@draht/ai 2026.7.7 → 2026.7.12-1

package/dist/utils/oauth/xai.js

@@ -0,0 +1,339 @@
+/**
+ * xAI (Grok) OAuth flow — SuperGrok / X Premium+ subscription login.
+ *
+ * Mirrors the flow the official Grok CLI uses: PKCE authorization code
+ * against auth.x.ai's OIDC discovery document. Tokens are scoped for the
+ * Grok CLI proxy (cli-chat-proxy.grok.com) rather than the pay-per-token
+ * api.x.ai endpoint, so `toAuth`/`modifyModels` route requests there.
+ *
+ * NOTE: This module uses Node.js http.createServer for the OAuth callback
+ * server. It is only intended for CLI use, not browser environments.
+ */
+import { getProviderEnvValue } from "../provider-env.js";
+import { oauthErrorHtml, oauthSuccessHtml } from "./oauth-page.js";
+import { generatePKCE } from "./pkce.js";
+let nodeApis = null;
+let nodeApisPromise = null;
+const OAUTH_ISSUER = "https://auth.x.ai";
+const DISCOVERY_URL = `${OAUTH_ISSUER}/.well-known/openid-configuration`;
+const CLIENT_ID = "b1a00492-073a-47ea-816f-4c329264a828";
+const SCOPE = "openid profile email offline_access grok-cli:access api:access";
+const CALLBACK_HOST = getProviderEnvValue("PI_OAUTH_CALLBACK_HOST") || "127.0.0.1";
+const CALLBACK_PORT = 56121;
+const CALLBACK_PATH = "/callback";
+const REFRESH_SKEW_MS = 2 * 60 * 1000;
+/** OAuth-authenticated requests are proxied through the Grok CLI endpoint, not api.x.ai directly. */
+export const XAI_OAUTH_BASE_URL = "https://cli-chat-proxy.grok.com/v1";
+async function getNodeApis() {
+    if (nodeApis)
+        return nodeApis;
+    if (!nodeApisPromise) {
+        if (typeof process === "undefined" || (!process.versions?.node && !process.versions?.bun)) {
+            throw new Error("xAI OAuth is only available in Node.js environments");
+        }
+        nodeApisPromise = import("node:http").then((httpModule) => ({
+            createServer: httpModule.createServer,
+        }));
+    }
+    nodeApis = await nodeApisPromise;
+    return nodeApis;
+}
+/** Reject discovery responses that don't point back at x.ai, in case DNS/discovery is ever compromised. */
+function validateXaiEndpoint(url) {
+    const parsed = new URL(url);
+    const host = parsed.hostname.toLowerCase();
+    if (parsed.protocol !== "https:" || (host !== "x.ai" && !host.endsWith(".x.ai"))) {
+        throw new Error(`xAI OAuth discovery returned an unexpected endpoint: ${url}`);
+    }
+    return url;
+}
+async function discoverXaiEndpoints() {
+    const response = await fetch(DISCOVERY_URL, { headers: { Accept: "application/json" } });
+    if (!response.ok) {
+        throw new Error(`xAI OAuth discovery failed: ${response.status} ${await response.text()}`);
+    }
+    const data = (await response.json());
+    if (!data.authorization_endpoint || !data.token_endpoint) {
+        throw new Error("xAI OAuth discovery response did not include authorization/token endpoints");
+    }
+    return {
+        authorizationEndpoint: validateXaiEndpoint(data.authorization_endpoint),
+        tokenEndpoint: validateXaiEndpoint(data.token_endpoint),
+    };
+}
+function parseAuthorizationInput(input) {
+    const value = input.trim();
+    if (!value)
+        return {};
+    try {
+        const url = new URL(value);
+        return {
+            code: url.searchParams.get("code") ?? undefined,
+            state: url.searchParams.get("state") ?? undefined,
+        };
+    }
+    catch {
+        // not a URL
+    }
+    if (value.includes("code=")) {
+        const params = new URLSearchParams(value);
+        return {
+            code: params.get("code") ?? undefined,
+            state: params.get("state") ?? undefined,
+        };
+    }
+    return { code: value };
+}
+async function startCallbackServer(expectedState) {
+    const { createServer } = await getNodeApis();
+    return new Promise((resolve, reject) => {
+        let settleWait;
+        const waitForCodePromise = new Promise((resolveWait) => {
+            let settled = false;
+            settleWait = (value) => {
+                if (settled)
+                    return;
+                settled = true;
+                resolveWait(value);
+            };
+        });
+        const server = createServer((req, res) => {
+            try {
+                const url = new URL(req.url || "", `http://${CALLBACK_HOST}`);
+                if (url.pathname !== CALLBACK_PATH) {
+                    res.writeHead(404, { "Content-Type": "text/html; charset=utf-8" });
+                    res.end(oauthErrorHtml("Callback route not found."));
+                    return;
+                }
+                const code = url.searchParams.get("code");
+                const state = url.searchParams.get("state");
+                const error = url.searchParams.get("error");
+                if (error) {
+                    res.writeHead(400, { "Content-Type": "text/html; charset=utf-8" });
+                    res.end(oauthErrorHtml("xAI authentication did not complete.", `Error: ${error}`));
+                    return;
+                }
+                if (!code || !state) {
+                    res.writeHead(400, { "Content-Type": "text/html; charset=utf-8" });
+                    res.end(oauthErrorHtml("Missing code or state parameter."));
+                    return;
+                }
+                if (state !== expectedState) {
+                    res.writeHead(400, { "Content-Type": "text/html; charset=utf-8" });
+                    res.end(oauthErrorHtml("State mismatch."));
+                    return;
+                }
+                res.writeHead(200, { "Content-Type": "text/html; charset=utf-8" });
+                res.end(oauthSuccessHtml("xAI authentication completed. You can close this window."));
+                settleWait?.({ code, state });
+            }
+            catch {
+                res.writeHead(500, { "Content-Type": "text/plain; charset=utf-8" });
+                res.end("Internal error");
+            }
+        });
+        server.on("error", (err) => {
+            reject(err);
+        });
+        server.listen(CALLBACK_PORT, CALLBACK_HOST, () => {
+            resolve({
+                server,
+                redirectUri: `http://${CALLBACK_HOST}:${CALLBACK_PORT}${CALLBACK_PATH}`,
+                cancelWait: () => {
+                    settleWait?.(null);
+                },
+                waitForCode: () => waitForCodePromise,
+            });
+        });
+    });
+}
+async function postForm(url, body) {
+    const response = await fetch(url, {
+        method: "POST",
+        headers: {
+            "Content-Type": "application/x-www-form-urlencoded",
+            Accept: "application/json",
+        },
+        body: new URLSearchParams(body).toString(),
+        signal: AbortSignal.timeout(30_000),
+    });
+    const responseBody = await response.text();
+    if (!response.ok) {
+        throw new Error(`xAI token request failed. status=${response.status}; url=${url}; body=${responseBody}`);
+    }
+    return responseBody;
+}
+function credentialsFromTokenPayload(data, fallbackRefresh = "") {
+    if (!data.access_token) {
+        throw new Error("xAI token response did not include an access token");
+    }
+    const refresh = data.refresh_token || fallbackRefresh;
+    if (!refresh) {
+        throw new Error("xAI token response did not include a refresh token");
+    }
+    return {
+        access: data.access_token,
+        refresh,
+        expires: Date.now() + (data.expires_in ?? 3600) * 1000 - REFRESH_SKEW_MS,
+    };
+}
+/**
+ * Login with xAI OAuth (SuperGrok / X Premium+ subscription, authorization code + PKCE)
+ */
+export async function loginXai(options) {
+    const discovery = await discoverXaiEndpoints();
+    const { verifier, challenge } = await generatePKCE();
+    const server = await startCallbackServer(verifier);
+    try {
+        const authParams = new URLSearchParams({
+            response_type: "code",
+            client_id: CLIENT_ID,
+            redirect_uri: server.redirectUri,
+            scope: SCOPE,
+            code_challenge: challenge,
+            code_challenge_method: "S256",
+            state: verifier,
+        });
+        options.onAuth({
+            url: `${discovery.authorizationEndpoint}?${authParams.toString()}`,
+            instructions: "Sign in with your SuperGrok or X Premium+ account. If the browser is on another machine, paste the final redirect URL here.",
+        });
+        let code;
+        if (options.onManualCodeInput) {
+            let manualInput;
+            let manualError;
+            const manualPromise = options
+                .onManualCodeInput()
+                .then((input) => {
+                manualInput = input;
+                server.cancelWait();
+            })
+                .catch((err) => {
+                manualError = err instanceof Error ? err : new Error(String(err));
+                server.cancelWait();
+            });
+            const result = await server.waitForCode();
+            if (manualError) {
+                throw manualError;
+            }
+            if (result?.code) {
+                code = result.code;
+            }
+            else if (manualInput) {
+                const parsed = parseAuthorizationInput(manualInput);
+                if (parsed.state && parsed.state !== verifier) {
+                    throw new Error("OAuth state mismatch");
+                }
+                code = parsed.code;
+            }
+            if (!code) {
+                await manualPromise;
+                if (manualError) {
+                    throw manualError;
+                }
+                if (manualInput) {
+                    const parsed = parseAuthorizationInput(manualInput);
+                    if (parsed.state && parsed.state !== verifier) {
+                        throw new Error("OAuth state mismatch");
+                    }
+                    code = parsed.code;
+                }
+            }
+        }
+        else {
+            const result = await server.waitForCode();
+            if (result?.code) {
+                code = result.code;
+            }
+        }
+        if (!code) {
+            const input = await options.onPrompt({
+                message: "Paste the authorization code or full redirect URL:",
+                placeholder: server.redirectUri,
+            });
+            const parsed = parseAuthorizationInput(input);
+            if (parsed.state && parsed.state !== verifier) {
+                throw new Error("OAuth state mismatch");
+            }
+            code = parsed.code;
+        }
+        if (!code) {
+            throw new Error("Missing authorization code");
+        }
+        options.onProgress?.("Exchanging xAI authorization code...");
+        const responseBody = await postForm(discovery.tokenEndpoint, {
+            grant_type: "authorization_code",
+            client_id: CLIENT_ID,
+            code,
+            redirect_uri: server.redirectUri,
+            code_verifier: verifier,
+        });
+        return credentialsFromTokenPayload(JSON.parse(responseBody));
+    }
+    finally {
+        server.server.close();
+    }
+}
+/**
+ * Refresh xAI OAuth token
+ */
+export async function refreshXaiToken(refreshToken) {
+    const discovery = await discoverXaiEndpoints();
+    const responseBody = await postForm(discovery.tokenEndpoint, {
+        grant_type: "refresh_token",
+        client_id: CLIENT_ID,
+        refresh_token: refreshToken,
+    });
+    return credentialsFromTokenPayload(JSON.parse(responseBody), refreshToken);
+}
+export const xaiOAuth = {
+    name: "xAI (Grok)",
+    async login(callbacks) {
+        // The manual_code prompt races the local callback server; abort it once
+        // the flow settles so the UI can dismiss the pending input.
+        const manualAbort = new AbortController();
+        try {
+            const credentials = await loginXai({
+                onAuth: (info) => callbacks.notify({ type: "auth_url", url: info.url, instructions: info.instructions }),
+                onProgress: (message) => callbacks.notify({ type: "progress", message }),
+                onPrompt: (prompt) => callbacks.prompt({ type: "text", message: prompt.message, placeholder: prompt.placeholder }),
+                onManualCodeInput: () => callbacks.prompt({
+                    type: "manual_code",
+                    message: "Complete login in your browser, or paste the authorization code / redirect URL here:",
+                    placeholder: `http://${CALLBACK_HOST}:${CALLBACK_PORT}${CALLBACK_PATH}`,
+                    signal: manualAbort.signal,
+                }),
+            });
+            return { ...credentials, type: "oauth" };
+        }
+        finally {
+            manualAbort.abort();
+        }
+    },
+    async refresh(credential) {
+        return { ...(await refreshXaiToken(credential.refresh)), type: "oauth" };
+    },
+    async toAuth(credential) {
+        return { apiKey: credential.access, baseUrl: XAI_OAUTH_BASE_URL };
+    },
+};
+export const xaiOAuthProvider = {
+    id: "xai",
+    name: "xAI (Grok)",
+    usesCallbackServer: true,
+    async login(callbacks) {
+        return loginXai({
+            onAuth: callbacks.onAuth,
+            onPrompt: callbacks.onPrompt,
+            onProgress: callbacks.onProgress,
+            onManualCodeInput: callbacks.onManualCodeInput,
+        });
+    },
+    async refreshToken(credentials) {
+        return refreshXaiToken(credentials.refresh);
+    },
+    getApiKey(credentials) {
+        return credentials.access;
+    },
+};
+//# sourceMappingURL=xai.js.map

