@zhachory1/mewrite-ai 0.65.3

package/dist/providers/anthropic-discovery.js

@@ -0,0 +1,218 @@
+/**
+ * Model-capability discovery for Anthropic-family models.
+ *
+ * Two backends, dispatched by provider:
+ *
+ *  - github-copilot: GET {baseUrl}/models — OpenAI-style listing with rich
+ *    `capabilities.supports` (adaptive_thinking, reasoning_effort, limits.
+ *    max_context_window_tokens). The 1M context tier on Copilot is exposed
+ *    as separate model ids (for example claude-opus-4.6-1m) rather than
+ *    via a beta header; the bare id (claude-opus-4.6) is hard-capped at
+ *    200k regardless of any beta header (which the relay rejects).
+ *    Discovery surfaces every id the response returns so the user can
+ *    select the right one in the picker.
+ *
+ *  - anthropic: GET {baseUrl}/v1/models — Anthropic-native with
+ *    `capabilities.thinking.types.adaptive.supported`,
+ *    `capabilities.effort.{max,xhigh}.supported`, `max_input_tokens`.
+ *    The 1M context tier is gated by `anthropic-beta: context-1m-2025-08-07`
+ *    on the same id; we detect it by making a second probe with that
+ *    header and observing whether `max_input_tokens` rises.
+ *
+ *  - others (amazon-bedrock, google-vertex, etc.): no equivalent
+ *    discovery endpoint; the static fallback table in
+ *    anthropic-capabilities.ts is used.
+ *
+ * Discovery runs at most once per (provider, baseUrl, apiKey-hash) per
+ * process. Concurrent calls join the same in-flight promise. Failures are
+ * silent — we leave the static fallback in place and try again next
+ * process. The discovery layer never throws back to the caller.
+ *
+ * The discovered registry entries (new model ids, corrected contextWindow)
+ * are also pushed into the in-memory model registry via
+ * `mergeDiscoveredModels` so the model picker reflects reality.
+ */
+import { CONTEXT_1M_BETA_HEADER, setDiscoveredCapabilities, } from "./anthropic-capabilities.js";
+// ----------------------------------------------------------------------------
+// Discovery state (per process)
+// ----------------------------------------------------------------------------
+const discoveryPromises = new Map();
+function discoveryKey(provider, baseUrl) {
+    return `${provider}::${baseUrl}`;
+}
+/**
+ * Trigger discovery for a provider+baseUrl. Returns a Promise that resolves
+ * once the capability cache has been populated (or the attempt has failed).
+ * Subsequent calls with the same (provider, baseUrl) join the in-flight
+ * promise.
+ *
+ * Never throws. On any error the static fallback remains in place.
+ */
+export function discoverAnthropicCapabilities(provider, baseUrl, apiKey, extraHeaders) {
+    const key = discoveryKey(provider, baseUrl);
+    const existing = discoveryPromises.get(key);
+    if (existing)
+        return existing;
+    const p = runDiscovery(provider, baseUrl, apiKey, extraHeaders).catch(() => {
+        // Swallow — discovery is best-effort; static table remains in place.
+    });
+    discoveryPromises.set(key, p);
+    return p;
+}
+/** Test-only: clear discovery memo so tests can re-run. */
+export function _clearDiscoveryStateForTests() {
+    discoveryPromises.clear();
+}
+// ----------------------------------------------------------------------------
+// Dispatcher
+// ----------------------------------------------------------------------------
+async function runDiscovery(provider, baseUrl, apiKey, extraHeaders) {
+    if (provider === "github-copilot") {
+        await discoverCopilot(baseUrl, apiKey, extraHeaders);
+        return;
+    }
+    if (provider === "anthropic") {
+        await discoverAnthropicNative(baseUrl, apiKey, extraHeaders);
+        return;
+    }
+    // No discovery endpoint for other providers (Bedrock, Vertex, etc.).
+}
+async function discoverCopilot(baseUrl, apiKey, extraHeaders) {
+    const res = await fetch(`${baseUrl}/models`, {
+        method: "GET",
+        headers: {
+            Authorization: `Bearer ${apiKey}`,
+            "User-Agent": "GitHubCopilotChat/0.35.0",
+            "Editor-Version": "vscode/1.107.0",
+            "Editor-Plugin-Version": "copilot-chat/0.35.0",
+            "Copilot-Integration-Id": "vscode-chat",
+            ...extraHeaders,
+        },
+    });
+    if (!res.ok)
+        return;
+    const body = (await res.json());
+    const entries = body?.data ?? [];
+    const discoveredModels = [];
+    for (const entry of entries) {
+        if (entry.vendor !== "Anthropic")
+            continue;
+        if (!entry.supported_endpoints?.includes("/v1/messages"))
+            continue;
+        const caps = entry.capabilities ?? {};
+        const sup = caps.supports ?? {};
+        const lim = caps.limits ?? {};
+        const adaptive = sup.adaptive_thinking === true;
+        const efforts = sup.reasoning_effort ?? [];
+        const xhighEffort = efforts.includes("max") || efforts.includes("xhigh");
+        setDiscoveredCapabilities("github-copilot", entry.id, {
+            thinkingSchema: adaptive ? "adaptive" : "legacy",
+            xhighEffort: xhighEffort || undefined,
+            contextWindow: lim.max_context_window_tokens,
+        });
+        // Build a registry entry so previously-unknown ids (e.g.
+        // claude-opus-4.6-1m) show up in the model picker.
+        discoveredModels.push({
+            id: entry.id,
+            name: entry.name || entry.id,
+            api: "anthropic-messages",
+            provider: "github-copilot",
+            baseUrl,
+            headers: {
+                "User-Agent": "GitHubCopilotChat/0.35.0",
+                "Editor-Version": "vscode/1.107.0",
+                "Editor-Plugin-Version": "copilot-chat/0.35.0",
+                "Copilot-Integration-Id": "vscode-chat",
+            },
+            reasoning: adaptive,
+            input: sup.vision ? ["text", "image"] : ["text"],
+            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+            contextWindow: lim.max_context_window_tokens ?? 128000,
+            maxTokens: lim.max_output_tokens ?? 8192,
+        });
+    }
+    mergeDiscoveredModels("github-copilot", discoveredModels);
+}
+async function discoverAnthropicNative(baseUrl, apiKey, extraHeaders) {
+    const baseHeaders = {
+        "x-api-key": apiKey,
+        "anthropic-version": "2023-06-01",
+        ...extraHeaders,
+    };
+    // First, capability map from the bare /v1/models endpoint.
+    const baseEntries = await fetchAnthropicModels(baseUrl, baseHeaders);
+    if (!baseEntries)
+        return;
+    // Then, retry with the context-1m beta and compare max_input_tokens.
+    // Models whose max_input_tokens grows with the beta need it to unlock
+    // the larger window; others reject the beta or ignore it.
+    const betaEntries = await fetchAnthropicModels(baseUrl, {
+        ...baseHeaders,
+        "anthropic-beta": CONTEXT_1M_BETA_HEADER,
+    });
+    const betaWindowById = new Map();
+    if (betaEntries) {
+        for (const entry of betaEntries) {
+            if (typeof entry.max_input_tokens === "number") {
+                betaWindowById.set(entry.id, entry.max_input_tokens);
+            }
+        }
+    }
+    for (const entry of baseEntries) {
+        const c = entry.capabilities ?? {};
+        const thinking = c.thinking ?? {};
+        const types = thinking.types ?? {};
+        const effort = c.effort ?? {};
+        const adaptive = types.adaptive?.supported === true;
+        const xhighEffort = effort.max?.supported === true || effort.xhigh?.supported === true;
+        const baseWindow = entry.max_input_tokens ?? 0;
+        const betaWindow = betaWindowById.get(entry.id) ?? 0;
+        const wantsBeta = betaWindow > baseWindow;
+        const caps = {
+            thinkingSchema: adaptive ? "adaptive" : "legacy",
+        };
+        if (xhighEffort)
+            caps.xhighEffort = true;
+        if (wantsBeta) {
+            caps.contextBeta = CONTEXT_1M_BETA_HEADER;
+            caps.contextWindow = betaWindow;
+        }
+        else if (baseWindow > 0) {
+            caps.contextWindow = baseWindow;
+        }
+        setDiscoveredCapabilities("anthropic", entry.id, caps);
+    }
+}
+async function fetchAnthropicModels(baseUrl, headers) {
+    const url = `${baseUrl.replace(/\/$/, "")}/v1/models?limit=1000`;
+    const res = await fetch(url, { method: "GET", headers });
+    if (!res.ok)
+        return null;
+    const body = (await res.json());
+    return body?.data ?? null;
+}
+// ----------------------------------------------------------------------------
+// Registry merge
+// ----------------------------------------------------------------------------
+/**
+ * Push freshly-discovered models into the in-memory registry via the hook
+ * wired by models.ts. New ids are inserted; existing ids have their
+ * capability-related fields refreshed.
+ */
+function mergeDiscoveredModels(provider, fresh) {
+    for (const m of fresh) {
+        registerModel(provider, m);
+    }
+}
+// Registry mutation is provided by models.ts at module load time to avoid a
+// circular import. Until it is wired, discovery still populates the
+// capability cache but cannot publish new model ids into the registry.
