vidspotai-shared 1.0.82 → 1.0.84

package/lib/services/agent/extendChain.js

@@ -0,0 +1,177 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.planExtendChain = planExtendChain;
+exports.resolveExtendChain = resolveExtendChain;
+const logger_1 = require("../../utils/logger");
+const tierHelpers_1 = require("../../globals/aiModels/tierHelpers");
+const extendVideo_tool_1 = require("./tools/extendVideo.tool");
+/**
+ * AG-2 / Stream L (L-4) — extend-chain orchestration.
+ *
+ * A single scene can ask for more seconds than the chosen video model emits in
+ * one clip (e.g. a 16s continuous establishing shot on Kling, whose max single
+ * clip is ~10s). For extend-capable providers (Kling / Pixverse) we don't drop
+ * the extra seconds or fall back to a different provider (which would visibly
+ * change the look) — we generate the initial clip at the model's max legal
+ * length, then call the provider's NATIVE extension endpoint N times, each
+ * continuing the PREVIOUS provider task.
+ *
+ * The hard part is ordering: `extend_video` takes a `previousTaskId` that must
+ * already be COMPLETE on the provider before the extension can start. The agent
+ * executor is otherwise fire-and-forget (it triggers every scene's generation
+ * in parallel and returns `pendingTaskId`s for the host to poll afterwards), so
+ * the chain can't live in that parallel fan-out.
+ *
+ * Split of responsibilities:
+ *   - `planExtendChain` (pure)        — decides whether a scene needs a chain
+ *                                       and how to slice the extra seconds. The
+ *                                       executor calls this at trigger time and
+ *                                       emits the segments on the SceneOutcome.
+ *   - `resolveExtendChain` (injected) — runs the chain SEQUENTIALLY: poll the
+ *                                       source → fire one extend → poll it →
+ *                                       repeat. The host injects `pollTask`
+ *                                       (a TaskPoller wrapper) + `runTool`.
+ *
+ * Keeping both here (in `shared`) rather than in the node worker means the
+ * decision logic is unit-testable and exercised by the eval harness's executor.
+ */
+/** Provider extension request bounds (mirror extendVideo.tool's InputSchema). */
+const EXTEND_MIN_SEC = 2;
+const EXTEND_MAX_SEC = 10;
+/** Preferred per-extension increment (Kling ~4-5s, Pixverse ~5-8s natively). */
+const PREFERRED_STEP_SEC = 5;
+/**
+ * Hard cap on extensions per scene. A mis-planned 60s "scene" should not fan
+ * into a dozen paid provider calls — beyond this we get as close as we can and
+ * log the shortfall. maxClip + MAX_EXTENSIONS×EXTEND_MAX covers ~70s, well past
+ * any sane single continuous shot.
+ */
+const MAX_EXTENSIONS = 6;
+/** Don't chain for a sub-second overflow — snapping absorbs it. */
+const CHAIN_TOLERANCE_SEC = 0.5;
+/**
+ * Decide whether a video scene needs an extend chain, and slice the extra
+ * seconds into provider-legal extension steps. Returns `null` when:
+ *   - the model has no declared duration constraint (unconstrained → one clip),
+ *   - the target already fits one clip (within snap tolerance), or
+ *   - the model has no native extension endpoint (caller clamps as before — a
+ *     non-extend model can't continue a clip, so the extra seconds are dropped
+ *     by the existing capability/snap path).
+ *
+ * `targetDurationSec` is the scene's full requested length (scene.durationMs/1000).
+ */
+function planExtendChain(modelKey, targetDurationSec) {
+    const allowed = (0, tierHelpers_1.getModelAllowedDurations)(modelKey);
+    if (!allowed || allowed.length === 0)
+        return null; // unconstrained
+    const maxClip = allowed[allowed.length - 1]; // getModelAllowedDurations sorts ascending
+    const remaining = targetDurationSec - maxClip;
+    if (remaining <= CHAIN_TOLERANCE_SEC)
+        return null; // fits one clip
+    if (!(0, extendVideo_tool_1.modelSupportsExtend)(modelKey))
+        return null; // can't extend on this model
+    let n = Math.ceil(remaining / PREFERRED_STEP_SEC);
+    n = Math.min(Math.max(n, 1), MAX_EXTENSIONS);
+    const per = clamp(Math.round(remaining / n), EXTEND_MIN_SEC, EXTEND_MAX_SEC);
+    const segments = Array.from({ length: n }, () => ({
+        extraDurationSec: per,
+    }));
+    const achievable = maxClip + n * per;
+    if (achievable + CHAIN_TOLERANCE_SEC < targetDurationSec) {
+        logger_1.logger.warn("extendChain: requested duration exceeds chain ceiling — clamping", {
+            modelKey,
+            targetDurationSec,
+            maxClip,
+            extensions: n,
+            perSec: per,
+            achievableSec: achievable,
+        });
+    }
+    return { initialDurationSec: maxClip, segments };
+}
+/**
+ * Run an extend chain to completion, enforcing source-task ordering: poll the
+ * current task → on completion, fire the next extension → poll that → repeat.
+ * Returns the source + extension clips in render order.
+ *
+ * Failure is partial-tolerant: if a mid-chain task fails/times out we stop and
+ * return the clips resolved so far with ok:false, so the caller can decide to
+ * ship a shorter scene rather than lose the whole thing. A failed SOURCE clip
+ * (no clips resolved) is a hard scene failure.
+ */
+async function resolveExtendChain(deps, args) {
+    const { runTool, pollTask } = deps;
+    const { sceneIndex, modelKey, segments, ctx } = args;
+    const prefix = args.outputFilenamePrefix ?? `scene${sceneIndex}`;
+    const clips = [];
+    let currentTaskId = args.initialTaskId;
+    let currentDurationSec = args.initialDurationSec;
+    // Step 0 is the source clip; steps 1..N are the extensions. We poll the
+    // current task, record its clip, then (if more remain) fire the next extend.
+    for (let step = 0; step <= segments.length; step++) {
+        const poll = await pollTask({
+            ref: `${sceneIndex}.${step}`,
+            taskId: currentTaskId,
+            modelKey,
+            outputFilename: `${prefix}-ext${step}`,
+        });
+        if (poll.status !== "completed" || !poll.videoUrl) {
+            logger_1.logger.warn("extendChain: chain task did not complete — stopping", {
+                sceneIndex,
+                step,
+                status: poll.status,
+                taskId: currentTaskId,
+                error: poll.errorMessage,
+            });
+            return {
+                ok: false,
+                clips,
+                error: {
+                    code: step === 0 ? "EXTEND_SOURCE_FAILED" : "EXTEND_STEP_FAILED",
+                    message: `Extend chain ${step === 0 ? "source" : `step ${step}`} ${poll.status}` +
+                        (poll.errorMessage ? `: ${poll.errorMessage}` : ""),
+                    failedTaskId: currentTaskId,
+                },
+            };
+        }
+        clips.push({
+            visualUrl: poll.videoUrl,
+            taskId: currentTaskId,
+            durationSec: currentDurationSec,
+        });
+        // All steps resolved — done.
+        if (step === segments.length)
+            break;
+        // Fire the next extension, continuing the task we just confirmed complete.
+        const seg = segments[step];
+        const extendOut = await runTool("extend_video", {
+            previousTaskId: currentTaskId,
+            modelKey,
+            extraDurationSec: seg.extraDurationSec,
+            ...(seg.prompt ? { prompt: seg.prompt } : {}),
+        }, { ...ctx, idempotencyKey: `${ctx.idempotencyKey ?? ctx.agentRunId}:ext${step + 1}` });
+        if (!extendOut.ok) {
+            logger_1.logger.warn("extendChain: extend_video call failed — stopping", {
+                sceneIndex,
+                step: step + 1,
+                sourceTaskId: currentTaskId,
+                error: extendOut.error,
+            });
+            return {
+                ok: false,
+                clips,
+                error: {
+                    code: "EXTEND_TRIGGER_FAILED",
+                    message: `extend_video step ${step + 1} failed: ${extendOut.error?.message ?? extendOut.error?.code ?? "unknown"}`,
+                    failedTaskId: currentTaskId,
+                },
+            };
+        }
+        currentTaskId = extendOut.output.taskId;
+        currentDurationSec = seg.extraDurationSec;
+    }
+    return { ok: true, clips };
+}
+function clamp(v, lo, hi) {
+    return Math.max(lo, Math.min(hi, v));
+}