package/dist/utils/oauth/xai.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"xai.js","sourceRoot":"","sources":["../../../src/utils/oauth/xai.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAIH,OAAO,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AACzD,OAAO,EAAE,cAAc,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAAE,YAAY,EAAE,MAAM,WAAW,CAAC;AAOzC,IAAI,QAAQ,GAAoB,IAAI,CAAC;AACrC,IAAI,eAAe,GAA6B,IAAI,CAAC;AAErD,MAAM,YAAY,GAAG,mBAAmB,CAAC;AACzC,MAAM,aAAa,GAAG,GAAG,YAAY,mCAAmC,CAAC;AACzE,MAAM,SAAS,GAAG,sCAAsC,CAAC;AACzD,MAAM,KAAK,GAAG,gEAAgE,CAAC;AAC/E,MAAM,aAAa,GAAG,mBAAmB,CAAC,wBAAwB,CAAC,IAAI,WAAW,CAAC;AACnF,MAAM,aAAa,GAAG,KAAK,CAAC;AAC5B,MAAM,aAAa,GAAG,WAAW,CAAC;AAClC,MAAM,eAAe,GAAG,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC;AAEtC,qGAAqG;AACrG,MAAM,CAAC,MAAM,kBAAkB,GAAG,oCAAoC,CAAC;AAKvE,KAAK,UAAU,WAAW;IACzB,IAAI,QAAQ;QAAE,OAAO,QAAQ,CAAC;IAC9B,IAAI,CAAC,eAAe,EAAE,CAAC;QACtB,IAAI,OAAO,OAAO,KAAK,WAAW,IAAI,CAAC,CAAC,OAAO,CAAC,QAAQ,EAAE,IAAI,IAAI,CAAC,OAAO,CAAC,QAAQ,EAAE,GAAG,CAAC,EAAE,CAAC;YAC3F,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;QACxE,CAAC;QACD,eAAe,GAAG,MAAM,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;YAC3D,YAAY,EAAE,UAAU,CAAC,YAAY;SACrC,CAAC,CAAC,CAAC;IACL,CAAC;IACD,QAAQ,GAAG,MAAM,eAAe,CAAC;IACjC,OAAO,QAAQ,CAAC;AACjB,CAAC;AAED,2GAA2G;AAC3G,SAAS,mBAAmB,CAAC,GAAW;IACvC,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC;IAC5B,MAAM,IAAI,GAAG,MAAM,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC;IAC3C,IAAI,MAAM,CAAC,QAAQ,KAAK,QAAQ,IAAI,CAAC,IAAI,KAAK,MAAM,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,EAAE,CAAC;QAClF,MAAM,IAAI,KAAK,CAAC,wDAAwD,GAAG,EAAE,CAAC,CAAC;IAChF,CAAC;IACD,OAAO,GAAG,CAAC;AACZ,CAAC;AAED,KAAK,UAAU,oBAAoB;IAClC,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,aAAa,EAAE,EAAE,OAAO,EAAE,EAAE,MAAM,EAAE,kBAAkB,EAAE,EAAE,CAAC,CAAC;IACzF,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QAClB,MAAM,IAAI,KAAK,CAAC,+BAA+B,QAAQ,CAAC,MAAM,IAAI,MAAM,QAAQ,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;IAC5F,CAAC;IAED,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAiE,CAAC;IACrG,IAAI,CAAC,IAAI,CAAC,sBAAsB,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,CAAC;QAC1D,MAAM,IAAI,KAAK,CAAC,4EAA4E,CAAC,CAAC;IAC/F,CAAC;IAED,OAAO;QACN,qBAAqB,EAAE,mBAAmB,CAAC,IAAI,CAAC,sBAAsB,CAAC;QACvE,aAAa,EAAE,mBAAmB,CAAC,IAAI,CAAC,cAAc,CAAC;KACvD,CAAC;AACH,CAAC;AAED,SAAS,uBAAuB,CAAC,KAAa;IAC7C,MAAM,KAAK,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC;IAC3B,IAAI,CAAC,KAAK;QAAE,OAAO,EAAE,CAAC;IAEtB,IAAI,CAAC;QACJ,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,CAAC;QAC3B,OAAO;YACN,IAAI,EAAE,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,SAAS;YAC/C,KAAK,EAAE,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,SAAS;SACjD,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACR,YAAY;IACb,CAAC;IAED,IAAI,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QAC7B,MAAM,MAAM,GAAG,IAAI,eAAe,CAAC,KAAK,CAAC,CAAC;QAC1C,OAAO;YACN,IAAI,EAAE,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,SAAS;YACrC,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,SAAS;SACvC,CAAC;IACH,CAAC;IAED,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC;AACxB,CAAC;AASD,KAAK,UAAU,mBAAmB,CAAC,aAAqB;IACvD,MAAM,EAAE,YAAY,EAAE,GAAG,MAAM,WAAW,EAAE,CAAC;IAE7C,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACtC,IAAI,UAAiF,CAAC;QACtF,MAAM,kBAAkB,GAAG,IAAI,OAAO,CAAyC,CAAC,WAAW,EAAE,EAAE;YAC9F,IAAI,OAAO,GAAG,KAAK,CAAC;YACpB,UAAU,GAAG,CAAC,KAAK,EAAE,EAAE;gBACtB,IAAI,OAAO;oBAAE,OAAO;gBACpB,OAAO,GAAG,IAAI,CAAC;gBACf,WAAW,CAAC,KAAK,CAAC,CAAC;YACpB,CAAC,CAAC;QACH,CAAC,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,YAAY,CAAC,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE;YACxC,IAAI,CAAC;gBACJ,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,GAAG,IAAI,EAAE,EAAE,UAAU,aAAa,EAAE,CAAC,CAAC;gBAC9D,IAAI,GAAG,CAAC,QAAQ,KAAK,aAAa,EAAE,CAAC;oBACpC,GAAG,CAAC,SAAS,CAAC,GAAG,EAAE,EAAE,cAAc,EAAE,0BAA0B,EAAE,CAAC,CAAC;oBACnE,GAAG,CAAC,GAAG,CAAC,cAAc,CAAC,2BAA2B,CAAC,CAAC,CAAC;oBACrD,OAAO;gBACR,CAAC;gBAED,MAAM,IAAI,GAAG,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;gBAC1C,MAAM,KAAK,GAAG,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;gBAC5C,MAAM,KAAK,GAAG,GAAG,CAAC,YAAY,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;gBAE5C,IAAI,KAAK,EAAE,CAAC;oBACX,GAAG,CAAC,SAAS,CAAC,GAAG,EAAE,EAAE,cAAc,EAAE,0BAA0B,EAAE,CAAC,CAAC;oBACnE,GAAG,CAAC,GAAG,CAAC,cAAc,CAAC,sCAAsC,EAAE,UAAU,KAAK,EAAE,CAAC,CAAC,CAAC;oBACnF,OAAO;gBACR,CAAC;gBAED,IAAI,CAAC,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;oBACrB,GAAG,CAAC,SAAS,CAAC,GAAG,EAAE,EAAE,cAAc,EAAE,0BAA0B,EAAE,CAAC,CAAC;oBACnE,GAAG,CAAC,GAAG,CAAC,cAAc,CAAC,kCAAkC,CAAC,CAAC,CAAC;oBAC5D,OAAO;gBACR,CAAC;gBAED,IAAI,KAAK,KAAK,aAAa,EAAE,CAAC;oBAC7B,GAAG,CAAC,SAAS,CAAC,GAAG,EAAE,EAAE,cAAc,EAAE,0BAA0B,EAAE,CAAC,CAAC;oBACnE,GAAG,CAAC,GAAG,CAAC,cAAc,CAAC,iBAAiB,CAAC,CAAC,CAAC;oBAC3C,OAAO;gBACR,CAAC;gBAED,GAAG,CAAC,SAAS,CAAC,GAAG,EAAE,EAAE,cAAc,EAAE,0BAA0B,EAAE,CAAC,CAAC;gBACnE,GAAG,CAAC,GAAG,CAAC,gBAAgB,CAAC,0DAA0D,CAAC,CAAC,CAAC;gBACtF,UAAU,EAAE,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC;YAC/B,CAAC;YAAC,MAAM,CAAC;gBACR,GAAG,CAAC,SAAS,CAAC,GAAG,EAAE,EAAE,cAAc,EAAE,2BAA2B,EAAE,CAAC,CAAC;gBACpE,GAAG,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC;YAC3B,CAAC;QACF,CAAC,CAAC,CAAC;QAEH,MAAM,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,GAAG,EAAE,EAAE;YAC1B,MAAM,CAAC,GAAG,CAAC,CAAC;QACb,CAAC,CAAC,CAAC;QAEH,MAAM,CAAC,MAAM,CAAC,aAAa,EAAE,aAAa,EAAE,GAAG,EAAE;YAChD,OAAO,CAAC;gBACP,MAAM;gBACN,WAAW,EAAE,UAAU,aAAa,IAAI,aAAa,GAAG,aAAa,EAAE;gBACvE,UAAU,EAAE,GAAG,EAAE;oBAChB,UAAU,EAAE,CAAC,IAAI,CAAC,CAAC;gBACpB,CAAC;gBACD,WAAW,EAAE,GAAG,EAAE,CAAC,kBAAkB;aACrC,CAAC,CAAC;QACJ,CAAC,CAAC,CAAC;IACJ,CAAC,CAAC,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,QAAQ,CAAC,GAAW,EAAE,IAA4B;IAChE,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,EAAE;QACjC,MAAM,EAAE,MAAM;QACd,OAAO,EAAE;YACR,cAAc,EAAE,mCAAmC;YACnD,MAAM,EAAE,kBAAkB;SAC1B;QACD,IAAI,EAAE,IAAI,eAAe,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE;QAC1C,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC;KACnC,CAAC,CAAC;IAEH,MAAM,YAAY,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;IAC3C,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QAClB,MAAM,IAAI,KAAK,CAAC,oCAAoC,QAAQ,CAAC,MAAM,SAAS,GAAG,UAAU,YAAY,EAAE,CAAC,CAAC;IAC1G,CAAC;IACD,OAAO,YAAY,CAAC;AACrB,CAAC;AAED,SAAS,2BAA2B,CAAC,IAAqB,EAAE,eAAe,GAAG,EAAE;IAC/E,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;QACxB,MAAM,IAAI,KAAK,CAAC,oDAAoD,CAAC,CAAC;IACvE,CAAC;IAED,MAAM,OAAO,GAAG,IAAI,CAAC,aAAa,IAAI,eAAe,CAAC;IACtD,IAAI,CAAC,OAAO,EAAE,CAAC;QACd,MAAM,IAAI,KAAK,CAAC,oDAAoD,CAAC,CAAC;IACvE,CAAC;IAED,OAAO;QACN,MAAM,EAAE,IAAI,CAAC,YAAY;QACzB,OAAO;QACP,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,CAAC,GAAG,IAAI,GAAG,eAAe;KACxE,CAAC;AACH,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,OAK9B;IACA,MAAM,SAAS,GAAG,MAAM,oBAAoB,EAAE,CAAC;IAC/C,MAAM,EAAE,QAAQ,EAAE,SAAS,EAAE,GAAG,MAAM,YAAY,EAAE,CAAC;IACrD,MAAM,MAAM,GAAG,MAAM,mBAAmB,CAAC,QAAQ,CAAC,CAAC;IAEnD,IAAI,CAAC;QACJ,MAAM,UAAU,GAAG,IAAI,eAAe,CAAC;YACtC,aAAa,EAAE,MAAM;YACrB,SAAS,EAAE,SAAS;YACpB,YAAY,EAAE,MAAM,CAAC,WAAW;YAChC,KAAK,EAAE,KAAK;YACZ,cAAc,EAAE,SAAS;YACzB,qBAAqB,EAAE,MAAM;YAC7B,KAAK,EAAE,QAAQ;SACf,CAAC,CAAC;QAEH,OAAO,CAAC,MAAM,CAAC;YACd,GAAG,EAAE,GAAG,SAAS,CAAC,qBAAqB,IAAI,UAAU,CAAC,QAAQ,EAAE,EAAE;YAClE,YAAY,EACX,6HAA6H;SAC9H,CAAC,CAAC;QAEH,IAAI,IAAwB,CAAC;QAE7B,IAAI,OAAO,CAAC,iBAAiB,EAAE,CAAC;YAC/B,IAAI,WAA+B,CAAC;YACpC,IAAI,WAA8B,CAAC;YACnC,MAAM,aAAa,GAAG,OAAO;iBAC3B,iBAAiB,EAAE;iBACnB,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE;gBACf,WAAW,GAAG,KAAK,CAAC;gBACpB,MAAM,CAAC,UAAU,EAAE,CAAC;YACrB,CAAC,CAAC;iBACD,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;gBACd,WAAW,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;gBAClE,MAAM,CAAC,UAAU,EAAE,CAAC;YACrB,CAAC,CAAC,CAAC;YAEJ,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,WAAW,EAAE,CAAC;YAE1C,IAAI,WAAW,EAAE,CAAC;gBACjB,MAAM,WAAW,CAAC;YACnB,CAAC;YAED,IAAI,MAAM,EAAE,IAAI,EAAE,CAAC;gBAClB,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC;YACpB,CAAC;iBAAM,IAAI,WAAW,EAAE,CAAC;gBACxB,MAAM,MAAM,GAAG,uBAAuB,CAAC,WAAW,CAAC,CAAC;gBACpD,IAAI,MAAM,CAAC,KAAK,IAAI,MAAM,CAAC,KAAK,KAAK,QAAQ,EAAE,CAAC;oBAC/C,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,CAAC;gBACzC,CAAC;gBACD,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC;YACpB,CAAC;YAED,IAAI,CAAC,IAAI,EAAE,CAAC;gBACX,MAAM,aAAa,CAAC;gBACpB,IAAI,WAAW,EAAE,CAAC;oBACjB,MAAM,WAAW,CAAC;gBACnB,CAAC;gBACD,IAAI,WAAW,EAAE,CAAC;oBACjB,MAAM,MAAM,GAAG,uBAAuB,CAAC,WAAW,CAAC,CAAC;oBACpD,IAAI,MAAM,CAAC,KAAK,IAAI,MAAM,CAAC,KAAK,KAAK,QAAQ,EAAE,CAAC;wBAC/C,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