+const registryHookHolder = { fn: () => { } };
+function registerModel(provider, model) {
+    registryHookHolder.fn(provider, model);
+}
+/** Not for external use: called by models.ts at module load to wire the registry hook. */
+export function _setRegistryHook(hook) {
+    registryHookHolder.fn = hook;
+}
+//# sourceMappingURL=anthropic-discovery.js.map

package/dist/providers/anthropic-discovery.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"anthropic-discovery.js","sourceRoot":"","sources":["../../src/providers/anthropic-discovery.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AAGH,OAAO,EAEN,sBAAsB,EACtB,yBAAyB,GACzB,MAAM,6BAA6B,CAAC;AAErC,+EAA+E;AAC/E,gCAAgC;AAChC,+EAA+E;AAE/E,MAAM,iBAAiB,GAAG,IAAI,GAAG,EAAyB,CAAC;AAE3D,SAAS,YAAY,CAAC,QAAgB,EAAE,OAAe,EAAU;IAChE,OAAO,GAAG,QAAQ,KAAK,OAAO,EAAE,CAAC;AAAA,CACjC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,6BAA6B,CAC5C,QAAgB,EAChB,OAAe,EACf,MAAc,EACd,YAAqC,EACrB;IAChB,MAAM,GAAG,GAAG,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IAC5C,MAAM,QAAQ,GAAG,iBAAiB,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAC5C,IAAI,QAAQ;QAAE,OAAO,QAAQ,CAAC;IAE9B,MAAM,CAAC,GAAG,YAAY,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,YAAY,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC;QAC3E,uEAAqE;IADO,CAE5E,CAAC,CAAC;IACH,iBAAiB,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;IAC9B,OAAO,CAAC,CAAC;AAAA,CACT;AAED,2DAA2D;AAC3D,MAAM,UAAU,4BAA4B,GAAS;IACpD,iBAAiB,CAAC,KAAK,EAAE,CAAC;AAAA,CAC1B;AAED,+EAA+E;AAC/E,aAAa;AACb,+EAA+E;AAE/E,KAAK,UAAU,YAAY,CAC1B,QAAgB,EAChB,OAAe,EACf,MAAc,EACd,YAAqC,EACrB;IAChB,IAAI,QAAQ,KAAK,gBAAgB,EAAE,CAAC;QACnC,MAAM,eAAe,CAAC,OAAO,EAAE,MAAM,EAAE,YAAY,CAAC,CAAC;QACrD,OAAO;IACR,CAAC;IACD,IAAI,QAAQ,KAAK,WAAW,EAAE,CAAC;QAC9B,MAAM,uBAAuB,CAAC,OAAO,EAAE,MAAM,EAAE,YAAY,CAAC,CAAC;QAC7D,OAAO;IACR,CAAC;IACD,qEAAqE;AADpE,CAED;AA8BD,KAAK,UAAU,eAAe,CAAC,OAAe,EAAE,MAAc,EAAE,YAAqC,EAAiB;IACrH,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,OAAO,SAAS,EAAE;QAC5C,MAAM,EAAE,KAAK;QACb,OAAO,EAAE;YACR,aAAa,EAAE,UAAU,MAAM,EAAE;YACjC,YAAY,EAAE,0BAA0B;YACxC,gBAAgB,EAAE,gBAAgB;YAClC,uBAAuB,EAAE,qBAAqB;YAC9C,wBAAwB,EAAE,aAAa;YACvC,GAAG,YAAY;SACf;KACD,CAAC,CAAC;IACH,IAAI,CAAC,GAAG,CAAC,EAAE;QAAE,OAAO;IACpB,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAmC,CAAC;IAClE,MAAM,OAAO,GAAG,IAAI,EAAE,IAAI,IAAI,EAAE,CAAC;IAEjC,MAAM,gBAAgB,GAAkC,EAAE,CAAC;IAE3D,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,KAAK,CAAC,MAAM,KAAK,WAAW;YAAE,SAAS;QAC3C,IAAI,CAAC,KAAK,CAAC,mBAAmB,EAAE,QAAQ,CAAC,cAAc,CAAC;YAAE,SAAS;QAEnE,MAAM,IAAI,GAAG,KAAK,CAAC,YAAY,IAAI,EAAE,CAAC;QACtC,MAAM,GAAG,GAAG,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC;QAChC,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,IAAI,EAAE,CAAC;QAE9B,MAAM,QAAQ,GAAG,GAAG,CAAC,iBAAiB,KAAK,IAAI,CAAC;QAChD,MAAM,OAAO,GAAG,GAAG,CAAC,gBAAgB,IAAI,EAAE,CAAC;QAC3C,MAAM,WAAW,GAAG,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QAEzE,yBAAyB,CAAC,gBAAgB,EAAE,KAAK,CAAC,EAAE,EAAE;YACrD,cAAc,EAAE,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ;YAChD,WAAW,EAAE,WAAW,IAAI,SAAS;YACrC,aAAa,EAAE,GAAG,CAAC,yBAAyB;SAC5C,CAAC,CAAC;QAEH,yDAAyD;QACzD,mDAAmD;QACnD,gBAAgB,CAAC,IAAI,CAAC;YACrB,EAAE,EAAE,KAAK,CAAC,EAAE;YACZ,IAAI,EAAE,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC,EAAE;YAC5B,GAAG,EAAE,oBAAoB;YACzB,QAAQ,EAAE,gBAAgB;YAC1B,OAAO;YACP,OAAO,EAAE;gBACR,YAAY,EAAE,0BAA0B;gBACxC,gBAAgB,EAAE,gBAAgB;gBAClC,uBAAuB,EAAE,qBAAqB;gBAC9C,wBAAwB,EAAE,aAAa;aACvC;YACD,SAAS,EAAE,QAAQ;YACnB,KAAK,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;YAChD,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE;YAC1D,aAAa,EAAE,GAAG,CAAC,yBAAyB,IAAI,MAAM;YACtD,SAAS,EAAE,GAAG,CAAC,iBAAiB,IAAI,IAAI;SACxC,CAAC,CAAC;IACJ,CAAC;IAED,qBAAqB,CAAC,gBAAgB,EAAE,gBAAgB,CAAC,CAAC;AAAA,CAC1D;AAkCD,KAAK,UAAU,uBAAuB,CACrC,OAAe,EACf,MAAc,EACd,YAAqC,EACrB;IAChB,MAAM,WAAW,GAA2B;QAC3C,WAAW,EAAE,MAAM;QACnB,mBAAmB,EAAE,YAAY;QACjC,GAAG,YAAY;KACf,CAAC;IAEF,2DAA2D;IAC3D,MAAM,WAAW,GAAG,MAAM,oBAAoB,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;IACrE,IAAI,CAAC,WAAW;QAAE,OAAO;IAEzB,qEAAqE;IACrE,sEAAsE;IACtE,0DAA0D;IAC1D,MAAM,WAAW,GAAG,MAAM,oBAAoB,CAAC,OAAO,EAAE;QACvD,GAAG,WAAW;QACd,gBAAgB,EAAE,sBAAsB;KACxC,CAAC,CAAC;IACH,MAAM,cAAc,GAAG,IAAI,GAAG,EAAkB,CAAC;IACjD,IAAI,WAAW,EAAE,CAAC;QACjB,KAAK,MAAM,KAAK,IAAI,WAAW,EAAE,CAAC;YACjC,IAAI,OAAO,KAAK,CAAC,gBAAgB,KAAK,QAAQ,EAAE,CAAC;gBAChD,cAAc,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,gBAAgB,CAAC,CAAC;YACtD,CAAC;QACF,CAAC;IACF,CAAC;IAED,KAAK,MAAM,KAAK,IAAI,WAAW,EAAE,CAAC;QACjC,MAAM,CAAC,GAAG,KAAK,CAAC,YAAY,IAAI,EAAE,CAAC;QACnC,MAAM,QAAQ,GAAG,CAAC,CAAC,QAAQ,IAAI,EAAE,CAAC;QAClC,MAAM,KAAK,GAAG,QAAQ,CAAC,KAAK,IAAI,EAAE,CAAC;QACnC,MAAM,MAAM,GAAG,CAAC,CAAC,MAAM,IAAI,EAAE,CAAC;QAE9B,MAAM,QAAQ,GAAG,KAAK,CAAC,QAAQ,EAAE,SAAS,KAAK,IAAI,CAAC;QACpD,MAAM,WAAW,GAAG,MAAM,CAAC,GAAG,EAAE,SAAS,KAAK,IAAI,IAAI,MAAM,CAAC,KAAK,EAAE,SAAS,KAAK,IAAI,CAAC;QAEvF,MAAM,UAAU,GAAG,KAAK,CAAC,gBAAgB,IAAI,CAAC,CAAC;QAC/C,MAAM,UAAU,GAAG,cAAc,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC;QACrD,MAAM,SAAS,GAAG,UAAU,GAAG,UAAU,CAAC;QAE1C,MAAM,IAAI,GAA+B;YACxC,cAAc,EAAE,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ;SAChD,CAAC;QACF,IAAI,WAAW;YAAE,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC;QACzC,IAAI,SAAS,EAAE,CAAC;YACf,IAAI,CAAC,WAAW,GAAG,sBAAsB,CAAC;YAC1C,IAAI,CAAC,aAAa,GAAG,UAAU,CAAC;QACjC,CAAC;aAAM,IAAI,UAAU,GAAG,CAAC,EAAE,CAAC;YAC3B,IAAI,CAAC,aAAa,GAAG,UAAU,CAAC;QACjC,CAAC;QAED,yBAAyB,CAAC,WAAW,EAAE,KAAK,CAAC,EAAE,EAAE,IAAI,CAAC,CAAC;IACxD,CAAC;AAAA,CACD;AAED,KAAK,UAAU,oBAAoB,CAClC,OAAe,EACf,OAA+B,EACS;IACxC,MAAM,GAAG,GAAG,GAAG,OAAO,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,uBAAuB,CAAC;IACjE,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC,CAAC;IACzD,IAAI,CAAC,GAAG,CAAC,EAAE;QAAE,OAAO,IAAI,CAAC;IACzB,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAqC,CAAC;IACpE,OAAO,IAAI,EAAE,IAAI,IAAI,IAAI,CAAC;AAAA,CAC1B;AAED,+EAA+E;AAC/E,iBAAiB;AACjB,+EAA+E;AAE/E;;;;GAIG;AACH,SAAS,qBAAqB,CAAC,QAAgB,EAAE,KAAmB,EAAQ;IAC3E,KAAK,MAAM,CAAC,IAAI,KAAK,EAAE,CAAC;QACvB,aAAa,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;IAC5B,CAAC;AAAA,CACD;AAED,4EAA4E;AAC5E,oEAAoE;AACpE,uEAAuE;AACvE,MAAM,kBAAkB,GAA0D,EAAE,EAAE,EAAE,GAAG,EAAE,CAAC,EAAC,CAAC,EAAE,CAAC;AAEnG,SAAS,aAAa,CAAC,QAAgB,EAAE,KAAiB,EAAQ;IACjE,kBAAkB,CAAC,EAAE,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;AAAA,CACvC;AAED,0FAA0F;AAC1F,MAAM,UAAU,gBAAgB,CAAC,IAAmD,EAAQ;IAC3F,kBAAkB,CAAC,EAAE,GAAG,IAAI,CAAC;AAAA,CAC7B","sourcesContent":["/**\n * Model-capability discovery for Anthropic-family models.