package/lib/services/agent/index.d.ts

@@ -2,13 +2,19 @@ export * from "./toolRegistry";
 export * from "./tools";
 export * from "./modelRouter";
 export * from "./llmCaller";
-export * from "./llmCallerGateway";
+export * from "./llmRetry";
+export * from "./llmModelRegistry";
 export * from "./llmCallerAnthropic";
+export * from "./llmCallerOpenai";
+export * from "./llmCallerGemini";
+export * from "./llmCallerRouting";
+export * from "./llmCallerFactory";
 export * from "./personas";
 export * from "./llmCallerRegistry";
 export * from "./costPreflight";
 export * from "./planner";
 export * from "./planMutations";
+export * from "./priorProject";
 export * from "./bibleBuilder";
 export * from "./referenceImageRenderer";
 export * from "./bibleImageVision";
@@ -16,6 +22,10 @@ export * from "./sceneLayoutVision";
 export * from "./beatSnap";
 export * from "./musicSelect";
 export * from "./executor";
+export * from "./extendChain";
+export * from "./editExisting";
+export * from "./repurpose";
+export * from "./covers";
 export * from "./providerTaskCache";
 export * from "./planDiff";
 export * from "./taskPoller";

package/lib/services/agent/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/services/agent/index.ts"],"names":[],"mappings":"AAAA,cAAc,gBAAgB,CAAC;AAC/B,cAAc,SAAS,CAAC;AACxB,cAAc,eAAe,CAAC;AAC9B,cAAc,aAAa,CAAC;AAC5B,cAAc,oBAAoB,CAAC;AACnC,cAAc,sBAAsB,CAAC;AACrC,cAAc,YAAY,CAAC;AAC3B,cAAc,qBAAqB,CAAC;AACpC,cAAc,iBAAiB,CAAC;AAChC,cAAc,WAAW,CAAC;AAC1B,cAAc,iBAAiB,CAAC;AAChC,cAAc,gBAAgB,CAAC;AAC/B,cAAc,0BAA0B,CAAC;AACzC,cAAc,oBAAoB,CAAC;AACnC,cAAc,qBAAqB,CAAC;AACpC,cAAc,YAAY,CAAC;AAC3B,cAAc,eAAe,CAAC;AAC9B,cAAc,YAAY,CAAC;AAC3B,cAAc,qBAAqB,CAAC;AACpC,cAAc,YAAY,CAAC;AAC3B,cAAc,cAAc,CAAC;AAC7B,cAAc,UAAU,CAAC;AACzB,cAAc,QAAQ,CAAC;AACvB,cAAc,2BAA2B,CAAC;AAC1C,cAAc,uBAAuB,CAAC;AACtC,cAAc,8BAA8B,CAAC;AAC7C,cAAc,kBAAkB,CAAC;AACjC,cAAc,qBAAqB,CAAC;AACpC,cAAc,mBAAmB,CAAC;AAClC,cAAc,kBAAkB,CAAC;AACjC,cAAc,eAAe,CAAC;AAC9B,cAAc,eAAe,CAAC;AAC9B,cAAc,qBAAqB,CAAC;AACpC,cAAc,cAAc,CAAC;AAC7B,cAAc,aAAa,CAAC;AAC5B,cAAc,iBAAiB,CAAC;AAChC,cAAc,aAAa,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/services/agent/index.ts"],"names":[],"mappings":"AAAA,cAAc,gBAAgB,CAAC;AAC/B,cAAc,SAAS,CAAC;AACxB,cAAc,eAAe,CAAC;AAC9B,cAAc,aAAa,CAAC;AAC5B,cAAc,YAAY,CAAC;AAC3B,cAAc,oBAAoB,CAAC;AACnC,cAAc,sBAAsB,CAAC;AACrC,cAAc,mBAAmB,CAAC;AAClC,cAAc,mBAAmB,CAAC;AAClC,cAAc,oBAAoB,CAAC;AACnC,cAAc,oBAAoB,CAAC;AACnC,cAAc,YAAY,CAAC;AAC3B,cAAc,qBAAqB,CAAC;AACpC,cAAc,iBAAiB,CAAC;AAChC,cAAc,WAAW,CAAC;AAC1B,cAAc,iBAAiB,CAAC;AAChC,cAAc,gBAAgB,CAAC;AAC/B,cAAc,gBAAgB,CAAC;AAC/B,cAAc,0BAA0B,CAAC;AACzC,cAAc,oBAAoB,CAAC;AACnC,cAAc,qBAAqB,CAAC;AACpC,cAAc,YAAY,CAAC;AAC3B,cAAc,eAAe,CAAC;AAC9B,cAAc,YAAY,CAAC;AAC3B,cAAc,eAAe,CAAC;AAC9B,cAAc,gBAAgB,CAAC;AAC/B,cAAc,aAAa,CAAC;AAC5B,cAAc,UAAU,CAAC;AACzB,cAAc,qBAAqB,CAAC;AACpC,cAAc,YAAY,CAAC;AAC3B,cAAc,cAAc,CAAC;AAC7B,cAAc,UAAU,CAAC;AACzB,cAAc,QAAQ,CAAC;AACvB,cAAc,2BAA2B,CAAC;AAC1C,cAAc,uBAAuB,CAAC;AACtC,cAAc,8BAA8B,CAAC;AAC7C,cAAc,kBAAkB,CAAC;AACjC,cAAc,qBAAqB,CAAC;AACpC,cAAc,mBAAmB,CAAC;AAClC,cAAc,kBAAkB,CAAC;AACjC,cAAc,eAAe,CAAC;AAC9B,cAAc,eAAe,CAAC;AAC9B,cAAc,qBAAqB,CAAC;AACpC,cAAc,cAAc,CAAC;AAC7B,cAAc,aAAa,CAAC;AAC5B,cAAc,iBAAiB,CAAC;AAChC,cAAc,aAAa,CAAC"}

package/lib/services/agent/index.js

@@ -18,13 +18,19 @@ __exportStar(require("./toolRegistry"), exports);
