@deepstrike/sdk 0.2.51 → 0.2.60

package/dist/providers/anthropic.js

@@ -1,30 +1,34 @@
 import Anthropic from "@anthropic-ai/sdk";
 import { assistantReplayKey } from "../runtime/provider-replay.js";
 import { withServerRuntimeGuard } from "../runtime/server.js";
-import { CircuitBreaker, normalizeToolCall, omitExtensionKeys, toAnthropicMessages } from "./base.js";
-const CLAUDE_POLICIES = {
-    "claude-opus-4-1": { maxTurns: 50 },
-    "claude-opus-4-7": { maxTurns: 50 },
-    "claude-opus-4-6": { maxTurns: 50 },
-    "claude-opus-4-0": { maxTurns: 50 },
-    "claude-sonnet-4-6": { maxTurns: 25 },
-    "claude-sonnet-4-0": { maxTurns: 25 },
-    "claude-haiku-4-5": { maxTurns: 15 },
-    "claude-3-5-haiku-latest": { maxTurns: 15 },
-};
+import { CircuitBreaker } from "./base.js";
+import { AnthropicMessagesAdapter, } from "./anthropic-adapter.js";
+import { normalizeCanonicalAdapterInput, } from "./content-normalization.js";
+import { endpointProfiles } from "./endpoints.js";
+import { circuitOpenError, classifyProviderError } from "./provider-error.js";
 export class AnthropicProvider {
     client;
     circuit;
     maxRetries;
     baseDelay;
     model;
+    adapter = new AnthropicMessagesAdapter();
     nativeAssistantBlocks = new Map();
-    // Accepts the options object (`new AnthropicProvider({ apiKey, model, baseURL })`) or the legacy
-    // positional form (still used by the Anthropic-compatible backend subclasses' `super(...)` calls).
-    constructor(apiKeyOrConfig, model = "claude-sonnet-4-6", retry = { maxRetries: 3, baseDelay: 1000 }, options = {}) {
-        const c = typeof apiKeyOrConfig === "string"
-            ? { apiKey: apiKeyOrConfig, model, retry, ...options }
-            : { model: "claude-sonnet-4-6", retry: { maxRetries: 3, baseDelay: 1000 }, ...apiKeyOrConfig };
+    resolvedRuntimePolicy;
+    directNativeTokenCounting;
+    resolvedRuntime;
+    constructor(config) {
+        if (!config || typeof config !== "object" || Array.isArray(config)) {
+            throw new TypeError("AnthropicProvider requires a configuration object");
+        }
+        if (typeof config.apiKey !== "string" || config.apiKey.length === 0) {
+            throw new TypeError("AnthropicProvider requires a non-empty apiKey");
+        }
+        const c = {
+            model: "claude-sonnet-4-6",
+            retry: { maxRetries: 3, baseDelay: 1000 },
+            ...config,
+        };
         this.model = c.model ?? "claude-sonnet-4-6";
         this.client = withServerRuntimeGuard(() => new Anthropic({
             ...(c.authMode === "bearer"
@@ -35,11 +39,14 @@ export class AnthropicProvider {
         this.circuit = new CircuitBreaker();
         this.maxRetries = c.retry?.maxRetries ?? 3;
         this.baseDelay = c.retry?.baseDelay ?? 1000;
+        this.resolvedRuntimePolicy = c.runtimePolicy ?? {};
+        this.directNativeTokenCounting = c.baseURL === undefined
+            || c.baseURL === endpointProfiles["anthropic.messages"].baseURL;
     }
     runtimePolicy() {
-        return CLAUDE_POLICIES[this.model] ?? {};
+        return this.resolvedRuntimePolicy;
     }
-    /** Identity advertised in the descriptor; overridden by Anthropic-compatible vendors (e.g. MiniMax). */
+    /** Identity advertised in the descriptor; overridden by Anthropic-compatible vendors. */
     providerName() {
         return "anthropic";
     }
@@ -59,419 +66,186 @@ export class AnthropicProvider {
             },
         };
     }
+    bindResolvedRuntime(resolved) {
+        if (resolved.identity.protocol !== "anthropic-messages"
+            || resolved.identity.providerId !== this.providerName()
+            || resolved.identity.modelId !== this.model) {
+            throw new Error("AnthropicProvider received a mismatched resolved runtime");
+        }
+        this.resolvedRuntime = resolved;
+    }
     peekProviderReplay(message) {
         const blocks = this.nativeAssistantBlocks.get(assistantReplayKey(message));
-        return blocks?.length ? { native_blocks: blocks } : undefined;
+        return blocks?.length ? { protocol: "anthropic-messages", native_blocks: blocks } : undefined;
     }
     seedProviderReplay(message, replay) {
-        if (replay.native_blocks?.length) {
+        if (replay.protocol === "anthropic-messages" && replay.native_blocks?.length) {
             this.nativeAssistantBlocks.set(assistantReplayKey(message), replay.native_blocks);
-            return;
         }
-        // Legacy log without persisted native blocks: reconstruct neutral
-        // text + tool_use blocks from the transcript so a tool-use turn can be
-        // replayed. Thinking blocks were never persisted, so they are not recovered.