\n *\n * Two backends, dispatched by provider:\n *\n *  - github-copilot: GET {baseUrl}/models — OpenAI-style listing with rich\n *    `capabilities.supports` (adaptive_thinking, reasoning_effort, limits.\n *    max_context_window_tokens). The 1M context tier on Copilot is exposed\n *    as separate model ids (for example claude-opus-4.6-1m) rather than\n *    via a beta header; the bare id (claude-opus-4.6) is hard-capped at\n *    200k regardless of any beta header (which the relay rejects).\n *    Discovery surfaces every id the response returns so the user can\n *    select the right one in the picker.\n *\n *  - anthropic: GET {baseUrl}/v1/models — Anthropic-native with\n *    `capabilities.thinking.types.adaptive.supported`,\n *    `capabilities.effort.{max,xhigh}.supported`, `max_input_tokens`.\n *    The 1M context tier is gated by `anthropic-beta: context-1m-2025-08-07`\n *    on the same id; we detect it by making a second probe with that\n *    header and observing whether `max_input_tokens` rises.\n *\n *  - others (amazon-bedrock, google-vertex, etc.): no equivalent\n *    discovery endpoint; the static fallback table in\n *    anthropic-capabilities.ts is used.\n *\n * Discovery runs at most once per (provider, baseUrl, apiKey-hash) per\n * process. Concurrent calls join the same in-flight promise. Failures are\n * silent — we leave the static fallback in place and try again next\n * process. The discovery layer never throws back to the caller.\n *\n * The discovered registry entries (new model ids, corrected contextWindow)\n * are also pushed into the in-memory model registry via\n * `mergeDiscoveredModels` so the model picker reflects reality.\n */\n\nimport type { Api, Model } from \"../types.js\";\nimport {\n\ttype AnthropicModelCapabilities,\n\tCONTEXT_1M_BETA_HEADER,\n\tsetDiscoveredCapabilities,\n} from \"./anthropic-capabilities.js\";\n\n// ----------------------------------------------------------------------------\n// Discovery state (per process)\n// ----------------------------------------------------------------------------\n\nconst discoveryPromises = new Map<string, Promise<void>>();\n\nfunction discoveryKey(provider: string, baseUrl: string): string {\n\treturn `${provider}::${baseUrl}`;\n}\n\n/**\n * Trigger discovery for a provider+baseUrl. Returns a Promise that resolves\n * once the capability cache has been populated (or the attempt has failed).\n * Subsequent calls with the same (provider, baseUrl) join the in-flight\n * promise.\n *\n * Never throws. On any error the static fallback remains in place.\n */\nexport function discoverAnthropicCapabilities(\n\tprovider: string,\n\tbaseUrl: string,\n\tapiKey: string,\n\textraHeaders?: Record<string, string>,\n): Promise<void> {\n\tconst key = discoveryKey(provider, baseUrl);\n\tconst existing = discoveryPromises.get(key);\n\tif (existing) return existing;\n\n\tconst p = runDiscovery(provider, baseUrl, apiKey, extraHeaders).catch(() => {\n\t\t// Swallow — discovery is best-effort; static table remains in place.\n\t});\n\tdiscoveryPromises.set(key, p);\n\treturn p;\n}\n\n/** Test-only: clear discovery memo so tests can re-run. */\nexport function _clearDiscoveryStateForTests(): void {\n\tdiscoveryPromises.clear();\n}\n\n// ----------------------------------------------------------------------------\n// Dispatcher\n// ----------------------------------------------------------------------------\n\nasync function runDiscovery(\n\tprovider: string,\n\tbaseUrl: string,\n\tapiKey: string,\n\textraHeaders?: Record<string, string>,\n): Promise<void> {\n\tif (provider === \"github-copilot\") {\n\t\tawait discoverCopilot(baseUrl, apiKey, extraHeaders);\n\t\treturn;\n\t}\n\tif (provider === \"anthropic\") {\n\t\tawait discoverAnthropicNative(baseUrl, apiKey, extraHeaders);\n\t\treturn;\n\t}\n\t// No discovery endpoint for other providers (Bedrock, Vertex, etc.).\n}\n\n// ----------------------------------------------------------------------------\n// GitHub Copilot: GET {baseUrl}/models\n// ----------------------------------------------------------------------------\n\ninterface CopilotModelEntry {\n\tid: string;\n\tname?: string;\n\tvendor?: string;\n\tsupported_endpoints?: string[];\n\tmodel_picker_enabled?: boolean;\n\tcapabilities?: {\n\t\tfamily?: string;\n\t\ttype?: string;\n\t\tsupports?: {\n\t\t\tadaptive_thinking?: boolean;\n\t\t\treasoning_effort?: string[];\n\t\t\tstreaming?: boolean;\n\t\t\ttool_calls?: boolean;\n\t\t\tvision?: boolean;\n\t\t};\n\t\tlimits?: {\n\t\t\tmax_context_window_tokens?: number;\n\t\t\tmax_output_tokens?: number;\n\t\t\tmax_prompt_tokens?: number;\n\t\t};\n\t};\n}\n\nasync function discoverCopilot(baseUrl: string, apiKey: string, extraHeaders?: Record<string, string>): Promise<void> {\n\tconst res = await fetch(`${baseUrl}/models`, {\n\t\tmethod: \"GET\",\n\t\theaders: {\n\t\t\tAuthorization: `Bearer ${apiKey}`,\n\t\t\t\"User-Agent\": \"GitHubCopilotChat/0.35.0\",\n\t\t\t\"Editor-Version\": \"vscode/1.107.0\",\n\t\t\t\"Editor-Plugin-Version\": \"copilot-chat/0.35.0\",\n\t\t\t\"Copilot-Integration-Id\": \"vscode-chat\",\n\t\t\t...extraHeaders,\n\t\t},\n\t});\n\tif (!res.ok) return;\n\tconst body = (await res.json()) as { data?: CopilotModelEntry[] };\n\tconst entries = body?.data ?? [];\n\n\tconst discoveredModels: Model<\"anthropic-messages\">[] = [];\n\n\tfor (const entry of entries) {\n\t\tif (entry.vendor !== \"Anthropic\") continue;\n\t\tif (!entry.supported_endpoints?.includes(\"/v1/messages\")) continue;\n\n\t\tconst caps = entry.capabilities ?? {};\n\t\tconst sup = caps.supports ?? {};\n\t\tconst lim = caps.limits ?? {};\n\n\t\tconst adaptive = sup.adaptive_thinking === true;\n\t\tconst efforts = sup.reasoning_effort ?? [];\n\t\tconst xhighEffort = efforts.includes(\"max\") || efforts.includes(\"xhigh\");\n\n\t\tsetDiscoveredCapabilities(\"github-copilot\", entry.id, {\n\t\t\tthinkingSchema: adaptive ? \"adaptive\" : \"legacy\",\n\t\t\txhighEffort: xhighEffort || undefined,\n\t\t\tcontextWindow: lim.max_context_window_tokens,\n\t\t});\n\n\t\t// Build a registry entry so previously-unknown ids (e.g.\n\t\t// claude-opus-4.6-1m) show up in the model picker.\n\t\tdiscoveredModels.push({\n\t\t\tid: entry.id,\n\t\t\tname: entry.name || entry.id,\n\t\t\tapi: \"anthropic-messages\",\n\t\t\tprovider: \"github-copilot\",\n\t\t\tbaseUrl,\n\t\t\theaders: {\n\t\t\t\t\"User-Agent\": \"GitHubCopilotChat/0.35.0\",\n\t\t\t\t\"Editor-Version\": \"vscode/1.107.0\",\n\t\t\t\t\"Editor-Plugin-Version\": \"copilot-chat/0.35.0\",\n\t\t\t\t\"Copilot-Integration-Id\": \"vscode-chat\",\n\t\t\t},\n\t\t\treasoning: adaptive,\n\t\t\tinput: sup.vision ? [\"text\", \"image\"] : [\"text\"],\n\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },\n\t\t\tcontextWindow: lim.max_context_window_tokens ?? 