 __exportStar(require("./tools"), exports);
 __exportStar(require("./modelRouter"), exports);
 __exportStar(require("./llmCaller"), exports);
-__exportStar(require("./llmCallerGateway"), exports);
+__exportStar(require("./llmRetry"), exports);
+__exportStar(require("./llmModelRegistry"), exports);
 __exportStar(require("./llmCallerAnthropic"), exports);
+__exportStar(require("./llmCallerOpenai"), exports);
+__exportStar(require("./llmCallerGemini"), exports);
+__exportStar(require("./llmCallerRouting"), exports);
+__exportStar(require("./llmCallerFactory"), exports);
 __exportStar(require("./personas"), exports);
 __exportStar(require("./llmCallerRegistry"), exports);
 __exportStar(require("./costPreflight"), exports);
 __exportStar(require("./planner"), exports);
 __exportStar(require("./planMutations"), exports);
+__exportStar(require("./priorProject"), exports);
 __exportStar(require("./bibleBuilder"), exports);
 __exportStar(require("./referenceImageRenderer"), exports);
 __exportStar(require("./bibleImageVision"), exports);
@@ -32,6 +38,10 @@ __exportStar(require("./sceneLayoutVision"), exports);
 __exportStar(require("./beatSnap"), exports);
 __exportStar(require("./musicSelect"), exports);
 __exportStar(require("./executor"), exports);
+__exportStar(require("./extendChain"), exports);
+__exportStar(require("./editExisting"), exports);
+__exportStar(require("./repurpose"), exports);
+__exportStar(require("./covers"), exports);
 __exportStar(require("./providerTaskCache"), exports);
 __exportStar(require("./planDiff"), exports);
 __exportStar(require("./taskPoller"), exports);

package/lib/services/agent/llmCaller.d.ts

@@ -2,9 +2,10 @@ import { z } from "zod";
 import { ModelChoice } from "./modelRouter";
 /**
  * Provider-agnostic LLM caller. The host (Express server / Firebase Functions)
- * wires a concrete implementation backed by Vercel AI Gateway, OpenRouter, or
- * direct SDKs. Agent code only ever sees this interface, which keeps it
- * portable and lets us A/B-test models against the eval set without code changes.
+ * wires a concrete implementation — the native ProviderRoutingLlmCaller, which
+ * dispatches per request to a provider's own SDK (Anthropic / OpenAI / Gemini).
+ * Agent code only ever sees this interface, which keeps it portable and lets us
+ * A/B-test models against the eval set without code changes.
  */
 export interface LlmCaller {
     /** Free-form chat completion. */
@@ -30,11 +31,9 @@ export interface LlmCaller {
      * emitting the final structured payload.
      *
      * Only implemented by callers that natively support tool-use loops (today:
-     * `AnthropicLlmCaller`). `GatewayLlmCaller` does NOT implement this because
-     * Anthropic's built-in server tools are not exposable through the OpenAI-
-     * compatible Vercel AI Gateway. If a future need warrants it, we can add
-     * a custom-tools-only variant on the gateway (e.g. Tavily passed as a
-     * regular OpenAI-format tool) — see deferred work in `llmCallerAnthropic.ts`.
+     * `AnthropicLlmCaller`, for Anthropic's built-in server tools). The
+     * dispatcher routes structuredWithTools to the Anthropic caller and throws
+     * if a non-Anthropic provider is asked for it. OpenAI/Gemini callers omit it.
      */
     structuredWithTools?<T extends z.ZodTypeAny>(req: StructuredWithToolsRequest<T>): Promise<StructuredResponse<T>>;
 }

