAAY,EAAE,MAAM,GAAG,OAAO,CAErE;AAED;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAAC,YAAY,EAAE,MAAM,GAAG,eAAe,GAAG,WAAW,GAAG,SAAS,CAKpG"}
+{"version":3,"file":"retry.d.ts","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAMpD,eAAO,MAAM,0BAA0B,gBAAgB,CAAC;AAIxD,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,GAAG,MAAM,CAEzE;AAED,wBAAgB,uBAAuB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAE5D;AAUD;;;;;;;;;GASG;AACH,eAAO,MAAM,sBAAsB;IAClC,+EAA+E;aAC/E,KAAK,YAAG,qBAAqB,EAAE,oBAAoB;IACnD,+EAA+E;aAC/E,OAAO,YAAG,qBAAqB,EAAE,oBAAoB,EAAE,8BAA8B;CAC5E,CAAC;AA0CX,wBAAgB,wBAAwB,CAAC,YAAY,EAAE,MAAM,GAAG,SAAS,GAAG,OAAO,CAElF;AAoID;;;;;;GAMG;AACH,MAAM,WAAW,WAAW;IAC3B,OAAO,EAAE,OAAO,CAAC;IACjB,qFAAqF;IACrF,UAAU,EAAE,MAAM,CAAC;IACnB,0FAA0F;IAC1F,WAAW,EAAE,MAAM,CAAC;IACpB,+EAA+E;IAC/E,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,oEAAoE;IACpE,MAAM,CAAC,EAAE,MAAM,MAAM,CAAC;CACtB;AAED,eAAO,MAAM,gCAAgC,QAAS,CAAC;AAEvD;;;;;;GAMG;AACH,wBAAgB,YAAY,CAC3B,MAAM,EAAE,IAAI,CAAC,WAAW,EAAE,aAAa,GAAG,iBAAiB,GAAG,QAAQ,CAAC,EACvE,OAAO,EAAE,MAAM,GACb,MAAM,CAMR;AAED,kFAAkF;AAClF,MAAM,WAAW,cAAc;IAC9B,0EAA0E;IAC1E,gBAAgB,CAAC,EAAE,CAClB,OAAO,EAAE,MAAM,EACf,WAAW,EAAE,MAAM,EACnB,OAAO,EAAE,MAAM,EACf,YAAY,EAAE,MAAM,KAChB,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC1B,mFAAmF;IACnF,mBAAmB,CAAC,EAAE,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IACjD,mFAAmF;IACnF,eAAe,CAAC,EAAE,CAAC,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,MAAM,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CACnG;AA0BD;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,wBAAsB,kBAAkB,CAAC,CAAC,EACzC,OAAO,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,EACzB,WAAW,EAAE,CAAC,KAAK,EAAE,OAAO,KAAK,OAAO,EACxC,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,CAAC,CAAC,CAiCZ;AAOD;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAsB,kBAAkB,CACvC,OAAO,EAAE,MAAM,OAAO,CAAC,gBAAgB,CAAC,EACxC,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,EAC/B,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,gBAAgB,CAAC,CA6C3B;AAED;;;;;;;;GAQG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAU5E;AAgBD,wBAAgB,0BAA0B,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAE7E;AA8BD,MAAM,WAAW,+BAA+B;IAC/C,sDAAsD;IACtD,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,iFAAiF;IACjF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,QAAQ,CAAC,EAAE,wBAAwB,GAAG,iBAAiB,CAAC;CACxD;AAED,wBAAgB,mBAAmB,CAAC,SAAS,EAAE,MAAM,GAAG,MAAM,CAI7D;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,4BAA4B,CAC3C,YAAY,EAAE,MAAM,GAAG,SAAS,EAChC,OAAO,GAAE,+BAAoC,GAC3C,MAAM,GAAG,SAAS,CAuBpB;AAED;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAKzE;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CAAC,YAAY,EAAE,MAAM,GAAG,OAAO,CAErE;AAED;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAAC,YAAY,EAAE,MAAM,GAAG,eAAe,GAAG,WAAW,GAAG,SAAS,CAMpG"}

package/node_modules/@earendil-works/pi-ai/dist/utils/retry.js

@@ -1,4 +1,15 @@
 import { FORWARDED_EMPTY_RESPONSE_ERROR, FORWARDED_EMPTY_TOOL_USE_ERROR } from "./empty-response-errors.js";
+// A provider's support request id is opaque hex like `C6FD:AB660:AEB5548:6ABA4D81`.
+// It can contain `429` or `500`, which message classifiers read as HTTP statuses,
+// so every id is rendered behind this marker and removed before classification.
+export const PROVIDER_REQUEST_ID_MARKER = "request id:";
+const REQUEST_ID_SEGMENT = /request id: [^\s,;)]+/gi;
+export function formatProviderRequestId(label, id) {
+    return `${label} ${PROVIDER_REQUEST_ID_MARKER} ${id}`;
+}
+export function stripProviderRequestIds(text) {
+    return text.replace(REQUEST_ID_SEGMENT, PROVIDER_REQUEST_ID_MARKER);
+}
 function buildProviderErrorPattern(patterns) {
     return new RegExp(patterns.join("|"), "i");
 }
@@ -56,7 +67,7 @@ const QUOTA_EXHAUSTION_PATTERNS = [
 ];
 const QUOTA_EXHAUSTION_PATTERN = buildProviderErrorPattern(QUOTA_EXHAUSTION_PATTERNS);
 export function isQuotaExhaustionMessage(errorMessage) {
-    return errorMessage !== undefined && QUOTA_EXHAUSTION_PATTERN.test(errorMessage);
+    return errorMessage !== undefined && QUOTA_EXHAUSTION_PATTERN.test(stripProviderRequestIds(errorMessage));
 }
 const NON_RETRYABLE_PROVIDER_ERROR_PATTERN = buildProviderErrorPattern([
     ...QUOTA_EXHAUSTION_PATTERNS,
@@ -463,9 +474,10 @@ export function isRetryableErrorMessage(errorMessage) {
 export function classifyErrorMessage(errorMessage) {
     if (!errorMessage)
         return "unknown";
-    if (NON_RETRYABLE_PROVIDER_ERROR_PATTERN.test(errorMessage))
+    const text = stripProviderRequestIds(errorMessage);
+    if (NON_RETRYABLE_PROVIDER_ERROR_PATTERN.test(text))
         return "non-retryable";
-    if (RETRYABLE_PROVIDER_ERROR_PATTERN.test(errorMessage))
+    if (RETRYABLE_PROVIDER_ERROR_PATTERN.test(text))
         return "retryable";
     return "unknown";
 }