128000,\n\t\t\tmaxTokens: lim.max_output_tokens ?? 8192,\n\t\t});\n\t}\n\n\tmergeDiscoveredModels(\"github-copilot\", discoveredModels);\n}\n\n// ----------------------------------------------------------------------------\n// Direct Anthropic: GET {baseUrl}/v1/models\n// ----------------------------------------------------------------------------\n\ninterface AnthropicCapabilitySupport {\n\tsupported?: boolean;\n}\ninterface AnthropicModelEntry {\n\tid: string;\n\tdisplay_name?: string;\n\tmax_input_tokens?: number;\n\tmax_tokens?: number;\n\tcapabilities?: {\n\t\tthinking?: {\n\t\t\tsupported?: boolean;\n\t\t\ttypes?: {\n\t\t\t\tadaptive?: AnthropicCapabilitySupport;\n\t\t\t\tenabled?: AnthropicCapabilitySupport;\n\t\t\t};\n\t\t};\n\t\teffort?: {\n\t\t\tsupported?: boolean;\n\t\t\tlow?: AnthropicCapabilitySupport;\n\t\t\tmedium?: AnthropicCapabilitySupport;\n\t\t\thigh?: AnthropicCapabilitySupport;\n\t\t\tmax?: AnthropicCapabilitySupport;\n\t\t\txhigh?: AnthropicCapabilitySupport;\n\t\t};\n\t\timage_input?: AnthropicCapabilitySupport;\n\t};\n}\n\nasync function discoverAnthropicNative(\n\tbaseUrl: string,\n\tapiKey: string,\n\textraHeaders?: Record<string, string>,\n): Promise<void> {\n\tconst baseHeaders: Record<string, string> = {\n\t\t\"x-api-key\": apiKey,\n\t\t\"anthropic-version\": \"2023-06-01\",\n\t\t...extraHeaders,\n\t};\n\n\t// First, capability map from the bare /v1/models endpoint.\n\tconst baseEntries = await fetchAnthropicModels(baseUrl, baseHeaders);\n\tif (!baseEntries) return;\n\n\t// Then, retry with the context-1m beta and compare max_input_tokens.\n\t// Models whose max_input_tokens grows with the beta need it to unlock\n\t// the larger window; others reject the beta or ignore it.\n\tconst betaEntries = await fetchAnthropicModels(baseUrl, {\n\t\t...baseHeaders,\n\t\t\"anthropic-beta\": CONTEXT_1M_BETA_HEADER,\n\t});\n\tconst betaWindowById = new Map<string, number>();\n\tif (betaEntries) {\n\t\tfor (const entry of betaEntries) {\n\t\t\tif (typeof entry.max_input_tokens === \"number\") {\n\t\t\t\tbetaWindowById.set(entry.id, entry.max_input_tokens);\n\t\t\t}\n\t\t}\n\t}\n\n\tfor (const entry of baseEntries) {\n\t\tconst c = entry.capabilities ?? {};\n\t\tconst thinking = c.thinking ?? {};\n\t\tconst types = thinking.types ?? {};\n\t\tconst effort = c.effort ?? {};\n\n\t\tconst adaptive = types.adaptive?.supported === true;\n\t\tconst xhighEffort = effort.max?.supported === true || effort.xhigh?.supported === true;\n\n\t\tconst baseWindow = entry.max_input_tokens ?? 0;\n\t\tconst betaWindow = betaWindowById.get(entry.id) ?? 0;\n\t\tconst wantsBeta = betaWindow > baseWindow;\n\n\t\tconst caps: AnthropicModelCapabilities = {\n\t\t\tthinkingSchema: adaptive ? \"adaptive\" : \"legacy\",\n\t\t};\n\t\tif (xhighEffort) caps.xhighEffort = true;\n\t\tif (wantsBeta) {\n\t\t\tcaps.contextBeta = CONTEXT_1M_BETA_HEADER;\n\t\t\tcaps.contextWindow = betaWindow;\n\t\t} else if (baseWindow > 0) {\n\t\t\tcaps.contextWindow = baseWindow;\n\t\t}\n\n\t\tsetDiscoveredCapabilities(\"anthropic\", entry.id, caps);\n\t}\n}\n\nasync function fetchAnthropicModels(\n\tbaseUrl: string,\n\theaders: Record<string, string>,\n): Promise<AnthropicModelEntry[] | null> {\n\tconst url = `${baseUrl.replace(/\\/$/, \"\")}/v1/models?limit=1000`;\n\tconst res = await fetch(url, { method: \"GET\", headers });\n\tif (!res.ok) return null;\n\tconst body = (await res.json()) as { data?: AnthropicModelEntry[] };\n\treturn body?.data ?? null;\n}\n\n// ----------------------------------------------------------------------------\n// Registry merge\n// ----------------------------------------------------------------------------\n\n/**\n * Push freshly-discovered models into the in-memory registry via the hook\n * wired by models.ts. New ids are inserted; existing ids have their\n * capability-related fields refreshed.\n */\nfunction mergeDiscoveredModels(provider: string, fresh: Model<Api>[]): void {\n\tfor (const m of fresh) {\n\t\tregisterModel(provider, m);\n\t}\n}\n\n// Registry mutation is provided by models.ts at module load time to avoid a\n// circular import. Until it is wired, discovery still populates the\n// capability cache but cannot publish new model ids into the registry.\nconst registryHookHolder: { fn: (provider: string, model: Model<Api>) => void } = { fn: () => {} };\n\nfunction registerModel(provider: string, model: Model<Api>): void {\n\tregistryHookHolder.fn(provider, model);\n}\n\n/** Not for external use: called by models.ts at module load to wire the registry hook. */\nexport function _setRegistryHook(hook: (provider: string, model: Model<Api>) => void): void {\n\tregistryHookHolder.fn = hook;\n}\n"]}

package/dist/providers/anthropic.d.ts

@@ -0,0 +1,40 @@
+import Anthropic from "@anthropic-ai/sdk";
+import type { SimpleStreamOptions, StreamFunction, StreamOptions } from "../types.js";
+export type AnthropicEffort = "low" | "medium" | "high" | "max";
+export interface AnthropicOptions extends StreamOptions {
+    /**
+     * Enable extended thinking.
+     * For Opus 4.6 and Sonnet 4.6: uses adaptive thinking (model decides when/how much to think).
+     * For older models: uses budget-based thinking with thinkingBudgetTokens.
+     */
+    thinkingEnabled?: boolean;
+    /**
+     * Token budget for extended thinking (older models only).
+     * Ignored for Opus 4.6 and Sonnet 4.6, which use adaptive thinking.
+     */
+    thinkingBudgetTokens?: number;
+    /**
+     * Effort level for adaptive thinking (Opus 4.6 and Sonnet 4.6).
+     * Controls how much thinking Claude allocates:
+     * - "max": Always thinks with no constraints (Opus 4.6 only)
+     * - "high": Always thinks, deep reasoning (default)
+     * - "medium": Moderate thinking, may skip for simple queries
+     * - "low": Minimal thinking, skips for simple tasks
+     * Ignored for older models.
+     */
+    effort?: AnthropicEffort;
+    interleavedThinking?: boolean;
+    toolChoice?: "auto" | "any" | "none" | {
+        type: "tool";
+        name: string;
+    };
+    /**
+     * Pre-built Anthropic client instance. When provided, skips internal client
+     * construction entirely. Use this to inject alternative SDK clients such as
+     * `AnthropicVertex` that shares the same messaging API.