package/lib/services/agent/llmCaller.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"llmCaller.d.ts","sourceRoot":"","sources":["../../../src/services/agent/llmCaller.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,WAAW,EAAE,MAAM,eAAe,CAAC;AAE5C;;;;;GAKG;AACH,MAAM,WAAW,SAAS;IACxB,iCAAiC;IACjC,IAAI,CAAC,GAAG,EAAE,WAAW,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;IAE9C,yFAAyF;IACzF,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,EAC/B,GAAG,EAAE,iBAAiB,CAAC,CAAC,CAAC,GACxB,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC,CAAC;IAElC;;;;;;;;OAQG;IACH,gBAAgB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,EACtC,GAAG,EAAE,iBAAiB,CAAC,CAAC,CAAC,GACxB;QACD,MAAM,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC;QAC9B,MAAM,EAAE,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC,CAAC;KACxC,CAAC;IAEF;;;;;;;;;;;OAWG;IACH,mBAAmB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,EACzC,GAAG,EAAE,0BAA0B,CAAC,CAAC,CAAC,GACjC,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC,CAAC;CACnC;AAED,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,CAAC;IACtC,OAAO,EAAE,MAAM,CAAC;IAChB,wDAAwD;IACxD,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;CACtB;AAED,MAAM,WAAW,WAAW;IAC1B,KAAK,EAAE,WAAW,CAAC;IACnB,QAAQ,EAAE,WAAW,EAAE,CAAC;IACxB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,sBAAsB;IACtB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB;;;;OAIG;IACH,aAAa,CAAC,EAAE,WAAW,CAAC;CAC7B;AAED,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,kDAAkD;IAClD,KAAK,CAAC,EAAE;QAAE,YAAY,CAAC,EAAE,MAAM,CAAC;QAAC,gBAAgB,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAC7D,+CAA+C;IAC/C,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iFAAiF;IACjF,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,iEAAiE;IACjE,eAAe,CAAC,EAAE,WAAW,EAAE,CAAC;CACjC;AAED,MAAM,WAAW,iBAAiB,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU;IACvD,KAAK,EAAE,WAAW,CAAC;IACnB,QAAQ,EAAE,WAAW,EAAE,CAAC;IACxB,MAAM,EAAE,CAAC,CAAC;IACV,mEAAmE;IACnE,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,qCAAqC;IACrC,aAAa,CAAC,EAAE,WAAW,CAAC;CAC7B;AAED,MAAM,WAAW,0BAA0B,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,CAChE,SAAQ,iBAAiB,CAAC,CAAC,CAAC;IAC5B;;;;;OAKG;IACH,WAAW,CAAC,EAAE,KAAK,CAAC;QAClB,IAAI,EAAE,YAAY,CAAC;QACnB,6DAA6D;QAC7D,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,6CAA6C;QAC7C,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;QAC1B,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;KAC3B,CAAC,CAAC;IACH;;;;OAIG;IACH,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB;;;;;OAKG;IACH,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB;;;;;OAKG;IACH,QAAQ,CAAC,EAAE;QAAE,YAAY,EAAE,MAAM,CAAA;KAAE,CAAC;CACrC;AAED,MAAM,WAAW,kBAAkB,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU;IACxD,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IACjB,KAAK,CAAC,EAAE;QAAE,YAAY,CAAC,EAAE,MAAM,CAAC;QAAC,gBAAgB,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAC7D,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iFAAiF;IACjF,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,iEAAiE;IACjE,eAAe,CAAC,EAAE,WAAW,EAAE,CAAC;CACjC;AAED;;;GAGG;AACH,qBAAa,qBAAsB,YAAW,SAAS;IACrD,IAAI,IAAI,OAAO,CAAC,YAAY,CAAC;IAK7B,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,KAAK,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC;CAKrE"}
+{"version":3,"file":"llmCaller.d.ts","sourceRoot":"","sources":["../../../src/services/agent/llmCaller.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,WAAW,EAAE,MAAM,eAAe,CAAC;AAE5C;;;;;;GAMG;AACH,MAAM,WAAW,SAAS;IACxB,iCAAiC;IACjC,IAAI,CAAC,GAAG,EAAE,WAAW,GAAG,OAAO,CAAC,YAAY,CAAC,CAAC;IAE9C,yFAAyF;IACzF,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,EAC/B,GAAG,EAAE,iBAAiB,CAAC,CAAC,CAAC,GACxB,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC,CAAC;IAElC;;;;;;;;OAQG;IACH,gBAAgB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,EACtC,GAAG,EAAE,iBAAiB,CAAC,CAAC,CAAC,GACxB;QACD,MAAM,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC;QAC9B,MAAM,EAAE,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC,CAAC;KACxC,CAAC;IAEF;;;;;;;;;OASG;IACH,mBAAmB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,EACzC,GAAG,EAAE,0BAA0B,CAAC,CAAC,CAAC,GACjC,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC,CAAC;CACnC;AAED,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,CAAC;IACtC,OAAO,EAAE,MAAM,CAAC;IAChB,wDAAwD;IACxD,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;CACtB;AAED,MAAM,WAAW,WAAW;IAC1B,KAAK,EAAE,WAAW,CAAC;IACnB,QAAQ,EAAE,WAAW,EAAE,CAAC;IACxB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,sBAAsB;IACtB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB;;;;OAIG;IACH,aAAa,CAAC,EAAE,WAAW,CAAC;CAC7B;AAED,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,kDAAkD;IAClD,KAAK,CAAC,EAAE;QAAE,YAAY,CAAC,EAAE,MAAM,CAAC;QAAC,gBAAgB,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAC7D,+CAA+C;IAC/C,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iFAAiF;IACjF,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,iEAAiE;IACjE,eAAe,CAAC,EAAE,WAAW,EAAE,CAAC;CACjC;AAED,MAAM,WAAW,iBAAiB,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU;IACvD,KAAK,EAAE,WAAW,CAAC;IACnB,QAAQ,EAAE,WAAW,EAAE,CAAC;IACxB,MAAM,EAAE,CAAC,CAAC;IACV,mEAAmE;IACnE,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,qCAAqC;IACrC,aAAa,CAAC,EAAE,WAAW,CAAC;CAC7B;AAED,MAAM,WAAW,0BAA0B,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,CAChE,SAAQ,iBAAiB,CAAC,CAAC,CAAC;IAC5B;;;;;OAKG;IACH,WAAW,CAAC,EAAE,KAAK,CAAC;QAClB,IAAI,EAAE,YAAY,CAAC;QACnB,6DAA6D;QAC7D,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,6CAA6C;QAC7C,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;QAC1B,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;KAC3B,CAAC,CAAC;IACH;;;;OAIG;IACH,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB;;;;;OAKG;IACH,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB;;;;;OAKG;IACH,QAAQ,CAAC,EAAE;QAAE,YAAY,EAAE,MAAM,CAAA;KAAE,CAAC;CACrC;AAED,MAAM,WAAW,kBAAkB,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU;IACxD,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IACjB,KAAK,CAAC,EAAE;QAAE,YAAY,CAAC,EAAE,MAAM,CAAC;QAAC,gBAAgB,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IAC7D,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iFAAiF;IACjF,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,iEAAiE;IACjE,eAAe,CAAC,EAAE,WAAW,EAAE,CAAC;CACjC;AAED;;;GAGG;AACH,qBAAa,qBAAsB,YAAW,SAAS;IACrD,IAAI,IAAI,OAAO,CAAC,YAAY,CAAC;IAK7B,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,KAAK,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC;CAKrE"}