,CAAC;oBACzC,CAAC;oBACD,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC;gBACpB,CAAC;YACF,CAAC;QACF,CAAC;aAAM,CAAC;YACP,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,WAAW,EAAE,CAAC;YAC1C,IAAI,MAAM,EAAE,IAAI,EAAE,CAAC;gBAClB,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC;YACpB,CAAC;QACF,CAAC;QAED,IAAI,CAAC,IAAI,EAAE,CAAC;YACX,MAAM,KAAK,GAAG,MAAM,OAAO,CAAC,QAAQ,CAAC;gBACpC,OAAO,EAAE,oDAAoD;gBAC7D,WAAW,EAAE,MAAM,CAAC,WAAW;aAC/B,CAAC,CAAC;YACH,MAAM,MAAM,GAAG,uBAAuB,CAAC,KAAK,CAAC,CAAC;YAC9C,IAAI,MAAM,CAAC,KAAK,IAAI,MAAM,CAAC,KAAK,KAAK,QAAQ,EAAE,CAAC;gBAC/C,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,CAAC;YACzC,CAAC;YACD,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC;QACpB,CAAC;QAED,IAAI,CAAC,IAAI,EAAE,CAAC;YACX,MAAM,IAAI,KAAK,CAAC,4BAA4B,CAAC,CAAC;QAC/C,CAAC;QAED,OAAO,CAAC,UAAU,EAAE,CAAC,sCAAsC,CAAC,CAAC;QAC7D,MAAM,YAAY,GAAG,MAAM,QAAQ,CAAC,SAAS,CAAC,aAAa,EAAE;YAC5D,UAAU,EAAE,oBAAoB;YAChC,SAAS,EAAE,SAAS;YACpB,IAAI;YACJ,YAAY,EAAE,MAAM,CAAC,WAAW;YAChC,aAAa,EAAE,QAAQ;SACvB,CAAC,CAAC;QACH,OAAO,2BAA2B,CAAC,IAAI,CAAC,KAAK,CAAC,YAAY,CAAoB,CAAC,CAAC;IACjF,CAAC;YAAS,CAAC;QACV,MAAM,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC;IACvB,CAAC;AACF,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,eAAe,CAAC,YAAoB;IACzD,MAAM,SAAS,GAAG,MAAM,oBAAoB,EAAE,CAAC;IAC/C,MAAM,YAAY,GAAG,MAAM,QAAQ,CAAC,SAAS,CAAC,aAAa,EAAE;QAC5D,UAAU,EAAE,eAAe;QAC3B,SAAS,EAAE,SAAS;QACpB,aAAa,EAAE,YAAY;KAC3B,CAAC,CAAC;IACH,OAAO,2BAA2B,CAAC,IAAI,CAAC,KAAK,CAAC,YAAY,CAAoB,EAAE,YAAY,CAAC,CAAC;AAC/F,CAAC;AAED,MAAM,CAAC,MAAM,QAAQ,GAAc;IAClC,IAAI,EAAE,YAAY;IAElB,KAAK,CAAC,KAAK,CAAC,SAAS;QACpB,wEAAwE;QACxE,4DAA4D;QAC5D,MAAM,WAAW,GAAG,IAAI,eAAe,EAAE,CAAC;QAC1C,IAAI,CAAC;YACJ,MAAM,WAAW,GAAG,MAAM,QAAQ,CAAC;gBAClC,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,SAAS,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,GAAG,EAAE,IAAI,CAAC,GAAG,EAAE,YAAY,EAAE,IAAI,CAAC,YAAY,EAAE,CAAC;gBACxG,UAAU,EAAE,CAAC,OAAO,EAAE,EAAE,CAAC,SAAS,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,OAAO,EAAE,CAAC;gBACxE,QAAQ,EAAE,CAAC,MAAM,EAAE,EAAE,CACpB,SAAS,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,WAAW,EAAE,MAAM,CAAC,WAAW,EAAE,CAAC;gBAC7F,iBAAiB,EAAE,GAAG,EAAE,CACvB,SAAS,CAAC,MAAM,CAAC;oBAChB,IAAI,EAAE,aAAa;oBACnB,OAAO,EAAE,sFAAsF;oBAC/F,WAAW,EAAE,UAAU,aAAa,IAAI,aAAa,GAAG,aAAa,EAAE;oBACvE,MAAM,EAAE,WAAW,CAAC,MAAM;iBAC1B,CAAC;aACH,CAAC,CAAC;YACH,OAAO,EAAE,GAAG,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC;QAC1C,CAAC;gBAAS,CAAC;YACV,WAAW,CAAC,KAAK,EAAE,CAAC;QACrB,CAAC;IACF,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,UAAU;QACvB,OAAO,EAAE,GAAG,CAAC,MAAM,eAAe,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC;IAC1E,CAAC;IAED,KAAK,CAAC,MAAM,CAAC,UAAU;QACtB,OAAO,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM,EAAE,OAAO,EAAE,kBAAkB,EAAE,CAAC;IACnE,CAAC;CACD,CAAC;AAEF,MAAM,CAAC,MAAM,gBAAgB,GAA2B;IACvD,EAAE,EAAE,KAAK;IACT,IAAI,EAAE,YAAY;IAClB,kBAAkB,EAAE,IAAI;IAExB,KAAK,CAAC,KAAK,CAAC,SAA8B;QACzC,OAAO,QAAQ,CAAC;YACf,MAAM,EAAE,SAAS,CAAC,MAAM;YACxB,QAAQ,EAAE,SAAS,CAAC,QAAQ;YAC5B,UAAU,EAAE,SAAS,CAAC,UAAU;YAChC,iBAAiB,EAAE,SAAS,CAAC,iBAAiB;SAC9C,CAAC,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,YAAY,CAAC,WAA6B;QAC/C,OAAO,eAAe,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC;IAC7C,CAAC;IAED,SAAS,CAAC,WAA6B;QACtC,OAAO,WAAW,CAAC,MAAM,CAAC;IAC3B,CAAC;CACD,CAAC","sourcesContent":["/**\n * xAI (Grok) OAuth flow — SuperGrok / X Premium+ subscription login.\n *\n * Mirrors the flow the official Grok CLI uses: PKCE authorization code\n * against auth.x.ai's OIDC discovery document. Tokens are scoped for the\n * Grok CLI proxy (cli-chat-proxy.grok.com) rather than the pay-per-token\n * api.x.ai endpoint, so `toAuth`/`modifyModels` route requests there.\n *\n * NOTE: This module uses Node.js http.createServer for the OAuth callback\n * server. It is only intended for CLI use, not browser environments.\n */\n\nimport type { Server } from \"node:http\";\nimport type { OAuthAuth } from \"../../auth/types.ts\";\nimport { getProviderEnvValue } from \"../provider-env.ts\";\nimport { oauthErrorHtml, oauthSuccessHtml } from \"./oauth-page.ts\";\nimport { generatePKCE } from \"./pkce.ts\";\nimport type { OAuthCredentials, OAuthLoginCallbacks, OAuthPrompt, OAuthProviderInterface } from \"./types.ts\";\n\ntype NodeApis = {\n\tcreateServer: typeof import(\"node:http\").createServer;\n};\n\nlet nodeApis: NodeApis | null = null;\nlet nodeApisPromise: Promise<NodeApis> | null = null;\n\nconst OAUTH_ISSUER = \"https://auth.x.ai\";\nconst DISCOVERY_URL = `${OAUTH_ISSUER}/.well-known/openid-configuration`;\nconst CLIENT_ID = \"b1a00492-073a-47ea-816f-4c329264a828\";\nconst SCOPE = \"openid profile email offline_access grok-cli:access api:access\";\nconst CALLBACK_HOST = getProviderEnvValue(\"PI_OAUTH_CALLBACK_HOST\") || \"127.0.0.1\";\nconst CALLBACK_PORT = 56121;\nconst CALLBACK_PATH = \"/callback\";\nconst REFRESH_SKEW_MS = 2 * 60 * 1000;\n\n/** OAuth-authenticated requests are proxied through the Grok CLI endpoint, not api.x.ai directly. */\nexport const XAI_OAUTH_BASE_URL = \"https://cli-chat-proxy.grok.com/v1\";\n\ntype XaiDiscovery = { authorizationEndpoint: string; tokenEndpoint: string };\ntype XaiTokenPayload = { access_token?: string; refresh_token?: string; expires_in?: number };\n\nasync function getNodeApis(): Promise<NodeApis> {\n\tif (nodeApis) return nodeApis;\n\tif (!nodeApisPromise) {\n\t\tif (typeof process === \"undefined\" || (!process.versions?.node && !process.versions?.bun)) {\n\t\t\tthrow new Error(\"xAI OAuth is only available in Node.js environments\");\n\t\t}\n\t\tnodeApisPromise = import(\"node:http\").then((httpModule) => ({\n\t\t\tcreateServer: httpModule.createServer,\n\t\t}));\n\t}\n\tnodeApis = await nodeApisPromise;\n\treturn nodeApis;\n}\n\n/** Reject discovery responses that don't point back at x.ai, in case DNS/discovery is ever compromised. */\nfunction validateXaiEndpoint(url: string): string {\n\tconst parsed = new URL(url);\n\tconst host = parsed.hostname.toLowerCase();\n\tif (parsed.protocol !== \"https:\" || (host !== \"x.ai\" && !host.endsWith(\".x.ai\"))) {\n\t\tthrow new Error(`xAI OAuth discovery returned an unexpected endpoint: ${url}`);\n\t}\n\treturn url;\n}\n\nasync function discoverXaiEndpoints(): Promise<XaiDiscovery> {\n\tconst response = await fetch(DISCOVERY_URL, { headers: { Accept: \"application/json\" } });\n\tif (!response.ok) {\n\t\tthrow new Error(`xAI OAuth discovery failed: ${response.status} ${await response.text()}`);\n\t}\n\n\tconst data = (await response.json()) as { authorization_endpoint?: string; token_endpoint?: string };\n\tif (!data.authorization_endpoint || !data.token_endpoint) {\n\t\tthrow new Error(\"xAI OAuth discovery response did not include authorization/token endpoints\");\n\t}\n\n\treturn {\n\t\tauthorizationEndpoint: validateXaiEndpoint(data.authorization_endpoint),\n\t\ttokenEndpoint: validateXaiEndpoint(data.token_endpoint),\n\t};\n}\n\nfunction parseAuthorizationInput(input: string): { code?: string; state?: string } {\n\tconst value = input.trim();\n\tif (!value) return {};\n\n\ttry {\n\t\tconst url = new URL(value);\n\t\treturn {\n\t\t\tcode: url.searchParams.get(\"code\") ?? undefined,\n\t\t\tstate: url.searchParams.get(\"state\") ?? undefined,\n\t\t};\n\t} catch {\n\t\t// not a URL\n\t}\n\n\tif (value.includes(\"code=\")) {\n\t\tconst params = new URLSearchParams(value);\n\t\treturn {\n\t\t\tcode: params.get(\"code\") ?? undefined,\n\t\t\tstate: params.get(\"state\") ?? undefined,\n\t\t};\n\t}\n\n\treturn { code: value };\n}\n\ntype CallbackServerInfo = {\n\tserver: Server;\n\tredirectUri: string;\n\tcancelWait: () => void;\n\twaitForCode: () => Promise<{ code: string; state: string } | null>;\n};\n\nasync function startCallbackServer(expectedState: string): Promise<CallbackServerInfo> {\n\tconst { createServer } = await getNodeApis();\n\n\treturn new Promise((resolve, reject) => {\n\t\tlet settleWait: ((value: { code: string; state: string } | null) => void) | undefined;\n\t\tconst waitForCodePromise = new Promise<{ code: string; state: string } | null>((resolveWait) => {\n\t\t\tlet settled = false;\n\t\t\tsettleWait = (value) => {\n\t\t\t\tif (settled) return;\n\t\t\t\tsettled = true;\n\t\t\t\tresolveWait(value);\n\t\t\t};\n\t\t});\n\n\t\tconst server = createServer((req, res) => {\n\t\t\ttry {\n\t\t\t\tconst url = new URL(req.url || \"\", `http://${CALLBACK_HOST}`);\n\t\t\t\tif (url.pathname !== CALLBACK_PATH) {\n\t\t\t\t\tres.writeHead(404, { \"Content-Type\": \"text/html; charset=utf-8\" });\n\t\t\t\t\tres.end(oauthErrorHtml(\"Callback route not found.