+     */
+    client?: Anthropic;
+}
+export declare const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOptions>;
+export declare const streamSimpleAnthropic: StreamFunction<"anthropic-messages", SimpleStreamOptions>;
+//# sourceMappingURL=anthropic.d.ts.map

package/dist/providers/anthropic.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"anthropic.d.ts","sourceRoot":"","sources":["../../src/providers/anthropic.ts"],"names":[],"mappings":"AAAA,OAAO,SAAS,MAAM,mBAAmB,CAAC;AAQ1C,OAAO,KAAK,EAQX,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAMb,MAAM,aAAa,CAAC;AAmIrB,MAAM,MAAM,eAAe,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,KAAK,CAAC;AAEhE,MAAM,WAAW,gBAAiB,SAAQ,aAAa;IACtD;;;;OAIG;IACH,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B;;;OAGG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B;;;;;;;;OAQG;IACH,MAAM,CAAC,EAAE,eAAe,CAAC;IACzB,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,UAAU,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IACtE;;;;OAIG;IACH,MAAM,CAAC,EAAE,SAAS,CAAC;CACnB;AAYD,eAAO,MAAM,eAAe,EAAE,cAAc,CAAC,oBAAoB,EAAE,gBAAgB,CA2PlF,CAAC;AA2BF,eAAO,MAAM,qBAAqB,EAAE,cAAc,CAAC,oBAAoB,EAAE,mBAAmB,CAuC3F,CAAC","sourcesContent":["import Anthropic from \"@anthropic-ai/sdk\";\nimport type {\n\tContentBlockParam,\n\tMessageCreateParamsStreaming,\n\tMessageParam,\n} from \"@anthropic-ai/sdk/resources/messages.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { calculateCost } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tImageContent,\n\tMessage,\n\tModel,\n\tSimpleStreamOptions,\n\tStopReason,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingContent,\n\tTool,\n\tToolCall,\n\tToolResultMessage,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { parseStreamingJson } from \"../utils/json-parse.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\n\nimport { getAnthropicCapabilities, supportsAdaptiveThinking } from \"./anthropic-capabilities.js\";\nimport { discoverAnthropicCapabilities } from \"./anthropic-discovery.js\";\nimport { buildCopilotDynamicHeaders, hasCopilotVisionInput } from \"./github-copilot-headers.js\";\nimport { adjustMaxTokensForThinking, buildBaseOptions } from \"./simple-options.js\";\nimport { transformMessages } from \"./transform-messages.js\";\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses PI_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (typeof process !== \"undefined\" && process.env.PI_CACHE_RETENTION === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\nfunction getCacheControl(\n\tbaseUrl: string,\n\tcacheRetention?: CacheRetention,\n): { retention: CacheRetention; cacheControl?: { type: \"ephemeral\"; ttl?: \"1h\" } } {\n\tconst retention = resolveCacheRetention(cacheRetention);\n\tif (retention === \"none\") {\n\t\treturn { retention };\n\t}\n\tconst ttl = retention === \"long\" && baseUrl.includes(\"api.anthropic.com\") ? \"1h\" : undefined;\n\treturn {\n\t\tretention,\n\t\tcacheControl: { type: \"ephemeral\", ...(ttl && { ttl }) },\n\t};\n}\n\n// Stealth mode: Mimic Claude Code's tool naming exactly\nconst claudeCodeVersion = \"2.1.75\";\n\n// Claude Code 2.x tool names (canonical casing)\n// Source: https://cchistory.mariozechner.at/data/prompts-2.1.11.md\n// To update: https://github.com/badlogic/cchistory\nconst claudeCodeTools = [\n\t\"Read\",\n\t\"Write\",\n\t\"Edit\",\n\t\"Bash\",\n\t\"Grep\",\n\t\"Glob\",\n\t\"AskUserQuestion\",\n\t\"EnterPlanMode\",\n\t\"ExitPlanMode\",\n\t\"KillShell\",\n\t\"NotebookEdit\",\n\t\"Skill\",\n\t\"Task\",\n\t\"TaskOutput\",\n\t\"TodoWrite\",\n\t\"WebFetch\",\n\t\"WebSearch\",\n];\n\nconst ccToolLookup = new Map(claudeCodeTools.map((t) => [t.toLowerCase(), t]));\n\n// Convert tool name to CC canonical casing if it matches (case-insensitive)\nconst toClaudeCodeName = (name: string) => ccToolLookup.get(name.toLowerCase()) ?? name;\nconst fromClaudeCodeName = (name: string, tools?: Tool[]) => {\n\tif (tools && tools.length > 0) {\n\t\tconst lowerName = name.toLowerCase();\n\t\tconst matchedTool = tools.find((tool) => tool.name.toLowerCase() === lowerName);\n\t\tif (matchedTool) return matchedTool.name;\n\t}\n\treturn name;\n};\n\n/**\n * Convert content blocks to Anthropic API format\n */\nfunction convertContentBlocks(content: (TextContent | ImageContent)[]):\n\t| string\n\t| Array<\n\t\t\t| { type: \"text\"; text: string }\n\t\t\t| {\n\t\t\t\t\ttype: \"image\";\n\t\t\t\t\tsource: {\n\t\t\t\t\t\ttype: \"base64\";\n\t\t\t\t\t\tmedia_type: \"image/jpeg\" | \"image/png\" | \"image/gif\" | \"image/webp\";\n\t\t\t\t\t\tdata: string;\n\t\t\t\t\t};\n\t\t\t  }\n\t  > {\n\t// If only text blocks, return as concatenated string for simplicity\n\tconst hasImages = content.some((c) => c.type === \"image\");\n\tif (!hasImages) {\n\t\treturn sanitizeSurrogates(content.map((c) => (c as TextContent).text).join(\"\\n\"));\n\t}\n\n\t// If we have images, convert to content block array\n\tconst blocks = content.map((block) => {\n\t\tif (block.type === \"text\") {\n\t\t\treturn {\n\t\t\t\ttype: \"text\" as const,\n\t\t\t\ttext: sanitizeSurrogates(block.text),\n\t\t\t};\n\t\t}\n\t\treturn {\n\t\t\ttype: \"image\" as const,\n\t\t\tsource: {\n\t\t\t\ttype: \"base64\" as const,\n\t\t\t\tmedia_type: block.mimeType as \"image/jpeg\" | \"image/png\" | \"image/gif\" | \"image/webp\",\n\t\t\t\tdata: block.data,\n\t\t\t},\n\t\t};\n\t});\n\n\t// If only images (no text), add placeholder text block\n\tconst hasText = blocks.some((b) => b.type === \"text\");\n\tif (!hasText) {\n\t\tblocks.unshift({\n\t\t\ttype: \"text\" as const,\n\t\t\ttext: \"(see attached image)\",\n\t\t});\n\t}\n\n\treturn blocks;\n}\n\nexport type AnthropicEffort = \"low\" | \"medium\" | \"high\" | \"max\";\n\nexport interface AnthropicOptions extends StreamOptions {\n\t/**\n\t * Enable extended thinking.\n\t * For Opus 4.6 and Sonnet 4.6: uses adaptive thinking (model decides when/how much to think).\n\t * For older models: uses budget-based thinking with thinkingBudgetTokens.\n\t */\n\tthinkingEnabled?: boolean;\n\t/**\n\t * Token budget for extended thinking (older models only).\n\t * Ignored for Opus 4.6 and Sonnet 4.6, which use adaptive thinking.\n\t */\n\tthinkingBudgetTokens?: number;\n\t/**\n\t * Effort level for adaptive thinking (Opus 4.6 and Sonnet 4.6).\n\t * Controls how much thinking Claude allocates:\n\t * - \"max\": Always thinks with no constraints (Opus 4.6 only)\n\t * - \"high\": Always thinks, deep reasoning (default)\n\t * - \"medium\": Moderate thinking, may skip for simple queries\n\t * - \"low\": Minimal thinking, skips for simple tasks\n\t * Ignored for older models.\n\t */\n\teffort?: AnthropicEffort;\n\tinterleavedThinking?: boolean;\n\ttoolChoice?: \"auto\" | \"any\" | \"none\" | { type: \"tool\"; name: string };\n\t/**\n\t * Pre-built Anthropic client instance. When provided, skips internal client\n\t * construction entirely. Use this to inject alternative SDK clients such as\n\t * `AnthropicVertex` that shares the same messaging API.\n\t */\n\tclient?: Anthropic;\n}\n\nfunction mergeHeaders(...headerSources: (Record<string, string> | undefined)[]): Record<string, string> {\n\tconst merged: Record<string, string> = {};\n\tfor (const headers of headerSources) {\n\t\tif (headers) {\n\t\t\tObject.assign(merged, headers);\n\t\t}\n\t}\n\treturn merged;\n}\n\nexport const streamAnthropic: StreamFunction<\"anthropic-messages\", AnthropicOptions> = (\n\tmodel: Model<\"anthropic-messages\">,\n\tcontext: Context,\n\toptions?: AnthropicOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: model.api as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\tlet client: Anthropic;\n\t\t\tlet isOAuth: boolean;\n\n\t\t\tif (options?.client) {\n\t\t\t\tclient = options.client;\n\t\t\t\tisOAuth = false;\n\t\t\t} else {\n\t\t\t\tconst apiKey = options?.apiKey ?? getEnvApiKey(model.provider) ?? \"\";\n\n\t\t\t\t// Discover real per-account capabilities before building the\n\t\t\t\t// request so the first call uses the right thinking schema,\n\t\t\t\t// reasoning_effort ceiling, and context window. Memoized per\n\t\t\t\t// (provider, baseUrl) for the rest of the process; subsequent\n\t\t\t\t// calls resolve from cache immediately.