package/lib/services/agent/llmCallerAnthropic.d.ts

@@ -1,22 +1,18 @@
 import { z } from "zod";
 import { ChatRequest, ChatResponse, LlmCaller, StructuredRequest, StructuredResponse, StructuredWithToolsRequest } from "./llmCaller";
+import { TransientLlmError, isTransientLlmError } from "./llmRetry";
+export { TransientLlmError, isTransientLlmError };
 /**
- * Direct Anthropic SDK LlmCaller. Exists alongside GatewayLlmCaller because
- * Anthropic's server-side tools (`web_search_20260209`, `web_fetch`, etc.)
- * cannot be exposed through the OpenAI-compatible Vercel AI Gateway —
- * they're Anthropic-only constructs that the API executes server-side.
+ * Direct Anthropic SDK LlmCaller — the `anthropic` slice of the native
+ * ProviderRoutingLlmCaller (LLM_NATIVE_MIGRATION_PLAN). Serves all anthropic
+ * models the router picks (router / planner / executor) and is the only caller
+ * that implements `structuredWithTools`, because Anthropic's server-side tools
+ * (`web_search_20260209`, `web_fetch`, etc.) are Anthropic-only constructs the
+ * API executes server-side — no other provider exposes them.
  *
- * Use this caller when:
- *   - You need server-side `web_search` for planner grounding.
- *   - You want Anthropic-native features (prompt caching, extended thinking,
- *     citations) without a translation layer.
- *
- * Use the gateway caller when:
- *   - You want provider portability (GPT-5 / Gemini fallback chains).
- *   - You don't need built-in tools.
- *
- * The Planner reads from `getAnthropicCaller()` only when `useWebResearch`
- * is set; otherwise it stays on the gateway path.
+ * The Planner also reads this directly via `getAnthropicCaller()` when
+ * `useWebResearch` is set (server-side `web_search` grounding); otherwise it
+ * goes through the normal dispatcher, which already routes anthropic natively.
  *
  * Implemented features:
  *   - Prompt caching (`cachePrompt: true`): `cache_control: ephemeral` on
@@ -25,8 +21,10 @@ import { ChatRequest, ChatResponse, LlmCaller, StructuredRequest, StructuredResp
  *   - Extended thinking (`thinking.budgetTokens`): enables Claude's
  *     interleaved reasoning blocks for T3 / complex briefs. Forces
  *     temperature=1 and threads `thinking` blocks through the tool-use loop.
- *   - Transient errors propagate as `TransientLlmError` so callers can
- *     fall back to the gateway path (no web search) on Anthropic outages.
+ *   - Native incremental streaming (`structuredStream`): forces the
+ *     `submit_<schema>` tool and streams its `input_json_delta` partials.
+ *   - Transient errors propagate as `TransientLlmError` so the dispatcher can
+ *     run the cross-provider `fallbackModel` on Anthropic outages.
  *
  * Deferred work (v2):
  *   - Custom client-side tools (Tavily / image search): extend the loop
@@ -46,15 +44,6 @@ export interface AnthropicCallerConfig {
     /** Max retries on transient errors per call. Default 3. */
     maxRetries?: number;
 }
-/**
- * Marks an error worth retrying on the gateway path. The Planner catches
- * this specifically to degrade gracefully — drops research, keeps planning.
- */
-export declare class TransientLlmError extends Error {
-    readonly cause?: unknown | undefined;
-    constructor(message: string, cause?: unknown | undefined);
-}
-export declare function isTransientLlmError(err: unknown): err is TransientLlmError;
 export declare class AnthropicLlmCaller implements LlmCaller {
     private readonly client;
     private readonly timeoutMs;
@@ -63,6 +52,26 @@ export declare class AnthropicLlmCaller implements LlmCaller {
     chat(req: ChatRequest): Promise<ChatResponse>;
     structured<T extends z.ZodTypeAny>(req: StructuredRequest<T>): Promise<StructuredResponse<T>>;
     structuredWithTools<T extends z.ZodTypeAny>(req: StructuredWithToolsRequest<T>): Promise<StructuredResponse<T>>;
+    /**
+     * Native incremental streaming structured output. Forces the
+     * `submit_<schemaName>` tool and streams its argument JSON as
+     * `input_json_delta` chunks — yielding each partial for token-by-token UI —
+     * then takes the authoritative payload from `stream.finalMessage()` (more
+     * robust than re-concatenating partials) and Zod-validates it.
+     *
+     * Scope: the base StructuredRequest carries no server tools / thinking /
+     * caching (those live on StructuredWithToolsRequest and run the blocking
+     * tool-loop), so this is a single forced-tool turn. Failure handling mirrors
+     * the other streaming callers: a post-stream parse/validation miss rejects
+     * `result` and returns cleanly (so the caller's token loop completes and its
+     * `await result` retry path fires); a mid-stream/connect drop rejects AND
+     * throws out so the iteration aborts.