\"));\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\tconst code = url.searchParams.get(\"code\");\n\t\t\t\tconst state = url.searchParams.get(\"state\");\n\t\t\t\tconst error = url.searchParams.get(\"error\");\n\n\t\t\t\tif (error) {\n\t\t\t\t\tres.writeHead(400, { \"Content-Type\": \"text/html; charset=utf-8\" });\n\t\t\t\t\tres.end(oauthErrorHtml(\"xAI authentication did not complete.\", `Error: ${error}`));\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\tif (!code || !state) {\n\t\t\t\t\tres.writeHead(400, { \"Content-Type\": \"text/html; charset=utf-8\" });\n\t\t\t\t\tres.end(oauthErrorHtml(\"Missing code or state parameter.\"));\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\tif (state !== expectedState) {\n\t\t\t\t\tres.writeHead(400, { \"Content-Type\": \"text/html; charset=utf-8\" });\n\t\t\t\t\tres.end(oauthErrorHtml(\"State mismatch.\"));\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\tres.writeHead(200, { \"Content-Type\": \"text/html; charset=utf-8\" });\n\t\t\t\tres.end(oauthSuccessHtml(\"xAI authentication completed. You can close this window.\"));\n\t\t\t\tsettleWait?.({ code, state });\n\t\t\t} catch {\n\t\t\t\tres.writeHead(500, { \"Content-Type\": \"text/plain; charset=utf-8\" });\n\t\t\t\tres.end(\"Internal error\");\n\t\t\t}\n\t\t});\n\n\t\tserver.on(\"error\", (err) => {\n\t\t\treject(err);\n\t\t});\n\n\t\tserver.listen(CALLBACK_PORT, CALLBACK_HOST, () => {\n\t\t\tresolve({\n\t\t\t\tserver,\n\t\t\t\tredirectUri: `http://${CALLBACK_HOST}:${CALLBACK_PORT}${CALLBACK_PATH}`,\n\t\t\t\tcancelWait: () => {\n\t\t\t\t\tsettleWait?.(null);\n\t\t\t\t},\n\t\t\t\twaitForCode: () => waitForCodePromise,\n\t\t\t});\n\t\t});\n\t});\n}\n\nasync function postForm(url: string, body: Record<string, string>): Promise<string> {\n\tconst response = await fetch(url, {\n\t\tmethod: \"POST\",\n\t\theaders: {\n\t\t\t\"Content-Type\": \"application/x-www-form-urlencoded\",\n\t\t\tAccept: \"application/json\",\n\t\t},\n\t\tbody: new URLSearchParams(body).toString(),\n\t\tsignal: AbortSignal.timeout(30_000),\n\t});\n\n\tconst responseBody = await response.text();\n\tif (!response.ok) {\n\t\tthrow new Error(`xAI token request failed. status=${response.status}; url=${url}; body=${responseBody}`);\n\t}\n\treturn responseBody;\n}\n\nfunction credentialsFromTokenPayload(data: XaiTokenPayload, fallbackRefresh = \"\"): OAuthCredentials {\n\tif (!data.access_token) {\n\t\tthrow new Error(\"xAI token response did not include an access token\");\n\t}\n\n\tconst refresh = data.refresh_token || fallbackRefresh;\n\tif (!refresh) {\n\t\tthrow new Error(\"xAI token response did not include a refresh token\");\n\t}\n\n\treturn {\n\t\taccess: data.access_token,\n\t\trefresh,\n\t\texpires: Date.now() + (data.expires_in ?? 3600) * 1000 - REFRESH_SKEW_MS,\n\t};\n}\n\n/**\n * Login with xAI OAuth (SuperGrok / X Premium+ subscription, authorization code + PKCE)\n */\nexport async function loginXai(options: {\n\tonAuth: (info: { url: string; instructions?: string }) => void;\n\tonPrompt: (prompt: OAuthPrompt) => Promise<string>;\n\tonProgress?: (message: string) => void;\n\tonManualCodeInput?: () => Promise<string>;\n}): Promise<OAuthCredentials> {\n\tconst discovery = await discoverXaiEndpoints();\n\tconst { verifier, challenge } = await generatePKCE();\n\tconst server = await startCallbackServer(verifier);\n\n\ttry {\n\t\tconst authParams = new URLSearchParams({\n\t\t\tresponse_type: \"code\",\n\t\t\tclient_id: CLIENT_ID,\n\t\t\tredirect_uri: server.redirectUri,\n\t\t\tscope: SCOPE,\n\t\t\tcode_challenge: challenge,\n\t\t\tcode_challenge_method: \"S256\",\n\t\t\tstate: verifier,\n\t\t});\n\n\t\toptions.onAuth({\n\t\t\turl: `${discovery.authorizationEndpoint}?${authParams.toString()}`,\n\t\t\tinstructions:\n\t\t\t\t\"Sign in with your SuperGrok or X Premium+ account. If the browser is on another machine, paste the final redirect URL here.\",\n\t\t});\n\n\t\tlet code: string | undefined;\n\n\t\tif (options.onManualCodeInput) {\n\t\t\tlet manualInput: string | undefined;\n\t\t\tlet manualError: Error | undefined;\n\t\t\tconst manualPromise = options\n\t\t\t\t.onManualCodeInput()\n\t\t\t\t.then((input) => {\n\t\t\t\t\tmanualInput = input;\n\t\t\t\t\tserver.cancelWait();\n\t\t\t\t})\n\t\t\t\t.catch((err) => {\n\t\t\t\t\tmanualError = err instanceof Error ? err : new Error(String(err));\n\t\t\t\t\tserver.cancelWait();\n\t\t\t\t});\n\n\t\t\tconst result = await server.waitForCode();\n\n\t\t\tif (manualError) {\n\t\t\t\tthrow manualError;\n\t\t\t}\n\n\t\t\tif (result?.code) {\n\t\t\t\tcode = result.code;\n\t\t\t} else if (manualInput) {\n\t\t\t\tconst parsed = parseAuthorizationInput(manualInput);\n\t\t\t\tif (parsed.state && parsed.state !== verifier) {\n\t\t\t\t\tthrow new Error(\"OAuth state mismatch\");\n\t\t\t\t}\n\t\t\t\tcode = parsed.code;\n\t\t\t}\n\n\t\t\tif (!code) {\n\t\t\t\tawait manualPromise;\n\t\t\t\tif (manualError) {\n\t\t\t\t\tthrow manualError;\n\t\t\t\t}\n\t\t\t\tif (manualInput) {\n\t\t\t\t\tconst parsed = parseAuthorizationInput(manualInput);\n\t\t\t\t\tif (parsed.state && parsed.state !== verifier) {\n\t\t\t\t\t\tthrow new Error(\"OAuth state mismatch\");\n\t\t\t\t\t}\n\t\t\t\t\tcode = parsed.code;\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tconst result = await server.waitForCode();\n\t\t\tif (result?.code) {\n\t\t\t\tcode = result.code;\n\t\t\t}\n\t\t}\n\n\t\tif (!code) {\n\t\t\tconst input = await options.onPrompt({\n\t\t\t\tmessage: \"Paste the authorization code or full redirect URL:\",\n\t\t\t\tplaceholder: server.redirectUri,\n\t\t\t});\n\t\t\tconst parsed = parseAuthorizationInput(input);\n\t\t\tif (parsed.state && parsed.state !== verifier) {\n\t\t\t\tthrow new Error(\"OAuth state mismatch\");\n\t\t\t}\n\t\t\tcode = parsed.code;\n\t\t}\n\n\t\tif (!code) {\n\t\t\tthrow new Error(\"Missing authorization code\");\n\t\t}\n\n\t\toptions.onProgress?.(\"Exchanging xAI authorization code...\");\n\t\tconst responseBody = await postForm(discovery.tokenEndpoint, {\n\t\t\tgrant_type: \"authorization_code\",\n\t\t\tclient_id: CLIENT_ID,\n\t\t\tcode,\n\t\t\tredirect_uri: server.redirectUri,\n\t\t\tcode_verifier: verifier,\n\t\t});\n\t\treturn credentialsFromTokenPayload(JSON.parse(responseBody) as XaiTokenPayload);\n\t} finally {\n\t\tserver.server.close();\n\t}\n}\n\n/**\n * Refresh xAI OAuth token\n */\nexport async function refreshXaiToken(refreshToken: string): Promise<OAuthCredentials> {\n\tconst discovery = await discoverXaiEndpoints();\n\tconst responseBody = await postForm(discovery.tokenEndpoint, {\n\t\tgrant_type: \"refresh_token\",\n\t\tclient_id: CLIENT_ID,\n\t\trefresh_token: refreshToken,\n\t});\n\treturn credentialsFromTokenPayload(JSON.parse(responseBody) as XaiTokenPayload, refreshToken);\n}\n\nexport const xaiOAuth: OAuthAuth = {\n\tname: \"xAI (Grok)\",\n\n\tasync login(callbacks) {\n\t\t// The manual_code prompt races the local callback server; abort it once\n\t\t// the flow settles so the UI can dismiss the pending input.\n\t\tconst manualAbort = new AbortController();\n\t\ttry {\n\t\t\tconst credentials = await loginXai({\n\t\t\t\tonAuth: (info) => callbacks.notify({ type: \"auth_url\", url: info.url, instructions: info.instructions }),\n\t\t\t\tonProgress: (message) => callbacks.notify({ type: \"progress\", message }),\n\t\t\t\tonPrompt: (prompt) =>\n\t\t\t\t\tcallbacks.prompt({ type: \"text\", message: prompt.message, placeholder: prompt.placeholder }),\n\t\t\t\tonManualCodeInput: () =>\n\t\t\t\t\tcallbacks.prompt({\n\t\t\t\t\t\ttype: \"manual_code\",\n\t\t\t\t\t\tmessage: \"Complete login in your browser, or paste the authorization code / redirect URL here:\",\n\t\t\t\t\t\tplaceholder: `http://${CALLBACK_HOST}:${CALLBACK_PORT}${CALLBACK_PATH}`,\n\t\t\t\t\t\tsignal: manualAbort.signal,\n\t\t\t\t\t}),\n\t\t\t});\n\t\t\treturn { ...credentials, type: \"oauth\" };\n\t\t} finally {\n\t\t\tmanualAbort.abort();\n\t\t}\n\t},\n\n\tasync refresh(credential) {\n\t\treturn { ...(await refreshXaiToken(credential.refresh)), type: \"oauth\" };\n\t},\n\n\tasync toAuth(credential) {\n\t\treturn { apiKey: credential.access, baseUrl: XAI_OAUTH_BASE_URL };\n\t},\n};\n\nexport const xaiOAuthProvider: OAuthProviderInterface = {\n\tid: \"xai\",\n\tname: \"xAI (Grok)\",\n\tusesCallbackServer: true,\n\n\tasync login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials> {\n\t\treturn loginXai({\n\t\t\tonAuth: callbacks.onAuth,\n\t\t\tonPrompt: callbacks.onPrompt,\n\t\t\tonProgress: callbacks.onProgress,\n\t\t\tonManualCodeInput: callbacks.onManualCodeInput,\n\t\t});\n\t},\n\n\tasync refreshToken(credentials: OAuthCredentials): Promise<OAuthCredentials> {\n\t\treturn refreshXaiToken(credentials.refresh);\n\t},\n\n\tgetApiKey(credentials: OAuthCredentials): string {\n\t\treturn credentials.access;\n\t},\n};\n"]}