\n\t\t\t\tif (apiKey) {\n\t\t\t\t\tawait discoverAnthropicCapabilities(model.provider, model.baseUrl, apiKey);\n\t\t\t\t}\n\n\t\t\t\tlet copilotDynamicHeaders: Record<string, string> | undefined;\n\t\t\t\tif (model.provider === \"github-copilot\") {\n\t\t\t\t\tconst hasImages = hasCopilotVisionInput(context.messages);\n\t\t\t\t\tcopilotDynamicHeaders = buildCopilotDynamicHeaders({\n\t\t\t\t\t\tmessages: context.messages,\n\t\t\t\t\t\thasImages,\n\t\t\t\t\t});\n\t\t\t\t}\n\n\t\t\t\tconst created = createClient(\n\t\t\t\t\tmodel,\n\t\t\t\t\tapiKey,\n\t\t\t\t\toptions?.interleavedThinking ?? true,\n\t\t\t\t\toptions?.headers,\n\t\t\t\t\tcopilotDynamicHeaders,\n\t\t\t\t);\n\t\t\t\tclient = created.client;\n\t\t\t\tisOAuth = created.isOAuthToken;\n\t\t\t}\n\t\t\tlet params = buildParams(model, context, isOAuth, options);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as MessageCreateParamsStreaming;\n\t\t\t}\n\t\t\tconst anthropicStream = client.messages.stream({ ...params, stream: true }, { signal: options?.signal });\n\t\t\tstream.push({ type: \"start\", partial: output });\n\n\t\t\ttype Block = (ThinkingContent | TextContent | (ToolCall & { partialJson: string })) & { index: number };\n\t\t\tconst blocks = output.content as Block[];\n\n\t\t\tfor await (const event of anthropicStream) {\n\t\t\t\tif (event.type === \"message_start\") {\n\t\t\t\t\toutput.responseId = event.message.id;\n\t\t\t\t\t// Capture initial token usage from message_start event\n\t\t\t\t\t// This ensures we have input token counts even if the stream is aborted early\n\t\t\t\t\toutput.usage.input = event.message.usage.input_tokens || 0;\n\t\t\t\t\toutput.usage.output = event.message.usage.output_tokens || 0;\n\t\t\t\t\toutput.usage.cacheRead = event.message.usage.cache_read_input_tokens || 0;\n\t\t\t\t\toutput.usage.cacheWrite = event.message.usage.cache_creation_input_tokens || 0;\n\t\t\t\t\t// Anthropic doesn't provide total_tokens, compute from components\n\t\t\t\t\toutput.usage.totalTokens =\n\t\t\t\t\t\toutput.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;\n\t\t\t\t\tcalculateCost(model, output.usage);\n\t\t\t\t} else if (event.type === \"content_block_start\") {\n\t\t\t\t\tif (event.content_block.type === \"text\") {\n\t\t\t\t\t\tconst block: Block = {\n\t\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\t\ttext: \"\",\n\t\t\t\t\t\t\tindex: event.index,\n\t\t\t\t\t\t};\n\t\t\t\t\t\toutput.content.push(block);\n\t\t\t\t\t\tstream.push({ type: \"text_start\", contentIndex: output.content.length - 1, partial: output });\n\t\t\t\t\t} else if (event.content_block.type === \"thinking\") {\n\t\t\t\t\t\tconst block: Block = {\n\t\t\t\t\t\t\ttype: \"thinking\",\n\t\t\t\t\t\t\tthinking: \"\",\n\t\t\t\t\t\t\tthinkingSignature: \"\",\n\t\t\t\t\t\t\tindex: event.index,\n\t\t\t\t\t\t};\n\t\t\t\t\t\toutput.content.push(block);\n\t\t\t\t\t\tstream.push({ type: \"thinking_start\", contentIndex: output.content.length - 1, partial: output });\n\t\t\t\t\t} else if (event.content_block.type === \"redacted_thinking\") {\n\t\t\t\t\t\tconst block: Block = {\n\t\t\t\t\t\t\ttype: \"thinking\",\n\t\t\t\t\t\t\tthinking: \"[Reasoning redacted]\",\n\t\t\t\t\t\t\tthinkingSignature: event.content_block.data,\n\t\t\t\t\t\t\tredacted: true,\n\t\t\t\t\t\t\tindex: event.index,\n\t\t\t\t\t\t};\n\t\t\t\t\t\toutput.content.push(block);\n\t\t\t\t\t\tstream.push({ type: \"thinking_start\", contentIndex: output.content.length - 1, partial: output });\n\t\t\t\t\t} else if (event.content_block.type === \"tool_use\") {\n\t\t\t\t\t\tconst block: Block = {\n\t\t\t\t\t\t\ttype: \"toolCall\",\n\t\t\t\t\t\t\tid: event.content_block.id,\n\t\t\t\t\t\t\tname: isOAuth\n\t\t\t\t\t\t\t\t? fromClaudeCodeName(event.content_block.name, context.tools)\n\t\t\t\t\t\t\t\t: event.content_block.name,\n\t\t\t\t\t\t\targuments: (event.content_block.input as Record<string, any>) ?? {},\n\t\t\t\t\t\t\tpartialJson: \"\",\n\t\t\t\t\t\t\tindex: event.index,\n\t\t\t\t\t\t};\n\t\t\t\t\t\toutput.content.push(block);\n\t\t\t\t\t\tstream.push({ type: \"toolcall_start\", contentIndex: output.content.length - 1, partial: output });\n\t\t\t\t\t}\n\t\t\t\t} else if (event.type === \"content_block_delta\") {\n\t\t\t\t\tif (event.delta.type === \"text_delta\") {\n\t\t\t\t\t\tconst index = blocks.findIndex((b) => b.index === event.index);\n\t\t\t\t\t\tconst block = blocks[index];\n\t\t\t\t\t\tif (block && block.type === \"text\") {\n\t\t\t\t\t\t\tblock.text += event.delta.text;\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"text_delta\",\n\t\t\t\t\t\t\t\tcontentIndex: index,\n\t\t\t\t\t\t\t\tdelta: event.delta.text,\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t} else if (event.delta.type === \"thinking_delta\") {\n\t\t\t\t\t\tconst index = blocks.findIndex((b) => b.index === event.index);\n\t\t\t\t\t\tconst block = blocks[index];\n\t\t\t\t\t\tif (block && block.type === \"thinking\") {\n\t\t\t\t\t\t\tblock.thinking += event.delta.thinking;\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\t\t\t\t\tcontentIndex: index,\n\t\t\t\t\t\t\t\tdelta: event.delta.thinking,\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t} else if (event.delta.type === \"input_json_delta\") {\n\t\t\t\t\t\tconst index = blocks.findIndex((b) => b.index === event.index);\n\t\t\t\t\t\tconst block = blocks[index];\n\t\t\t\t\t\tif (block && block.type === \"toolCall\") {\n\t\t\t\t\t\t\tblock.partialJson += event.delta.partial_json;\n\t\t\t\t\t\t\tblock.arguments = parseStreamingJson(block.partialJson);\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"toolcall_delta\",\n\t\t\t\t\t\t\t\tcontentIndex: index,\n\t\t\t\t\t\t\t\tdelta: event.delta.partial_json,\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t} else if (event.delta.type === \"signature_delta\") {\n\t\t\t\t\t\tconst index = blocks.findIndex((b) => b.index === event.index);\n\t\t\t\t\t\tconst block = blocks[index];\n\t\t\t\t\t\tif (block && block.type === \"thinking\") {\n\t\t\t\t\t\t\tblock.thinkingSignature = block.thinkingSignature || \"\";\n\t\t\t\t\t\t\tblock.thinkingSignature += event.delta.signature;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t} else if (event.type === \"content_block_stop\") {\n\t\t\t\t\tconst index = blocks.findIndex((b) => b.index === event.index);\n\t\t\t\t\tconst block = blocks[index];\n\t\t\t\t\tif (block) {\n\t\t\t\t\t\tdelete (block as any).index;\n\t\t\t\t\t\tif (block.type === \"text\") {\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\t\tcontentIndex: index,\n\t\t\t\t\t\t\t\tcontent: block.text,\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t} else if (block.type === \"thinking\") {\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\t\tcontentIndex: index,\n\t\t\t\t\t\t\t\tcontent: block.thinking,\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t} else if (block.type === \"toolCall\") {\n\t\t\t\t\t\t\tblock.arguments = parseStreamingJson(block.partialJson);\n\t\t\t\t\t\t\tdelete (block as any).partialJson;\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"toolcall_end\",\n\t\t\t\t\t\t\t\tcontentIndex: index,\n\t\t\t\t\t\t\t\ttoolCall: block,\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t} else if (event.type === \"message_delta\") {\n\t\t\t\t\tif (event.delta.stop_reason) {\n\t\t\t\t\t\toutput.stopReason = mapStopReason(event.delta.stop_reason);\n\t\t\t\t\t}\n\t\t\t\t\t// Only update usage fields if present (not null).\n\t\t\t\t\t// Preserves input_tokens from message_start when proxies omit it in message_delta.\n\t\t\t\t\tif (event.usage.input_tokens != null) {\n\t\t\t\t\t\toutput.usage.input = event.usage.input_tokens;\n\t\t\t\t\t}\n\t\t\t\t\tif (event.usage.output_tokens != null) {\n\t\t\t\t\t\toutput.usage.output = event.usage.output_tokens;\n\t\t\t\t\t}\n\t\t\t\t\tif (event.usage.cache_read_input_tokens != null) {\n\t\t\t\t\t\toutput.usage.cacheRead = event.usage.cache_read_input_tokens;\n\t\t\t\t\t}\n\t\t\t\t\tif (event.usage.cache_creation_input_tokens != null) {\n\t\t\t\t\t\toutput.usage.cacheWrite = event.usage.cache_creation_input_tokens;\n\t\t\t\t\t}\n\t\t\t\t\t// Anthropic doesn't provide total_tokens, compute from components\n\t\t\t\t\toutput.usage.totalTokens =\n\t\t\t\t\t\toutput.