+     */
+    structuredStream<T extends z.ZodTypeAny>(req: StructuredRequest<T>): {
+        tokens: AsyncIterable<string>;
+        result: Promise<StructuredResponse<T>>;
+    };
+    private streamIterator;
     /**
      * Anthropic's API takes a top-level `system` (string or TextBlockParam[])
      * and messages of user/assistant only. Our LlmCaller carries system as
@@ -78,13 +87,17 @@ export declare class AnthropicLlmCaller implements LlmCaller {
     private splitSystem;
     private buildServerTools;
     /**
-     * Map our ModelChoice → Anthropic API model id. Pass-through today since
-     * routing uses Anthropic-shaped ids (e.g. "claude-sonnet-4-6"). If a
-     * future routing entry uses a non-Anthropic id we'd want to throw rather
-     * than silently send GPT names to the Anthropic API.
+     * Map our ModelChoice → Anthropic API model id via the shared registry
+     * (pass-through today). Throws on a non-Anthropic provider rather than
+     * silently sending GPT/Gemini names to the Anthropic API — the dispatcher
+     * routes by provider, so this guards against a mis-wire.
      */
     private resolveModelId;
+    /**
+     * Delegates to the shared transient-retry helper so all native callers share
+     * one backoff/classification policy (Phase 1 extraction). The SDK's own
+     * `status`/connection-error shapes are recognised by `isTransientError`.
+     */
     private withRetry;
-    private isTransient;
 }
 //# sourceMappingURL=llmCallerAnthropic.d.ts.map

package/lib/services/agent/llmCallerAnthropic.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"llmCallerAnthropic.d.ts","sourceRoot":"","sources":["../../../src/services/agent/llmCallerAnthropic.ts"],"names":[],"mappings":"AAWA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAEL,WAAW,EACX,YAAY,EACZ,SAAS,EACT,iBAAiB,EACjB,kBAAkB,EAClB,0BAA0B,EAC3B,MAAM,aAAa,CAAC;AAGrB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCG;AAEH,MAAM,WAAW,qBAAqB;IACpC,MAAM,EAAE,MAAM,CAAC;IACf,qCAAqC;IACrC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,2DAA2D;IAC3D,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAID;;;GAGG;AACH,qBAAa,iBAAkB,SAAQ,KAAK;IACb,QAAQ,CAAC,KAAK,CAAC,EAAE,OAAO;gBAAzC,OAAO,EAAE,MAAM,EAAW,KAAK,CAAC,EAAE,OAAO,YAAA;CAItD;AAED,wBAAgB,mBAAmB,CAAC,GAAG,EAAE,OAAO,GAAG,GAAG,IAAI,iBAAiB,CAE1E;AAED,qBAAa,kBAAmB,YAAW,SAAS;IAClD,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAY;IACnC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAS;IACnC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAS;gBAExB,GAAG,EAAE,qBAAqB;IAShC,IAAI,CAAC,GAAG,EAAE,WAAW,GAAG,OAAO,CAAC,YAAY,CAAC;IA4B7C,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,EACrC,GAAG,EAAE,iBAAiB,CAAC,CAAC,CAAC,GACxB,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC;IAI3B,mBAAmB,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,EAC9C,GAAG,EAAE,0BAA0B,CAAC,CAAC,CAAC,GACjC,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC;IAsHjC;;;;;;;;;;;OAWG;IACH,OAAO,CAAC,WAAW;IAiCnB,OAAO,CAAC,gBAAgB;IAkBxB;;;;;OAKG;IACH,OAAO,CAAC,cAAc;YASR,SAAS;IAyBvB,OAAO,CAAC,WAAW;CAYpB"}
+{"version":3,"file":"llmCallerAnthropic.d.ts","sourceRoot":"","sources":["../../../src/services/agent/llmCallerAnthropic.ts"],"names":[],"mappings":"AAWA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAEL,WAAW,EACX,YAAY,EACZ,SAAS,EACT,iBAAiB,EACjB,kBAAkB,EAClB,0BAA0B,EAC3B,MAAM,aAAa,CAAC;AAMrB,OAAO,EACL,iBAAiB,EACjB,mBAAmB,EAEpB,MAAM,YAAY,CAAC;AAEpB,OAAO,EAAE,iBAAiB,EAAE,mBAAmB,EAAE,CAAC;AAElD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAkCG;AAEH,MAAM,WAAW,qBAAqB;IACpC,MAAM,EAAE,MAAM,CAAC;IACf,qCAAqC;IACrC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,2DAA2D;IAC3D,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,kBAAmB,YAAW,SAAS;IAClD,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAY;IACnC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAS;IACnC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAS;gBAExB,GAAG,EAAE,qBAAqB;IAShC,IAAI,CAAC,GAAG,EAAE,WAAW,GAAG,OAAO,CAAC,YAAY,CAAC;IA4B7C,UAAU,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,EACrC,GAAG,EAAE,iBAAiB,CAAC,CAAC,CAAC,GACxB,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC;IAI3B,mBAAmB,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,EAC9C,GAAG,EAAE,0BAA0B,CAAC,CAAC,CAAC,GACjC,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC;IAoHjC;;;;;;;;;;;;;;OAcG;IACH,gBAAgB,CAAC,CAAC,SAAS,CAAC,CAAC,UAAU,EACrC,GAAG,EAAE,iBAAiB,CAAC,CAAC,CAAC,GACxB;QACD,MAAM,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC;QAC9B,MAAM,EAAE,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC,CAAC,CAAC;KACxC;YAgBc,cAAc;IAsF7B;;;;;;;;;;;OAWG;IACH,OAAO,CAAC,WAAW;IAkDnB,OAAO,CAAC,gBAAgB;IAkBxB;;;;;OAKG;IACH,OAAO,CAAC,cAAc;IAStB;;;;OAIG;IACH,OAAO,CAAC,SAAS;CAGlB"}