package/dist/utils/overflow.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"overflow.d.ts","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AA6EpD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAkDG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,gBAAgB,EAAE,aAAa,CAAC,EAAE,MAAM,GAAG,OAAO,CA6B5F;AAED;;GAEG;AACH,wBAAgB,mBAAmB,IAAI,MAAM,EAAE,CAE9C","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/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 * - 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 * - Ollama: Some deployments truncate silently, others return errors like \"prompt too long; exceeded max context length by X tokens\"\n */\nconst OVERFLOW_PATTERNS = [\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 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/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];\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.\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];\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles two 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 *\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 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 *\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}\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/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n"]}
+{"version":3,"file":"overflow.d.ts","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AA6EpD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAkDG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,gBAAgB,EAAE,aAAa,CAAC,EAAE,MAAM,GAAG,OAAO,CA6B5F;AAED;;GAEG;AACH,wBAAgB,mBAAmB,IAAI,MAAM,EAAE,CAE9C"}

package/dist/utils/overflow.js.map

@@ -1 +1 @@
-{"version":3,"file":"overflow.js","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAgCG;AACH,MAAM,iBAAiB,GAAG;IACzB,qBAAqB,EAAE,2BAA2B;IAClD,oBAAoB,EAAE,kDAAkD;IACxE,wCAAwC,EAAE,iBAAiB;IAC3D,6BAA6B,EAAE,uCAAuC;IACtE,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,gCAAgC,EAAE,mBAAmB;IACrD,kBAAkB,EAAE,mBAAmB;IACvC,uBAAuB,EAAE,mBAAmB;IAC5C,+CAA+C,EAAE,iCAAiC;CAClF,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,qBAAqB,GAAG;IAC7B,2CAA2C,EAAE,oFAAoF;IACjI,aAAa,EAAE,wBAAwB;IACvC,oBAAoB,EAAE,yBAAyB;CAC/C,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAkDG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAyB,EAAE,aAAsB,EAAW;IAC7F,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;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;AAAA,CACb;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB,GAAa;IAC/C,OAAO,CAAC,GAAG,iBAAiB,CAAC,CAAC;AAAA,CAC9B","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/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 * - 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 * - Ollama: Some deployments truncate silently, others return errors like \"prompt too long; exceeded max context length by X tokens\"\n */\nconst OVERFLOW_PATTERNS = [\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 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/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];\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.\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];\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles two 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 *\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 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 *\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}\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/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n"]}
+{"version":3,"file":"overflow.js","sourceRoot":"","sources":["../../src/utils/overflow.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAgCG;AACH,MAAM,iBAAiB,GAAG;IACzB,qBAAqB,EAAE,2BAA2B;IAClD,oBAAoB,EAAE,kDAAkD;IACxE,wCAAwC,EAAE,iBAAiB;IAC3D,6BAA6B,EAAE,uCAAuC;IACtE,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,gCAAgC,EAAE,mBAAmB;IACrD,kBAAkB,EAAE,mBAAmB;IACvC,uBAAuB,EAAE,mBAAmB;IAC5C,+CAA+C,EAAE,iCAAiC;CAClF,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,qBAAqB,GAAG;IAC7B,2CAA2C,EAAE,oFAAoF;IACjI,aAAa,EAAE,wBAAwB;IACvC,oBAAoB,EAAE,yBAAyB;CAC/C,CAAC;AAEF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAkDG;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;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;AAED;;GAEG;AACH,MAAM,UAAU,mBAAmB;IAClC,OAAO,CAAC,GAAG,iBAAiB,CAAC,CAAC;AAC/B,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/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 * - 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 * - Ollama: Some deployments truncate silently, others return errors like \"prompt too long; exceeded max context length by X tokens\"\n */\nconst OVERFLOW_PATTERNS = [\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 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/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];\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.\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];\n\n/**\n * Check if an assistant message represents a context overflow error.\n *\n * This handles two 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 *\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 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 *\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}\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/**\n * Get the overflow patterns for testing purposes.\n */\nexport function getOverflowPatterns(): RegExp[] {\n\treturn [...OVERFLOW_PATTERNS];\n}\n"]}

package/dist/utils/provider-env.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"provider-env.d.ts","sourceRoot":"","sources":["../../src/utils/provider-env.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAwC/C;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,GAAG,SAAS,CAOvF","sourcesContent":["import type { ProviderEnv } from \"../types.ts\";\n\nlet procEnvCache: Map<string, string> | null = null;\n\n/**\n * Fallback for https://github.com/oven-sh/bun/issues/27802.\n * Bun compiled binaries can expose an empty process.env inside Linux sandboxes\n * even though /proc/self/environ contains the environment.\n *\n * This intentionally duplicates restoreSandboxEnv() in\n * packages/coding-agent/src/bun/restore-sandbox-env.ts. The ai package can be\n * used directly, without going through that entrypoint, so provider env lookup\n * must not depend on process.env having been patched.\n */\nfunction getBunSandboxEnvValue(name: string): string | undefined {\n\tif (typeof process === \"undefined\" || !process.versions?.bun || Object.keys(process.env).length > 0) {\n\t\treturn undefined;\n\t}\n\n\tif (procEnvCache === null) {\n\t\tprocEnvCache = new Map();\n\t\ttry {\n\t\t\tconst { readFileSync } = require(\"node:fs\") as {\n\t\t\t\treadFileSync(path: string, encoding: BufferEncoding): string;\n\t\t\t};\n\t\t\tconst data = readFileSync(\"/proc/self/environ\", \"utf-8\");\n\t\t\tfor (const entry of data.split(\"\\0\")) {\n\t\t\t\tconst idx = entry.indexOf(\"=\");\n\t\t\t\tif (idx > 0) {\n\t\t\t\t\tprocEnvCache.set(entry.slice(0, idx), entry.slice(idx + 1));\n\t\t\t\t}\n\t\t\t}\n\t\t} catch {\n\t\t\t// /proc/self/environ may not exist or may not be readable.\n\t\t}\n\t}\n\n\treturn procEnvCache.get(name);\n}\n\n/**\n * Resolve a provider env value from scoped overrides, normal process.env, then\n * the duplicated Bun sandbox fallback for direct pi-ai consumers.\n */\nexport function getProviderEnvValue(name: string, env?: ProviderEnv): string | undefined {\n\treturn (\n\t\tenv?.[name] ||\n\t\t(typeof process !== \"undefined\" ? process.env[name] : undefined) ||\n\t\tgetBunSandboxEnvValue(name) ||\n\t\tundefined\n\t);\n}\n"]}
+{"version":3,"file":"provider-env.d.ts","sourceRoot":"","sources":["../../src/utils/provider-env.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAwC/C;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,CAAC,EAAE,WAAW,GAAG,MAAM,GAAG,SAAS,CAOvF"}