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;\n\t\t\t\t\tcalculateCost(model, output.usage);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) delete (block as any).index;\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\n/**\n * Map ThinkingLevel to Anthropic effort levels for adaptive thinking.\n * Note: effort \"max\" is only valid on models whose capability table\n * entry sets `xhighEffort: true` (today: Opus 4.6 / 4.7).\n */\nfunction mapThinkingLevelToEffort(\n\tlevel: SimpleStreamOptions[\"reasoning\"],\n\tmodel: Model<\"anthropic-messages\">,\n): AnthropicEffort {\n\tswitch (level) {\n\t\tcase \"minimal\":\n\t\t\treturn \"low\";\n\t\tcase \"low\":\n\t\t\treturn \"low\";\n\t\tcase \"medium\":\n\t\t\treturn \"medium\";\n\t\tcase \"high\":\n\t\t\treturn \"high\";\n\t\tcase \"xhigh\":\n\t\t\treturn getAnthropicCapabilities(model.id, model.provider).xhighEffort ? \"max\" : \"high\";\n\t\tdefault:\n\t\t\treturn \"high\";\n\t}\n}\n\nexport const streamSimpleAnthropic: StreamFunction<\"anthropic-messages\", SimpleStreamOptions> = (\n\tmodel: Model<\"anthropic-messages\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tif (!options?.reasoning) {\n\t\treturn streamAnthropic(model, context, { ...base, thinkingEnabled: false } satisfies AnthropicOptions);\n\t}\n\n\t// For Opus 4.6 and Sonnet 4.6: use adaptive thinking with effort level\n\t// For older models: use budget-based thinking\n\tif (supportsAdaptiveThinking(model.id, model.provider)) {\n\t\tconst effort = mapThinkingLevelToEffort(options.reasoning, model);\n\t\treturn streamAnthropic(model, context, {\n\t\t\t...base,\n\t\t\tthinkingEnabled: true,\n\t\t\teffort,\n\t\t} satisfies AnthropicOptions);\n\t}\n\n\tconst adjusted = adjustMaxTokensForThinking(\n\t\tbase.maxTokens || 0,\n\t\tmodel.maxTokens,\n\t\toptions.reasoning,\n\t\toptions.thinkingBudgets,\n\t);\n\n\treturn streamAnthropic(model, context, {\n\t\t...base,\n\t\tmaxTokens: adjusted.maxTokens,\n\t\tthinkingEnabled: true,\n\t\tthinkingBudgetTokens: adjusted.thinkingBudget,\n\t} satisfies AnthropicOptions);\n};\n\nfunction isOAuthToken(apiKey: string): boolean {\n\treturn apiKey.includes(\"sk-ant-oat\");\n}\n\nfunction createClient(\n\tmodel: Model<\"anthropic-messages\">,\n\tapiKey: string,\n\tinterleavedThinking: boolean,\n\toptionsHeaders?: Record<string, string>,\n\tdynamicHeaders?: Record<string, string>,\n): { client: Anthropic; isOAuthToken: boolean } {\n\t// Adaptive thinking models (Opus 4.6, Sonnet 4.6) have interleaved thinking built-in.\n\t// The beta header is deprecated on Opus 4.6 and redundant on Sonnet 4.6, so skip it.\n\tconst needsInterleavedBeta = interleavedThinking && !supportsAdaptiveThinking(model.id, model.provider);\n\n\t// Opt into per-model capability betas (e.g. 1M context window).\n\tconst caps = getAnthropicCapabilities(model.id, model.provider);\n\tconst extraBetas: string[] = [];\n\tif (caps.contextBeta) {\n\t\textraBetas.push(caps.contextBeta);\n\t}\n\n\t// Copilot: Bearer auth, selective betas (no fine-grained-tool-streaming)\n\tif (model.provider === \"github-copilot\") {\n\t\tconst betaFeatures: string[] = [];\n\t\tif (needsInterleavedBeta) {\n\t\t\tbetaFeatures.push(\"interleaved-thinking-2025-05-14\");\n\t\t}\n\t\tbetaFeatures.push(...extraBetas);\n\n\t\tconst client = new Anthropic({\n\t\t\tapiKey: null,\n\t\t\tauthToken: apiKey,\n\t\t\tbaseURL: model.baseUrl,\n\t\t\tdangerouslyAllowBrowser: true,\n\t\t\tdefaultHeaders: mergeHeaders(\n\t\t\t\t{\n\t\t\t\t\taccept: \"application/json\",\n\t\t\t\t\t\"anthropic-dangerous-direct-browser-access\": \"true\",\n\t\t\t\t\t...(betaFeatures.length > 0 ? { \"anthropic-beta\": betaFeatures.join(\",\") } : {}),\n\t\t\t\t},\n\t\t\t\tmodel.headers,\n\t\t\t\tdynamicHeaders,\n\t\t\t\toptionsHeaders,\n\t\t\t),\n\t\t});\n\n\t\treturn { client, isOAuthToken: false };\n\t}\n\n\tconst betaFeatures = [\"fine-grained-tool-streaming-2025-05-14\"];\n\tif (needsInterleavedBeta) {\n\t\tbetaFeatures.push(\"interleaved-thinking-2025-05-14\");\n\t}\n\tbetaFeatures.push(...extraBetas);\n\n\t// OAuth: Bearer auth, Claude Code identity headers\n\tif (isOAuthToken(apiKey)) {\n\t\tconst client = new Anthropic({\n\t\t\tapiKey: null,\n\t\t\tauthToken: apiKey,\n\t\t\tbaseURL: model.baseUrl,\n\t\t\tdangerouslyAllowBrowser: true,\n\t\t\tdefaultHeaders: mergeHeaders(\n\t\t\t\t{\n\t\t\t\t\taccept: \"application/json\",\n\t\t\t\t\t\"anthropic-dangerous-direct-browser-access\": \"true\",\n\t\t\t\t\t\"anthropic-beta\": `claude-code-20250219,oauth-2025-04-20,${betaFeatures.join(\",\")}`,\n\t\t\t\t\t\"user-agent\": `claude-cli/${claudeCodeVersion}`,\n\t\t\t\t\t\"x-app\": \"cli\",\n\t\t\t\t},\n\t\t\t\tmodel.headers,\n\t\t\t\toptionsHeaders,\n\t\t\t),\n\t\t});\n\n\t\treturn { client, isOAuthToken: true };\n\t}\n\n\t// API key auth\n\tconst client = new Anthropic({\n\t\tapiKey,\n\t\tbaseURL: model.baseUrl,\n\t\tdangerouslyAllowBrowser: true,\n\t\tdefaultHeaders: mergeHeaders(\n\t\t\t{\n\t\t\t\taccept: \"application/json\",\n\t\t\t\t\"anthropic-dangerous-direct-browser-access\": \"true\",\n\t\t\t\t\"anthropic-beta\": betaFeatures.join(\",\"),\n\t\t\t},\n\t\t\tmodel.headers,\n\t\t\toptionsHeaders,\n\t\t),\n\t});\n\n\treturn { client, isOAuthToken: false };\n}\n\nfunction buildParams(\n\tmodel: Model<\"anthropic-messages\">,\n\tcontext: Context,\n\tisOAuthToken: boolean,\n\toptions?: AnthropicOptions,\n): MessageCreateParamsStreaming {\n\tconst { cacheControl } = getCacheControl(model.baseUrl, options?.cacheRetention);\n\tconst params: MessageCreateParamsStreaming = {\n\t\tmodel: model.id,\n\t\tmessages: convertMessages(context.messages, model, isOAuthToken, cacheControl),\n\t\tmax_tokens: options?.maxTokens || (model.maxTokens / 3) | 0,\n\t\tstream: true,\n\t};\n\n\t// For OAuth tokens, we MUST include Claude Code identity\n\tif (isOAuthToken) {\n\t\tparams.system = [\n\t\t\t{\n\t\t\t\ttype: \"text\",\n\t\t\t\ttext: \"You are Claude Code, Anthropic's official CLI for Claude.\",\n\t\t\t\t...(cacheControl ? { cache_control: cacheControl } : {}),\n\t\t\t},\n\t\t];\n\t\tif (context.systemPrompt) {\n\t\t\tparams.system.push({\n\t\t\t\ttype: \"text\",\n\t\t\t\ttext: sanitizeSurrogates(context.systemPrompt),\n\t\t\t\t...(cacheControl ? { cache_control: cacheControl } : {}),\n\t\t\t});\n\t\t}\n\t} else if (context.systemPrompt) {\n\t\t// Add cache control to system prompt for non-OAuth tokens\n\t\tparams.system = [\n\t\t\t{\n\t\t\t\ttype: \"text\",\n\t\t\t\ttext: sanitizeSurrogates(context.systemPrompt),\n\t\t\t\t...(cacheControl ? { cache_control: cacheControl } : {}),\n\t\t\t},\n\t\t];\n\t}\n\n\t// Temperature is incompatible with extended thinking (adaptive or budget-based).\n\tif (options?.temperature !== undefined && !options?.thinkingEnabled) {\n\t\tparams.temperature = options.temperature;\n\t}\n\n\tif (context.tools) {\n\t\tparams.tools = convertTools(context.tools, isOAuthToken);\n\t\t// Experimental, env-gated (#42 cache probe): anchor a cache breakpoint on the\n\t\t// last tool so [system+tools] is a named stable segment, independent of the\n\t\t// rolling-tail breakpoint. Default OFF — no behavior change. Used only to\n\t\t// measure whether this beats Anthropic's automatic prefix-matching.\n\t\tif (cacheControl && process.env.CAVE_TOOLS_CACHE_BREAKPOINT === \"1\" && params.tools.length > 0) {\n\t\t\t(params.tools[params.tools.length - 1] as { cache_control?: typeof cacheControl }).cache_control =\n\t\t\t\tcacheControl;\n\t\t}\n\t}\n\n\t// Configure thinking mode: adaptive (Opus 4.6 and Sonnet 4.6),\n\t// budget-based (older models), or explicitly disabled.