package/lib/services/agent/llmCallerAnthropic.js

@@ -3,26 +3,16 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.AnthropicLlmCaller = exports.TransientLlmError = void 0;
-exports.isTransientLlmError = isTransientLlmError;
+exports.AnthropicLlmCaller = exports.isTransientLlmError = exports.TransientLlmError = void 0;
 const sdk_1 = __importDefault(require("@anthropic-ai/sdk"));
 const zod_1 = require("zod");
-const TRANSIENT_STATUSES = new Set([408, 425, 429, 500, 502, 503, 504]);
-/**
- * Marks an error worth retrying on the gateway path. The Planner catches
- * this specifically to degrade gracefully — drops research, keeps planning.
- */
-class TransientLlmError extends Error {
-    constructor(message, cause) {
-        super(message);
-        this.cause = cause;
-        this.name = "TransientLlmError";
-    }
-}
-exports.TransientLlmError = TransientLlmError;
-function isTransientLlmError(err) {
-    return err instanceof Error && err.name === "TransientLlmError";
-}
+const llmModelRegistry_1 = require("./llmModelRegistry");
+// Transient-retry primitives are shared across all native callers (Phase 1 of
+// LLM_NATIVE_MIGRATION_PLAN). Re-exported here for backward compat with existing
+// importers that pulled `TransientLlmError` / `isTransientLlmError` from this file.
+const llmRetry_1 = require("./llmRetry");
+Object.defineProperty(exports, "TransientLlmError", { enumerable: true, get: function () { return llmRetry_1.TransientLlmError; } });
+Object.defineProperty(exports, "isTransientLlmError", { enumerable: true, get: function () { return llmRetry_1.isTransientLlmError; } });
 class AnthropicLlmCaller {
     constructor(cfg) {
         this.client = new sdk_1.default({
@@ -149,6 +139,101 @@ class AnthropicLlmCaller {
         }
         throw new Error(`AnthropicLlmCaller.structuredWithTools: exceeded ${maxRounds} tool rounds without ${submitToolName}`);
     }
+    /**
+     * Native incremental streaming structured output. Forces the
+     * `submit_<schemaName>` tool and streams its argument JSON as
+     * `input_json_delta` chunks — yielding each partial for token-by-token UI —
+     * then takes the authoritative payload from `stream.finalMessage()` (more
+     * robust than re-concatenating partials) and Zod-validates it.
+     *
+     * Scope: the base StructuredRequest carries no server tools / thinking /
+     * caching (those live on StructuredWithToolsRequest and run the blocking
+     * tool-loop), so this is a single forced-tool turn. Failure handling mirrors
+     * the other streaming callers: a post-stream parse/validation miss rejects
+     * `result` and returns cleanly (so the caller's token loop completes and its
+     * `await result` retry path fires); a mid-stream/connect drop rejects AND
+     * throws out so the iteration aborts.
+     */
+    structuredStream(req) {
+        let resolveResult;
+        let rejectResult;
+        const result = new Promise((resolve, reject) => {
+            resolveResult = resolve;
+            rejectResult = reject;
+        });
+        const self = this;
+        const tokens = {
+            [Symbol.asyncIterator]() {
+                return self.streamIterator(req, resolveResult, rejectResult);
+            },
+        };
+        return { tokens, result };
+    }
+    async *streamIterator(req, resolveResult, rejectResult) {
+        const submitToolName = `submit_${req.schemaName}`;
+        const jsonSchema = zod_1.z.toJSONSchema(req.schema, { target: "draft-7" });
+        const tools = [
+            {
+                name: submitToolName,
+                description: `Submit the final ${req.schemaName} payload. Call this exactly once when ready.`,
+                input_schema: jsonSchema,
+            },
+        ];
+        const { system, messages } = this.splitSystem(req.messages, false);
+        let requestId;
+        let inputTokens;
+        let outputTokens;
+        try {
+            const stream = this.client.messages.stream({
+                model: this.resolveModelId(req.model),
+                max_tokens: req.maxTokens ?? 8192,
+                temperature: req.temperature,
+                system,
+                tools,
+                tool_choice: { type: "tool", name: submitToolName },
+                messages,
+            });
+            for await (const event of stream) {
+                if (event.type === "message_start") {
+                    requestId = event.message.id;
+                    inputTokens = event.message.usage?.input_tokens;
+                }
+                else if (event.type === "content_block_delta" &&
+                    event.delta.type === "input_json_delta") {
+                    const partial = event.delta.partial_json;
+                    if (partial)
+                        yield partial;
+                }
+                else if (event.type === "message_delta") {
+                    outputTokens = event.usage?.output_tokens ?? outputTokens;
+                }
+            }
+            // Authoritative payload from the assembled final message (the forced
+            // tool_use block's input), not the concatenated partials.
+            const final = await stream.finalMessage();
+            const submit = final.content.find((b) => b.type === "tool_use" && b.name === submitToolName);