package/dist/utils/provider-env.js.map

@@ -1 +1 @@
-{"version":3,"file":"provider-env.js","sourceRoot":"","sources":["../../src/utils/provider-env.ts"],"names":[],"mappings":"AAEA,IAAI,YAAY,GAA+B,IAAI,CAAC;AAEpD;;;;;;;;;GASG;AACH,SAAS,qBAAqB,CAAC,IAAY,EAAsB;IAChE,IAAI,OAAO,OAAO,KAAK,WAAW,IAAI,CAAC,OAAO,CAAC,QAAQ,EAAE,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACrG,OAAO,SAAS,CAAC;IAClB,CAAC;IAED,IAAI,YAAY,KAAK,IAAI,EAAE,CAAC;QAC3B,YAAY,GAAG,IAAI,GAAG,EAAE,CAAC;QACzB,IAAI,CAAC;YACJ,MAAM,EAAE,YAAY,EAAE,GAAG,OAAO,CAAC,SAAS,CAEzC,CAAC;YACF,MAAM,IAAI,GAAG,YAAY,CAAC,oBAAoB,EAAE,OAAO,CAAC,CAAC;YACzD,KAAK,MAAM,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;gBACtC,MAAM,GAAG,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;gBAC/B,IAAI,GAAG,GAAG,CAAC,EAAE,CAAC;oBACb,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,KAAK,CAAC,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;gBAC7D,CAAC;YACF,CAAC;QACF,CAAC;QAAC,MAAM,CAAC;YACR,2DAA2D;QAC5D,CAAC;IACF,CAAC;IAED,OAAO,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;AAAA,CAC9B;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAC,IAAY,EAAE,GAAiB,EAAsB;IACxF,OAAO,CACN,GAAG,EAAE,CAAC,IAAI,CAAC;QACX,CAAC,OAAO,OAAO,KAAK,WAAW,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QAChE,qBAAqB,CAAC,IAAI,CAAC;QAC3B,SAAS,CACT,CAAC;AAAA,CACF","sourcesContent":["import type { ProviderEnv } from \"../types.ts\";\n\nlet procEnvCache: Map<string, string> | null = null;\n\n/**\n * Fallback for https://github.com/oven-sh/bun/issues/27802.\n * Bun compiled binaries can expose an empty process.env inside Linux sandboxes\n * even though /proc/self/environ contains the environment.\n *\n * This intentionally duplicates restoreSandboxEnv() in\n * packages/coding-agent/src/bun/restore-sandbox-env.ts. The ai package can be\n * used directly, without going through that entrypoint, so provider env lookup\n * must not depend on process.env having been patched.\n */\nfunction getBunSandboxEnvValue(name: string): string | undefined {\n\tif (typeof process === \"undefined\" || !process.versions?.bun || Object.keys(process.env).length > 0) {\n\t\treturn undefined;\n\t}\n\n\tif (procEnvCache === null) {\n\t\tprocEnvCache = new Map();\n\t\ttry {\n\t\t\tconst { readFileSync } = require(\"node:fs\") as {\n\t\t\t\treadFileSync(path: string, encoding: BufferEncoding): string;\n\t\t\t};\n\t\t\tconst data = readFileSync(\"/proc/self/environ\", \"utf-8\");\n\t\t\tfor (const entry of data.split(\"\\0\")) {\n\t\t\t\tconst idx = entry.indexOf(\"=\");\n\t\t\t\tif (idx > 0) {\n\t\t\t\t\tprocEnvCache.set(entry.slice(0, idx), entry.slice(idx + 1));\n\t\t\t\t}\n\t\t\t}\n\t\t} catch {\n\t\t\t// /proc/self/environ may not exist or may not be readable.\n\t\t}\n\t}\n\n\treturn procEnvCache.get(name);\n}\n\n/**\n * Resolve a provider env value from scoped overrides, normal process.env, then\n * the duplicated Bun sandbox fallback for direct pi-ai consumers.\n */\nexport function getProviderEnvValue(name: string, env?: ProviderEnv): string | undefined {\n\treturn (\n\t\tenv?.[name] ||\n\t\t(typeof process !== \"undefined\" ? process.env[name] : undefined) ||\n\t\tgetBunSandboxEnvValue(name) ||\n\t\tundefined\n\t);\n}\n"]}
+{"version":3,"file":"provider-env.js","sourceRoot":"","sources":["../../src/utils/provider-env.ts"],"names":[],"mappings":"AAEA,IAAI,YAAY,GAA+B,IAAI,CAAC;AAEpD;;;;;;;;;GASG;AACH,SAAS,qBAAqB,CAAC,IAAY;IAC1C,IAAI,OAAO,OAAO,KAAK,WAAW,IAAI,CAAC,OAAO,CAAC,QAAQ,EAAE,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACrG,OAAO,SAAS,CAAC;IAClB,CAAC;IAED,IAAI,YAAY,KAAK,IAAI,EAAE,CAAC;QAC3B,YAAY,GAAG,IAAI,GAAG,EAAE,CAAC;QACzB,IAAI,CAAC;YACJ,MAAM,EAAE,YAAY,EAAE,GAAG,OAAO,CAAC,SAAS,CAEzC,CAAC;YACF,MAAM,IAAI,GAAG,YAAY,CAAC,oBAAoB,EAAE,OAAO,CAAC,CAAC;YACzD,KAAK,MAAM,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;gBACtC,MAAM,GAAG,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;gBAC/B,IAAI,GAAG,GAAG,CAAC,EAAE,CAAC;oBACb,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,KAAK,CAAC,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;gBAC7D,CAAC;YACF,CAAC;QACF,CAAC;QAAC,MAAM,CAAC;YACR,2DAA2D;QAC5D,CAAC;IACF,CAAC;IAED,OAAO,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;AAC/B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAC,IAAY,EAAE,GAAiB;IAClE,OAAO,CACN,GAAG,EAAE,CAAC,IAAI,CAAC;QACX,CAAC,OAAO,OAAO,KAAK,WAAW,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QAChE,qBAAqB,CAAC,IAAI,CAAC;QAC3B,SAAS,CACT,CAAC;AACH,CAAC","sourcesContent":["import type { ProviderEnv } from \"../types.ts\";\n\nlet procEnvCache: Map<string, string> | null = null;\n\n/**\n * Fallback for https://github.com/oven-sh/bun/issues/27802.\n * Bun compiled binaries can expose an empty process.env inside Linux sandboxes\n * even though /proc/self/environ contains the environment.\n *\n * This intentionally duplicates restoreSandboxEnv() in\n * packages/coding-agent/src/bun/restore-sandbox-env.ts. The ai package can be\n * used directly, without going through that entrypoint, so provider env lookup\n * must not depend on process.env having been patched.\n */\nfunction getBunSandboxEnvValue(name: string): string | undefined {\n\tif (typeof process === \"undefined\" || !process.versions?.bun || Object.keys(process.env).length > 0) {\n\t\treturn undefined;\n\t}\n\n\tif (procEnvCache === null) {\n\t\tprocEnvCache = new Map();\n\t\ttry {\n\t\t\tconst { readFileSync } = require(\"node:fs\") as {\n\t\t\t\treadFileSync(path: string, encoding: BufferEncoding): string;\n\t\t\t};\n\t\t\tconst data = readFileSync(\"/proc/self/environ\", \"utf-8\");\n\t\t\tfor (const entry of data.split(\"\\0\")) {\n\t\t\t\tconst idx = entry.indexOf(\"=\");\n\t\t\t\tif (idx > 0) {\n\t\t\t\t\tprocEnvCache.set(entry.slice(0, idx), entry.slice(idx + 1));\n\t\t\t\t}\n\t\t\t}\n\t\t} catch {\n\t\t\t// /proc/self/environ may not exist or may not be readable.\n\t\t}\n\t}\n\n\treturn procEnvCache.get(name);\n}\n\n/**\n * Resolve a provider env value from scoped overrides, normal process.env, then\n * the duplicated Bun sandbox fallback for direct pi-ai consumers.\n */\nexport function getProviderEnvValue(name: string, env?: ProviderEnv): string | undefined {\n\treturn (\n\t\tenv?.[name] ||\n\t\t(typeof process !== \"undefined\" ? process.env[name] : undefined) ||\n\t\tgetBunSandboxEnvValue(name) ||\n\t\tundefined\n\t);\n}\n"]}

package/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;AAkFpD;;;;;;;;GAQG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAK5E","sourcesContent":["import type { AssistantMessage } from \"../types.ts\";\n\nfunction buildProviderErrorPattern(patterns: readonly string[]): RegExp {\n\treturn new RegExp(patterns.join(\"|\"), \"i\");\n}\n\nconst NON_RETRYABLE_PROVIDER_LIMIT_ERROR_PATTERN = buildProviderErrorPattern([\n\t// OpenCode Go/free-tier limits returned as 429 JSON error types by OpenCode's\n\t// Zen API. These are subscription/account limits, not transient throttles.\n\t\"GoUsageLimitError\",\n\t\"FreeUsageLimitError\",\n\n\t// OpenCode Go subscription-limit text asks users to enable available-balance\n\t// usage after rolling/weekly/monthly limits are reached.\n\t\"Monthly usage limit reached\",\n\t\"available balance\",\n\n\t// Generic quota/budget/billing exhaustion. `insufficient_quota` is OpenAI's\n\t// quota/billing error code; the other strings cover common gateway wording.\n\t\"insufficient_quota\",\n\t\"out of budget\",\n\t\"quota exceeded\",\n\t\"billing\",\n]);\n\nconst RETRYABLE_PROVIDER_ERROR_PATTERN = buildProviderErrorPattern([\n\t// Generic provider load, HTTP status, and server-side transient failures.\n\t\"overloaded\",\n\t\"rate.?limit\",\n\t\"too many requests\",\n\t\"429\",\n\t\"500\",\n\t\"502\",\n\t\"503\",\n\t\"504\",\n\t\"524\",\n\t\"service.?unavailable\",\n\t\"server.?error\",\n\t\"internal.?error\",\n\n\t// Wrapper/provider text for transient upstream failures, including OpenRouter\n\t// \"Provider returned error\" responses (#2264).\n\t\"provider.?returned.?error\",\n\n\t// Network, proxy, and fetch transport failures. This includes OpenAI Codex\n\t// raw-fetch failures such as \"upstream connect\", \"connection refused\", and\n\t// \"reset before headers\" (#733), plus OpenRouter connection drops (#3317).\n\t\"network.?error\",\n\t\"connection.?error\",\n\t\"connection.?refused\",\n\t\"connection.?lost\",\n\t\"other side closed\",\n\t\"fetch failed\",\n\t\"upstream.?connect\",\n\t\"reset before headers\",\n\t\"socket hang up\",\n\t\"timed? out\",\n\t\"timeout\",\n\t\"terminated\",\n\n\t// WebSocket transports can report close/error text instead of HTTP/fetch text.\n\t\"websocket.?closed\",\n\t\"websocket.?error\",\n\n\t// Premature stream endings from SDKs and transports. Anthropic can throw\n\t// \"stream ended without ...\" and \"Anthropic stream ended before message_stop\"\n\t// (#4433); Bedrock/Smithy can throw an HTTP/2 no-response error (#3594).\n\t\"ended without\",\n\t\"stream ended before message_stop\",\n\t\"http2 request did not get a response\",\n\n\t// Provider-requested retry delay cap failures should flow through the outer\n\t// retry policy so callers can surface/abort the backoff (#1123).\n\t\"retry delay\",\n\n\t// Explicit retry guidance emitted mid-stream by OpenAI Responses and Bedrock\n\t// stream exceptions (#6019).\n\t\"you can retry your request\",\n\t\"try your request again\",\n\t\"please retry your request\",\n]);\n\n/**\n * Classifies whether a failed assistant message looks like a transient provider\n * or transport error, so callers can decide if the last assistant turn should be\n * restarted.\n *\n * This does not implement retry policy. Callers should first handle context\n * overflow separately, then apply their own retry budget, backoff, and reporting\n * before restarting the assistant turn.\n */\nexport function isRetryableAssistantError(message: AssistantMessage): boolean {\n\tif (message.stopReason !== \"error\" || !message.errorMessage) return false;\n\tconst errorMessage = message.errorMessage;\n\tif (NON_RETRYABLE_PROVIDER_LIMIT_ERROR_PATTERN.test(errorMessage)) return false;\n\treturn RETRYABLE_PROVIDER_ERROR_PATTERN.test(errorMessage);\n}\n"]}
+{"version":3,"file":"retry.d.ts","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAsFpD;;;;;;;;GAQG;AACH,wBAAgB,yBAAyB,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAK5E"}

package/dist/utils/retry.js

@@ -46,6 +46,7 @@ const RETRYABLE_PROVIDER_ERROR_PATTERN = buildProviderErrorPattern([
     "upstream.?connect",
     "reset before headers",
     "socket hang up",
+    "socket connection was closed",
     "timed? out",
     "timeout",
     "terminated",
@@ -66,6 +67,8 @@ const RETRYABLE_PROVIDER_ERROR_PATTERN = buildProviderErrorPattern([
     "you can retry your request",
     "try your request again",
     "please retry your request",
+    // gRPC based providers (e.g. NVIDIA NIM)
+    "ResourceExhausted",
 ]);
 /**
  * Classifies whether a failed assistant message looks like a transient provider