\n\tif (model.reasoning) {\n\t\tif (options?.thinkingEnabled) {\n\t\t\tif (supportsAdaptiveThinking(model.id, model.provider)) {\n\t\t\t\t// Adaptive thinking: Claude decides when and how much to think\n\t\t\t\tparams.thinking = { type: \"adaptive\" };\n\t\t\t\tif (options.effort) {\n\t\t\t\t\tparams.output_config = { effort: options.effort };\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\t// Budget-based thinking for older models\n\t\t\t\tparams.thinking = {\n\t\t\t\t\ttype: \"enabled\",\n\t\t\t\t\tbudget_tokens: options.thinkingBudgetTokens || 1024,\n\t\t\t\t};\n\t\t\t}\n\t\t} else if (options?.thinkingEnabled === false) {\n\t\t\tparams.thinking = { type: \"disabled\" };\n\t\t}\n\t}\n\n\tif (options?.metadata) {\n\t\tconst userId = options.metadata.user_id;\n\t\tif (typeof userId === \"string\") {\n\t\t\tparams.metadata = { user_id: userId };\n\t\t}\n\t}\n\n\tif (options?.toolChoice) {\n\t\tif (typeof options.toolChoice === \"string\") {\n\t\t\tparams.tool_choice = { type: options.toolChoice };\n\t\t} else {\n\t\t\tparams.tool_choice = options.toolChoice;\n\t\t}\n\t}\n\n\treturn params;\n}\n\n// Normalize tool call IDs to match Anthropic's required pattern and length\nfunction normalizeToolCallId(id: string): string {\n\treturn id.replace(/[^a-zA-Z0-9_-]/g, \"_\").slice(0, 64);\n}\n\nfunction convertMessages(\n\tmessages: Message[],\n\tmodel: Model<\"anthropic-messages\">,\n\tisOAuthToken: boolean,\n\tcacheControl?: { type: \"ephemeral\"; ttl?: \"1h\" },\n): MessageParam[] {\n\tconst params: MessageParam[] = [];\n\n\t// Transform messages for cross-provider compatibility\n\tconst transformedMessages = transformMessages(messages, model, normalizeToolCallId);\n\n\tfor (let i = 0; i < transformedMessages.length; i++) {\n\t\tconst msg = transformedMessages[i];\n\n\t\tif (msg.role === \"user\") {\n\t\t\tif (typeof msg.content === \"string\") {\n\t\t\t\tif (msg.content.trim().length > 0) {\n\t\t\t\t\tparams.push({\n\t\t\t\t\t\trole: \"user\",\n\t\t\t\t\t\tcontent: sanitizeSurrogates(msg.content),\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tconst blocks: ContentBlockParam[] = msg.content.map((item) => {\n\t\t\t\t\tif (item.type === \"text\") {\n\t\t\t\t\t\treturn {\n\t\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\t\ttext: sanitizeSurrogates(item.text),\n\t\t\t\t\t\t};\n\t\t\t\t\t} else {\n\t\t\t\t\t\treturn {\n\t\t\t\t\t\t\ttype: \"image\",\n\t\t\t\t\t\t\tsource: {\n\t\t\t\t\t\t\t\ttype: \"base64\",\n\t\t\t\t\t\t\t\tmedia_type: item.mimeType as \"image/jpeg\" | \"image/png\" | \"image/gif\" | \"image/webp\",\n\t\t\t\t\t\t\t\tdata: item.data,\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t};\n\t\t\t\t\t}\n\t\t\t\t});\n\t\t\t\tlet filteredBlocks = !model?.input.includes(\"image\") ? blocks.filter((b) => b.type !== \"image\") : blocks;\n\t\t\t\tfilteredBlocks = filteredBlocks.filter((b) => {\n\t\t\t\t\tif (b.type === \"text\") {\n\t\t\t\t\t\treturn b.text.trim().length > 0;\n\t\t\t\t\t}\n\t\t\t\t\treturn true;\n\t\t\t\t});\n\t\t\t\tif (filteredBlocks.length === 0) continue;\n\t\t\t\tparams.push({\n\t\t\t\t\trole: \"user\",\n\t\t\t\t\tcontent: filteredBlocks,\n\t\t\t\t});\n\t\t\t}\n\t\t} else if (msg.role === \"assistant\") {\n\t\t\tconst blocks: ContentBlockParam[] = [];\n\n\t\t\tfor (const block of msg.content) {\n\t\t\t\tif (block.type === \"text\") {\n\t\t\t\t\tif (block.text.trim().length === 0) continue;\n\t\t\t\t\tblocks.push({\n\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\ttext: sanitizeSurrogates(block.text),\n\t\t\t\t\t});\n\t\t\t\t} else if (block.type === \"thinking\") {\n\t\t\t\t\t// Redacted thinking: pass the opaque payload back as redacted_thinking\n\t\t\t\t\tif (block.redacted) {\n\t\t\t\t\t\tblocks.push({\n\t\t\t\t\t\t\ttype: \"redacted_thinking\",\n\t\t\t\t\t\t\tdata: block.thinkingSignature!,\n\t\t\t\t\t\t});\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\t\t\t\t\tif (block.thinking.trim().length === 0) continue;\n\t\t\t\t\t// If thinking signature is missing/empty (e.g., from aborted stream),\n\t\t\t\t\t// convert to plain text block without <thinking> tags to avoid API rejection\n\t\t\t\t\t// and prevent Claude from mimicking the tags in responses\n\t\t\t\t\tif (!block.thinkingSignature || block.thinkingSignature.trim().length === 0) {\n\t\t\t\t\t\tblocks.push({\n\t\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\t\ttext: sanitizeSurrogates(block.thinking),\n\t\t\t\t\t\t});\n\t\t\t\t\t} else {\n\t\t\t\t\t\tblocks.push({\n\t\t\t\t\t\t\ttype: \"thinking\",\n\t\t\t\t\t\t\tthinking: sanitizeSurrogates(block.thinking),\n\t\t\t\t\t\t\tsignature: block.thinkingSignature,\n\t\t\t\t\t\t});\n\t\t\t\t\t}\n\t\t\t\t} else if (block.type === \"toolCall\") {\n\t\t\t\t\tblocks.push({\n\t\t\t\t\t\ttype: \"tool_use\",\n\t\t\t\t\t\tid: block.id,\n\t\t\t\t\t\tname: isOAuthToken ? toClaudeCodeName(block.name) : block.name,\n\t\t\t\t\t\tinput: block.arguments ?? {},\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (blocks.length === 0) continue;\n\t\t\tparams.push({\n\t\t\t\trole: \"assistant\",\n\t\t\t\tcontent: blocks,\n\t\t\t});\n\t\t} else if (msg.role === \"toolResult\") {\n\t\t\t// Collect all consecutive toolResult messages, needed for z.ai Anthropic endpoint\n\t\t\tconst toolResults: ContentBlockParam[] = [];\n\n\t\t\t// Add the current tool result\n\t\t\ttoolResults.push({\n\t\t\t\ttype: \"tool_result\",\n\t\t\t\ttool_use_id: msg.toolCallId,\n\t\t\t\tcontent: convertContentBlocks(msg.content),\n\t\t\t\tis_error: msg.isError,\n\t\t\t});\n\n\t\t\t// Look ahead for consecutive toolResult messages\n\t\t\tlet j = i + 1;\n\t\t\twhile (j < transformedMessages.length && transformedMessages[j].role === \"toolResult\") {\n\t\t\t\tconst nextMsg = transformedMessages[j] as ToolResultMessage; // We know it's a toolResult\n\t\t\t\ttoolResults.push({\n\t\t\t\t\ttype: \"tool_result\",\n\t\t\t\t\ttool_use_id: nextMsg.toolCallId,\n\t\t\t\t\tcontent: convertContentBlocks(nextMsg.content),\n\t\t\t\t\tis_error: nextMsg.isError,\n\t\t\t\t});\n\t\t\t\tj++;\n\t\t\t}\n\n\t\t\t// Skip the messages we've already processed\n\t\t\ti = j - 1;\n\n\t\t\t// Add a single user message with all tool results\n\t\t\tparams.push({\n\t\t\t\trole: \"user\",\n\t\t\t\tcontent: toolResults,\n\t\t\t});\n\t\t}\n\t}\n\n\t// Add cache_control to the last user message to cache conversation history\n\tif (cacheControl && params.length > 0) {\n\t\tconst lastMessage = params[params.length - 1];\n\t\tif (lastMessage.role === \"user\") {\n\t\t\tif (Array.isArray(lastMessage.content)) {\n\t\t\t\tconst lastBlock = lastMessage.content[lastMessage.content.length - 1];\n\t\t\t\tif (\n\t\t\t\t\tlastBlock &&\n\t\t\t\t\t(lastBlock.type === \"text\" || lastBlock.type === \"image\" || lastBlock.type === \"tool_result\")\n\t\t\t\t) {\n\t\t\t\t\t(lastBlock as any).cache_control = cacheControl;\n\t\t\t\t}\n\t\t\t} else if (typeof lastMessage.content === \"string\") {\n\t\t\t\tlastMessage.content = [\n\t\t\t\t\t{\n\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\ttext: lastMessage.content,\n\t\t\t\t\t\tcache_control: cacheControl,\n\t\t\t\t\t},\n\t\t\t\t] as any;\n\t\t\t}\n\t\t}\n\t}\n\n\treturn params;\n}\n\nfunction convertTools(tools: Tool[], isOAuthToken: boolean): Anthropic.Messages.Tool[] {\n\tif (!tools) return [];\n\n\treturn tools.map((tool) => {\n\t\tconst jsonSchema = tool.parameters as any; // TypeBox already generates JSON Schema\n\n\t\treturn {\n\t\t\tname: isOAuthToken ? toClaudeCodeName(tool.name) : tool.name,\n\t\t\tdescription: tool.description,\n\t\t\tinput_schema: {\n\t\t\t\ttype: \"object\" as const,\n\t\t\t\tproperties: jsonSchema.properties || {},\n\t\t\t\trequired: jsonSchema.required || [],\n\t\t\t},\n\t\t};\n\t});\n}\n\nfunction mapStopReason(reason: Anthropic.Messages.StopReason | string): StopReason {\n\tswitch (reason) {\n\t\tcase \"end_turn\":\n\t\t\treturn \"stop\";\n\t\tcase \"max_tokens\":\n\t\t\treturn \"length\";\n\t\tcase \"tool_use\":\n\t\t\treturn \"toolUse\";\n\t\tcase \"refusal\":\n\t\t\treturn \"error\";\n\t\tcase \"pause_turn\": // Stop is good enough -> resubmit\n\t\t\treturn \"stop\";\n\t\tcase \"stop_sequence\":\n\t\t\treturn \"stop\"; // We don't supply stop sequences, so this should never happen\n\t\tcase \"sensitive\": // Content flagged by safety filters (not yet in SDK types)\n\t\t\treturn \"error\";\n\t\tdefault:\n\t\t\t// Handle unknown stop reasons gracefully (API may add new values)\n\t\t\tthrow new Error(`Unhandled stop reason: ${reason}`);\n\t}\n}\n"]}