+            if (!submit) {
+                rejectResult(new Error(`AnthropicLlmCaller.structuredStream: no ${submitToolName} tool_use in final message`));
+                return;
+            }
+            const parsed = req.schema.safeParse(submit.input);
+            if (!parsed.success) {
+                rejectResult(new Error(`AnthropicLlmCaller.structuredStream: schema validation failed for ${req.schemaName}: ${parsed.error.message}`));
+                return;
+            }
+            resolveResult({
+                data: parsed.data,
+                usage: { promptTokens: inputTokens, completionTokens: outputTokens },
+                requestId,
+                attemptedModels: [req.model],
+                usedFallback: false,
+            });
+        }
+        catch (e) {
+            rejectResult(e);
+            throw e;
+        }
+    }
     // ---------- helpers ----------
     /**
      * Anthropic's API takes a top-level `system` (string or TextBlockParam[])
@@ -170,9 +255,24 @@ class AnthropicLlmCaller {
                 systemText = systemText ? `${systemText}\n\n${m.content}` : m.content;
                 continue;
             }
-            // Image attachments are not yet plumbed for the Anthropic caller —
-            // planner doesn't need them today. When the critic moves over (it's
-            // vision-heavy), wire image blocks here.
+            // Vision: when a message carries imageUrls, emit a content-block array
+            // (text + image blocks) instead of a bare string. Uses Anthropic's native
+            // URL image source (`source.type: "url"`) — no base64 round-trip — which
+            // the direct API supports (unlike the gateway's Anthropic translator,
+            // which rejected raw https URLs). Needed for the critic fallback
+            // (claude-sonnet-4-6, vision) now that it routes here instead of the
+            // gateway.
+            if (m.imageUrls?.length) {
+                const blocks = [
+                    { type: "text", text: m.content },
+                    ...m.imageUrls.map((url) => ({
+                        type: "image",
+                        source: { type: "url", url },
+                    })),
+                ];
+                out.push({ role: m.role, content: blocks });
+                continue;
+            }
             out.push({ role: m.role, content: m.content });
         }
         if (!systemText)
@@ -207,49 +307,24 @@ class AnthropicLlmCaller {
         });
     }
     /**
-     * Map our ModelChoice → Anthropic API model id. Pass-through today since
-     * routing uses Anthropic-shaped ids (e.g. "claude-sonnet-4-6"). If a
-     * future routing entry uses a non-Anthropic id we'd want to throw rather
-     * than silently send GPT names to the Anthropic API.
+     * Map our ModelChoice → Anthropic API model id via the shared registry
+     * (pass-through today). Throws on a non-Anthropic provider rather than
+     * silently sending GPT/Gemini names to the Anthropic API — the dispatcher
+     * routes by provider, so this guards against a mis-wire.
      */
     resolveModelId(m) {
-        if (m.provider !== "anthropic") {
-            throw new Error(`AnthropicLlmCaller cannot serve provider=${m.provider}; route to GatewayLlmCaller instead.`);
-        }
-        return m.modelId;
-    }
-    async withRetry(fn) {
-        let lastErr;
-        for (let attempt = 0; attempt < this.maxRetries; attempt++) {
-            try {
-                return await fn();
-            }
-            catch (err) {
-                lastErr = err;
-                const transient = this.isTransient(err);
-                if (!transient || attempt === this.maxRetries - 1) {
-                    if (transient) {
-                        throw new TransientLlmError(`Anthropic transient error after ${attempt + 1} attempts: ${String(err?.message ?? err)}`, err);
-                    }
-                    throw err;
-                }
-                await new Promise((r) => setTimeout(r, 250 * Math.pow(4, attempt)));
-            }
+        if ((0, llmModelRegistry_1.normalizeProvider)(m.provider) !== "anthropic") {
+            throw new Error(`AnthropicLlmCaller cannot serve provider=${m.provider}; route via the dispatcher.`);
         }
-        throw lastErr;
+        return (0, llmModelRegistry_1.nativeModelId)(m);
     }
-    isTransient(err) {
-        if (!err || typeof err !== "object")
-            return false;
-        const e = err;
-        if (typeof e.status === "number" && TRANSIENT_STATUSES.has(e.status)) {
-            return true;
-        }
-        if (e.name === "APIConnectionError" || e.name === "APIConnectionTimeoutError") {
-            return true;
-        }
-        const msg = String(e.message ?? "");
-        return /timeout|ECONNRESET|ETIMEDOUT|network|socket hang up/i.test(msg);
+    /**
+     * Delegates to the shared transient-retry helper so all native callers share
+     * one backoff/classification policy (Phase 1 extraction). The SDK's own
+     * `status`/connection-error shapes are recognised by `isTransientError`.
+     */
+    withRetry(fn) {
+        return (0, llmRetry_1.withRetry)(fn, { maxRetries: this.maxRetries });
     }
 }
 exports.AnthropicLlmCaller = AnthropicLlmCaller;