package/dist/utils/retry.js.map

@@ -1 +1 @@
-{"version":3,"file":"retry.js","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAEA,SAAS,yBAAyB,CAAC,QAA2B,EAAU;IACvE,OAAO,IAAI,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,CAAC;AAAA,CAC3C;AAED,MAAM,0CAA0C,GAAG,yBAAyB,CAAC;IAC5E,8EAA8E;IAC9E,2EAA2E;IAC3E,mBAAmB;IACnB,qBAAqB;IAErB,6EAA6E;IAC7E,yDAAyD;IACzD,6BAA6B;IAC7B,mBAAmB;IAEnB,4EAA4E;IAC5E,4EAA4E;IAC5E,oBAAoB;IACpB,eAAe;IACf,gBAAgB;IAChB,SAAS;CACT,CAAC,CAAC;AAEH,MAAM,gCAAgC,GAAG,yBAAyB,CAAC;IAClE,0EAA0E;IAC1E,YAAY;IACZ,aAAa;IACb,mBAAmB;IACnB,KAAK;IACL,KAAK;IACL,KAAK;IACL,KAAK;IACL,KAAK;IACL,KAAK;IACL,sBAAsB;IACtB,eAAe;IACf,iBAAiB;IAEjB,8EAA8E;IAC9E,+CAA+C;IAC/C,2BAA2B;IAE3B,2EAA2E;IAC3E,2EAA2E;IAC3E,2EAA2E;IAC3E,gBAAgB;IAChB,mBAAmB;IACnB,qBAAqB;IACrB,kBAAkB;IAClB,mBAAmB;IACnB,cAAc;IACd,mBAAmB;IACnB,sBAAsB;IACtB,gBAAgB;IAChB,YAAY;IACZ,SAAS;IACT,YAAY;IAEZ,+EAA+E;IAC/E,mBAAmB;IACnB,kBAAkB;IAElB,yEAAyE;IACzE,8EAA8E;IAC9E,yEAAyE;IACzE,eAAe;IACf,kCAAkC;IAClC,sCAAsC;IAEtC,4EAA4E;IAC5E,iEAAiE;IACjE,aAAa;IAEb,6EAA6E;IAC7E,6BAA6B;IAC7B,4BAA4B;IAC5B,wBAAwB;IACxB,2BAA2B;CAC3B,CAAC,CAAC;AAEH;;;;;;;;GAQG;AACH,MAAM,UAAU,yBAAyB,CAAC,OAAyB,EAAW;IAC7E,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,OAAO,CAAC,YAAY;QAAE,OAAO,KAAK,CAAC;IAC1E,MAAM,YAAY,GAAG,OAAO,CAAC,YAAY,CAAC;IAC1C,IAAI,0CAA0C,CAAC,IAAI,CAAC,YAAY,CAAC;QAAE,OAAO,KAAK,CAAC;IAChF,OAAO,gCAAgC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;AAAA,CAC3D","sourcesContent":["import type { AssistantMessage } from \"../types.ts\";\n\nfunction buildProviderErrorPattern(patterns: readonly string[]): RegExp {\n\treturn new RegExp(patterns.join(\"|\"), \"i\");\n}\n\nconst NON_RETRYABLE_PROVIDER_LIMIT_ERROR_PATTERN = buildProviderErrorPattern([\n\t// OpenCode Go/free-tier limits returned as 429 JSON error types by OpenCode's\n\t// Zen API. These are subscription/account limits, not transient throttles.\n\t\"GoUsageLimitError\",\n\t\"FreeUsageLimitError\",\n\n\t// OpenCode Go subscription-limit text asks users to enable available-balance\n\t// usage after rolling/weekly/monthly limits are reached.\n\t\"Monthly usage limit reached\",\n\t\"available balance\",\n\n\t// Generic quota/budget/billing exhaustion. `insufficient_quota` is OpenAI's\n\t// quota/billing error code; the other strings cover common gateway wording.\n\t\"insufficient_quota\",\n\t\"out of budget\",\n\t\"quota exceeded\",\n\t\"billing\",\n]);\n\nconst RETRYABLE_PROVIDER_ERROR_PATTERN = buildProviderErrorPattern([\n\t// Generic provider load, HTTP status, and server-side transient failures.\n\t\"overloaded\",\n\t\"rate.?limit\",\n\t\"too many requests\",\n\t\"429\",\n\t\"500\",\n\t\"502\",\n\t\"503\",\n\t\"504\",\n\t\"524\",\n\t\"service.?unavailable\",\n\t\"server.?error\",\n\t\"internal.?error\",\n\n\t// Wrapper/provider text for transient upstream failures, including OpenRouter\n\t// \"Provider returned error\" responses (#2264).\n\t\"provider.?returned.?error\",\n\n\t// Network, proxy, and fetch transport failures. This includes OpenAI Codex\n\t// raw-fetch failures such as \"upstream connect\", \"connection refused\", and\n\t// \"reset before headers\" (#733), plus OpenRouter connection drops (#3317).\n\t\"network.?error\",\n\t\"connection.?error\",\n\t\"connection.?refused\",\n\t\"connection.?lost\",\n\t\"other side closed\",\n\t\"fetch failed\",\n\t\"upstream.?connect\",\n\t\"reset before headers\",\n\t\"socket hang up\",\n\t\"timed? out\",\n\t\"timeout\",\n\t\"terminated\",\n\n\t// WebSocket transports can report close/error text instead of HTTP/fetch text.\n\t\"websocket.?closed\",\n\t\"websocket.?error\",\n\n\t// Premature stream endings from SDKs and transports. Anthropic can throw\n\t// \"stream ended without ...\" and \"Anthropic stream ended before message_stop\"\n\t// (#4433); Bedrock/Smithy can throw an HTTP/2 no-response error (#3594).\n\t\"ended without\",\n\t\"stream ended before message_stop\",\n\t\"http2 request did not get a response\",\n\n\t// Provider-requested retry delay cap failures should flow through the outer\n\t// retry policy so callers can surface/abort the backoff (#1123).\n\t\"retry delay\",\n\n\t// Explicit retry guidance emitted mid-stream by OpenAI Responses and Bedrock\n\t// stream exceptions (#6019).\n\t\"you can retry your request\",\n\t\"try your request again\",\n\t\"please retry your request\",\n]);\n\n/**\n * Classifies whether a failed assistant message looks like a transient provider\n * or transport error, so callers can decide if the last assistant turn should be\n * restarted.\n *\n * This does not implement retry policy. Callers should first handle context\n * overflow separately, then apply their own retry budget, backoff, and reporting\n * before restarting the assistant turn.\n */\nexport function isRetryableAssistantError(message: AssistantMessage): boolean {\n\tif (message.stopReason !== \"error\" || !message.errorMessage) return false;\n\tconst errorMessage = message.errorMessage;\n\tif (NON_RETRYABLE_PROVIDER_LIMIT_ERROR_PATTERN.test(errorMessage)) return false;\n\treturn RETRYABLE_PROVIDER_ERROR_PATTERN.test(errorMessage);\n}\n"]}
+{"version":3,"file":"retry.js","sourceRoot":"","sources":["../../src/utils/retry.ts"],"names":[],"mappings":"AAEA,SAAS,yBAAyB,CAAC,QAA2B;IAC7D,OAAO,IAAI,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,CAAC;AAC5C,CAAC;AAED,MAAM,0CAA0C,GAAG,yBAAyB,CAAC;IAC5E,8EAA8E;IAC9E,2EAA2E;IAC3E,mBAAmB;IACnB,qBAAqB;IAErB,6EAA6E;IAC7E,yDAAyD;IACzD,6BAA6B;IAC7B,mBAAmB;IAEnB,4EAA4E;IAC5E,4EAA4E;IAC5E,oBAAoB;IACpB,eAAe;IACf,gBAAgB;IAChB,SAAS;CACT,CAAC,CAAC;AAEH,MAAM,gCAAgC,GAAG,yBAAyB,CAAC;IAClE,0EAA0E;IAC1E,YAAY;IACZ,aAAa;IACb,mBAAmB;IACnB,KAAK;IACL,KAAK;IACL,KAAK;IACL,KAAK;IACL,KAAK;IACL,KAAK;IACL,sBAAsB;IACtB,eAAe;IACf,iBAAiB;IAEjB,8EAA8E;IAC9E,+CAA+C;IAC/C,2BAA2B;IAE3B,2EAA2E;IAC3E,2EAA2E;IAC3E,2EAA2E;IAC3E,gBAAgB;IAChB,mBAAmB;IACnB,qBAAqB;IACrB,kBAAkB;IAClB,mBAAmB;IACnB,cAAc;IACd,mBAAmB;IACnB,sBAAsB;IACtB,gBAAgB;IAChB,8BAA8B;IAC9B,YAAY;IACZ,SAAS;IACT,YAAY;IAEZ,+EAA+E;IAC/E,mBAAmB;IACnB,kBAAkB;IAElB,yEAAyE;IACzE,8EAA8E;IAC9E,yEAAyE;IACzE,eAAe;IACf,kCAAkC;IAClC,sCAAsC;IAEtC,4EAA4E;IAC5E,iEAAiE;IACjE,aAAa;IAEb,6EAA6E;IAC7E,6BAA6B;IAC7B,4BAA4B;IAC5B,wBAAwB;IACxB,2BAA2B;IAE3B,yCAAyC;IACzC,mBAAmB;CACnB,CAAC,CAAC;AAEH;;;;;;;;GAQG;AACH,MAAM,UAAU,yBAAyB,CAAC,OAAyB;IAClE,IAAI,OAAO,CAAC,UAAU,KAAK,OAAO,IAAI,CAAC,OAAO,CAAC,YAAY;QAAE,OAAO,KAAK,CAAC;IAC1E,MAAM,YAAY,GAAG,OAAO,CAAC,YAAY,CAAC;IAC1C,IAAI,0CAA0C,CAAC,IAAI,CAAC,YAAY,CAAC;QAAE,OAAO,KAAK,CAAC;IAChF,OAAO,gCAAgC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;AAC5D,CAAC","sourcesContent":["import type { AssistantMessage } from \"../types.ts\";\n\nfunction buildProviderErrorPattern(patterns: readonly string[]): RegExp {\n\treturn new RegExp(patterns.join(\"|\"), \"i\");\n}\n\nconst NON_RETRYABLE_PROVIDER_LIMIT_ERROR_PATTERN = buildProviderErrorPattern([\n\t// OpenCode Go/free-tier limits returned as 429 JSON error types by OpenCode's\n\t// Zen API. These are subscription/account limits, not transient throttles.\n\t\"GoUsageLimitError\",\n\t\"FreeUsageLimitError\",\n\n\t// OpenCode Go subscription-limit text asks users to enable available-balance\n\t// usage after rolling/weekly/monthly limits are reached.\n\t\"Monthly usage limit reached\",\n\t\"available balance\",\n\n\t// Generic quota/budget/billing exhaustion. `insufficient_quota` is OpenAI's\n\t// quota/billing error code; the other strings cover common gateway wording.\n\t\"insufficient_quota\",\n\t\"out of budget\",\n\t\"quota exceeded\",\n\t\"billing\",\n]);\n\nconst RETRYABLE_PROVIDER_ERROR_PATTERN = buildProviderErrorPattern([\n\t// Generic provider load, HTTP status, and server-side transient failures.\n\t\"overloaded\",\n\t\"rate.?limit\",\n\t\"too many requests\",\n\t\"429\",\n\t\"500\",\n\t\"502\",\n\t\"503\",\n\t\"504\",\n\t\"524\",\n\t\"service.?unavailable\",\n\t\"server.?error\",\n\t\"internal.?error\",\n\n\t// Wrapper/provider text for transient upstream failures, including OpenRouter\n\t// \"Provider returned error\" responses (#2264).\n\t\"provider.?returned.?error\",\n\n\t// Network, proxy, and fetch transport failures. This includes OpenAI Codex\n\t// raw-fetch failures such as \"upstream connect\", \"connection refused\", and\n\t// \"reset before headers\" (#733), plus OpenRouter connection drops (#3317).\n\t\"network.?error\",\n\t\"connection.?error\",\n\t\"connection.?refused\",\n\t\"connection.?lost\",\n\t\"other side closed\",\n\t\"fetch failed\",\n\t\"upstream.?connect\",\n\t\"reset before headers\",\n\t\"socket hang up\",\n\t\"socket connection was closed\",\n\t\"timed? out\",\n\t\"timeout\",\n\t\"terminated\",\n\n\t// WebSocket transports can report close/error text instead of HTTP/fetch text.\n\t\"websocket.?closed\",\n\t\"websocket.?error\",\n\n\t// Premature stream endings from SDKs and transports. Anthropic can throw\n\t// \"stream ended without ...\" and \"Anthropic stream ended before message_stop\"\n\t// (#4433); Bedrock/Smithy can throw an HTTP/2 no-response error (#3594).\n\t\"ended without\",\n\t\"stream ended before message_stop\",\n\t\"http2 request did not get a response\",\n\n\t// Provider-requested retry delay cap failures should flow through the outer\n\t// retry policy so callers can surface/abort the backoff (#1123).\n\t\"retry delay\",\n\n\t// Explicit retry guidance emitted mid-stream by OpenAI Responses and Bedrock\n\t// stream exceptions (#6019).\n\t\"you can retry your request\",\n\t\"try your request again\",\n\t\"please retry your request\",\n\n\t// gRPC based providers (e.g. NVIDIA NIM)\n\t\"ResourceExhausted\",\n]);\n\n/**\n * Classifies whether a failed assistant message looks like a transient provider\n * or transport error, so callers can decide if the last assistant turn should be\n * restarted.\n *\n * This does not implement retry policy. Callers should first handle context\n * overflow separately, then apply their own retry budget, backoff, and reporting\n * before restarting the assistant turn.\n */\nexport function isRetryableAssistantError(message: AssistantMessage): boolean {\n\tif (message.stopReason !== \"error\" || !message.errorMessage) return false;\n\tconst errorMessage = message.errorMessage;\n\tif (NON_RETRYABLE_PROVIDER_LIMIT_ERROR_PATTERN.test(errorMessage)) return false;\n\treturn RETRYABLE_PROVIDER_ERROR_PATTERN.test(errorMessage);\n}\n"]}

package/dist/utils/sanitize-unicode.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"sanitize-unicode.d.ts","sourceRoot":"","sources":["../../src/utils/sanitize-unicode.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AACH,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAIvD","sourcesContent":["/**\n * Removes unpaired Unicode surrogate characters from a string.\n *\n * Unpaired surrogates (high surrogates 0xD800-0xDBFF without matching low surrogates 0xDC00-0xDFFF,\n * or vice versa) cause JSON serialization errors in many API providers.\n *\n * Valid emoji and other characters outside the Basic Multilingual Plane use properly paired\n * surrogates and will NOT be affected by this function.\n *\n * @param text - The text to sanitize\n * @returns The sanitized text with unpaired surrogates removed\n *\n * @example\n * // Valid emoji (properly paired surrogates) are preserved\n * sanitizeSurrogates(\"Hello 🙈 World\") // => \"Hello 🙈 World\"\n *\n * // Unpaired high surrogate is removed\n * const unpaired = String.fromCharCode(0xD83D); // high surrogate without low\n * sanitizeSurrogates(`Text ${unpaired} here`) // => \"Text  here\"\n */\nexport function sanitizeSurrogates(text: string): string {\n\t// Replace unpaired high surrogates (0xD800-0xDBFF not followed by low surrogate)\n\t// Replace unpaired low surrogates (0xDC00-0xDFFF not preceded by high surrogate)\n\treturn text.replace(/[\\uD800-\\uDBFF](?![\\uDC00-\\uDFFF])|(?<![\\uD800-\\uDBFF])[\\uDC00-\\uDFFF]/g, \"\");\n}\n"]}
+{"version":3,"file":"sanitize-unicode.d.ts","sourceRoot":"","sources":["../../src/utils/sanitize-unicode.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AACH,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAIvD"}

package/dist/utils/sanitize-unicode.js.map

@@ -1 +1 @@
-{"version":3,"file":"sanitize-unicode.js","sourceRoot":"","sources":["../../src/utils/sanitize-unicode.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AACH,MAAM,UAAU,kBAAkB,CAAC,IAAY,EAAU;IACxD,iFAAiF;IACjF,iFAAiF;IACjF,OAAO,IAAI,CAAC,OAAO,CAAC,yEAAyE,EAAE,EAAE,CAAC,CAAC;AAAA,CACnG","sourcesContent":["/**\n * Removes unpaired Unicode surrogate characters from a string.\n *\n * Unpaired surrogates (high surrogates 0xD800-0xDBFF without matching low surrogates 0xDC00-0xDFFF,\n * or vice versa) cause JSON serialization errors in many API providers.\n *\n * Valid emoji and other characters outside the Basic Multilingual Plane use properly paired\n * surrogates and will NOT be affected by this function.\n *\n * @param text - The text to sanitize\n * @returns The sanitized text with unpaired surrogates removed\n *\n * @example\n * // Valid emoji (properly paired surrogates) are preserved\n * sanitizeSurrogates(\"Hello 🙈 World\") // => \"Hello 🙈 World\"\n *\n * // Unpaired high surrogate is removed\n * const unpaired = String.fromCharCode(0xD83D); // high surrogate without low\n * sanitizeSurrogates(`Text ${unpaired} here`) // => \"Text  here\"\n */\nexport function sanitizeSurrogates(text: string): string {\n\t// Replace unpaired high surrogates (0xD800-0xDBFF not followed by low surrogate)\n\t// Replace unpaired low surrogates (0xDC00-0xDFFF not preceded by high surrogate)\n\treturn text.replace(/[\\uD800-\\uDBFF](?![\\uDC00-\\uDFFF])|(?<![\\uD800-\\uDBFF])[\\uDC00-\\uDFFF]/g, \"\");\n}\n"]}
+{"version":3,"file":"sanitize-unicode.js","sourceRoot":"","sources":["../../src/utils/sanitize-unicode.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AACH,MAAM,UAAU,kBAAkB,CAAC,IAAY;IAC9C,iFAAiF;IACjF,iFAAiF;IACjF,OAAO,IAAI,CAAC,OAAO,CAAC,yEAAyE,EAAE,EAAE,CAAC,CAAC;AACpG,CAAC","sourcesContent":["/**\n * Removes unpaired Unicode surrogate characters from a string.\n *\n * Unpaired surrogates (high surrogates 0xD800-0xDBFF without matching low surrogates 0xDC00-0xDFFF,\n * or vice versa) cause JSON serialization errors in many API providers.\n *\n * Valid emoji and other characters outside the Basic Multilingual Plane use properly paired\n * surrogates and will NOT be affected by this function.\n *\n * @param text - The text to sanitize\n * @returns The sanitized text with unpaired surrogates removed\n *\n * @example\n * // Valid emoji (properly paired surrogates) are preserved\n * sanitizeSurrogates(\"Hello 🙈 World\") // => \"Hello 🙈 World\"\n *\n * // Unpaired high surrogate is removed\n * const unpaired = String.fromCharCode(0xD83D); // high surrogate without low\n * sanitizeSurrogates(`Text ${unpaired} here`) // => \"Text  here\"\n */\nexport function sanitizeSurrogates(text: string): string {\n\t// Replace unpaired high surrogates (0xD800-0xDBFF not followed by low surrogate)\n\t// Replace unpaired low surrogates (0xDC00-0xDFFF not preceded by high surrogate)\n\treturn text.replace(/[\\uD800-\\uDBFF](?![\\uDC00-\\uDFFF])|(?<![\\uD800-\\uDBFF])[\\uDC00-\\uDFFF]/g, \"\");\n}\n"]}

package/dist/utils/typebox-helpers.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"typebox-helpers.d.ts","sourceRoot":"","sources":["../../src/utils/typebox-helpers.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,OAAO,EAAQ,MAAM,SAAS,CAAC;AAE7C;;;;;;;;;;GAUG;AACH,wBAAgB,UAAU,CAAC,CAAC,SAAS,SAAS,MAAM,EAAE,EACrD,MAAM,EAAE,CAAC,EACT,OAAO,CAAC,EAAE;IAAE,WAAW,CAAC,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC,CAAC,MAAM,CAAC,CAAA;CAAE,GACrD,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAOpB","sourcesContent":["import { type TUnsafe, Type } from \"typebox\";\n\n/**\n * Creates a string enum schema compatible with Google's API and other providers\n * that don't support anyOf/const patterns.\n *\n * @example\n * const OperationSchema = StringEnum([\"add\", \"subtract\", \"multiply\", \"divide\"], {\n *   description: \"The operation to perform\"\n * });\n *\n * type Operation = Static<typeof OperationSchema>; // \"add\" | \"subtract\" | \"multiply\" | \"divide\"\n */\nexport function StringEnum<T extends readonly string[]>(\n\tvalues: T,\n\toptions?: { description?: string; default?: T[number] },\n): TUnsafe<T[number]> {\n\treturn Type.Unsafe<T[number]>({\n\t\ttype: \"string\",\n\t\tenum: values as any,\n\t\t...(options?.description && { description: options.description }),\n\t\t...(options?.default && { default: options.default }),\n\t});\n}\n"]}
+{"version":3,"file":"typebox-helpers.d.ts","sourceRoot":"","sources":["../../src/utils/typebox-helpers.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,OAAO,EAAQ,MAAM,SAAS,CAAC;AAE7C;;;;;;;;;;GAUG;AACH,wBAAgB,UAAU,CAAC,CAAC,SAAS,SAAS,MAAM,EAAE,EACrD,MAAM,EAAE,CAAC,EACT,OAAO,CAAC,EAAE;IAAE,WAAW,CAAC,EAAE,MAAM,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC,CAAC,MAAM,CAAC,CAAA;CAAE,GACrD,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAOpB"}

package/dist/utils/typebox-helpers.js.map

@@ -1 +1 @@
-{"version":3,"file":"typebox-helpers.js","sourceRoot":"","sources":["../../src/utils/typebox-helpers.ts"],"names":[],"mappings":"AAAA,OAAO,EAAgB,IAAI,EAAE,MAAM,SAAS,CAAC;AAE7C;;;;;;;;;;GAUG;AACH,MAAM,UAAU,UAAU,CACzB,MAAS,EACT,OAAuD,EAClC;IACrB,OAAO,IAAI,CAAC,MAAM,CAAY;QAC7B,IAAI,EAAE,QAAQ;QACd,IAAI,EAAE,MAAa;QACnB,GAAG,CAAC,OAAO,EAAE,WAAW,IAAI,EAAE,WAAW,EAAE,OAAO,CAAC,WAAW,EAAE,CAAC;QACjE,GAAG,CAAC,OAAO,EAAE,OAAO,IAAI,EAAE,OAAO,EAAE,OAAO,CAAC,OAAO,EAAE,CAAC;KACrD,CAAC,CAAC;AAAA,CACH","sourcesContent":["import { type TUnsafe, Type } from \"typebox\";\n\n/**\n * Creates a string enum schema compatible with Google's API and other providers\n * that don't support anyOf/const patterns.\n *\n * @example\n * const OperationSchema = StringEnum([\"add\", \"subtract\", \"multiply\", \"divide\"], {\n *   description: \"The operation to perform\"\n * });\n *\n * type Operation = Static<typeof OperationSchema>; // \"add\" | \"subtract\" | \"multiply\" | \"divide\"\n */\nexport function StringEnum<T extends readonly string[]>(\n\tvalues: T,\n\toptions?: { description?: string; default?: T[number] },\n): TUnsafe<T[number]> {\n\treturn Type.Unsafe<T[number]>({\n\t\ttype: \"string\",\n\t\tenum: values as any,\n\t\t...(options?.description && { description: options.description }),\n\t\t...(options?.default && { default: options.default }),\n\t});\n}\n"]}
+{"version":3,"file":"typebox-helpers.js","sourceRoot":"","sources":["../../src/utils/typebox-helpers.ts"],"names":[],"mappings":"AAAA,OAAO,EAAgB,IAAI,EAAE,MAAM,SAAS,CAAC;AAE7C;;;;;;;;;;GAUG;AACH,MAAM,UAAU,UAAU,CACzB,MAAS,EACT,OAAuD;IAEvD,OAAO,IAAI,CAAC,MAAM,CAAY;QAC7B,IAAI,EAAE,QAAQ;QACd,IAAI,EAAE,MAAa;QACnB,GAAG,CAAC,OAAO,EAAE,WAAW,IAAI,EAAE,WAAW,EAAE,OAAO,CAAC,WAAW,EAAE,CAAC;QACjE,GAAG,CAAC,OAAO,EAAE,OAAO,IAAI,EAAE,OAAO,EAAE,OAAO,CAAC,OAAO,EAAE,CAAC;KACrD,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import { type TUnsafe, Type } from \"typebox\";\n\n/**\n * Creates a string enum schema compatible with Google's API and other providers\n * that don't support anyOf/const patterns.\n *\n * @example\n * const OperationSchema = StringEnum([\"add\", \"subtract\", \"multiply\", \"divide\"], {\n *   description: \"The operation to perform\"\n * });\n *\n * type Operation = Static<typeof OperationSchema>; // \"add\" | \"subtract\" | \"multiply\" | \"divide\"\n */\nexport function StringEnum<T extends readonly string[]>(\n\tvalues: T,\n\toptions?: { description?: string; default?: T[number] },\n): TUnsafe<T[number]> {\n\treturn Type.Unsafe<T[number]>({\n\t\ttype: \"string\",\n\t\tenum: values as any,\n\t\t...(options?.description && { description: options.description }),\n\t\t...(options?.default && { default: options.default }),\n\t});\n}\n"]}