-        const blocks = reconstructAnthropicBlocks(message);
-        if (blocks.length)
-            this.nativeAssistantBlocks.set(assistantReplayKey(message), blocks);
     }
-    /**
-     * Build tool definitions. A cache breakpoint is anchored on the final tool
-     * only when the system blocks won't carry one (`anchorCache`). When structured
-     * system blocks are present, their breakpoints already cache the tools prefix
-     * (tools render before system), so a redundant tool breakpoint would only burn
-     * one of Anthropic's 4 cache_control slots — slots the message history needs.
-     */
-    buildTools(tools, anchorCache, strategy) {
-        // Tool cache_control is emitted under "default" and "tools-only". "system-only",
-        // "frozen-prefix", and "none" all skip it.
-        const emitOnLastTool = anchorCache &&
-            (strategy === "default" || strategy === "tools-only");
-        return tools.map((t, i) => ({
-            name: t.name,
-            description: t.description,
-            input_schema: JSON.parse(t.parameters),
-            ...(emitOnLastTool && i === tools.length - 1 ? { cache_control: { type: "ephemeral" } } : {}),
-        }));
+    adapterInput(context, tools, extensions) {
+        const providerId = this.providerName();
+        const endpoint = Object.values(endpointProfiles).find(profile => profile.providerId === providerId && profile.protocol === "anthropic-messages") ?? endpointProfiles["anthropic.messages"];
+        const resolved = this.resolvedRuntime ?? {
+            identity: {
+                providerId,
+                modelId: this.model,
+                endpointId: endpoint.id,
+                protocol: "anthropic-messages",
+            },
+            model: {
+                id: `${providerId}/${this.model}`,
+                providerId,
+                kind: "generation",
+                intrinsic: {},
+            },
+            endpoint,
+            adapter: this,
+            effectiveCapabilities: compatibilityCapabilities(),
+        };
+        return normalizeCanonicalAdapterInput({
+            context,
+            tools,
+            resolved,
+            extensions,
+            replayForMessage: message => this.peekProviderReplay(message),
+        });
+    }
+    buildPlan(context, tools, extensions) {
+        const input = this.adapterInput(context, tools, extensions);
+        return { input, plan: this.adapter.buildRequest(input) };
     }
     async complete(context, tools, extensions) {
+        const provider = this.providerName();
         if (this.circuit.isOpen())
-            throw new Error("Circuit breaker open");
-        const strategy = resolveCacheBreakpointStrategy(extensions);
-        const system = this.buildSystem(context, strategy);
-        const msgs = this.buildMessages(context, strategy);
-        assertCacheBudget(system, tools.length);
-        const requestExtensions = this.requestExtensions(extensions);
+            throw circuitOpenError(provider);
+        let input;
+        let plan;
+        try {
+            ;
+            ({ input, plan } = this.buildPlan(context, tools, extensions));
+        }
+        catch (error) {
+            throw classifyProviderError(provider, error);
+        }
         let lastErr;
-        for (let i = 0; i < this.maxRetries; i++) {
+        for (let attempt = 0; attempt < this.maxRetries; attempt++) {
             try {
-                const resp = await this.createMessage({
-                    ...requestExtensions,
-                    model: this.model,
-                    max_tokens: typeof extensions?.max_tokens === "number" ? extensions.max_tokens : 8096,
-                    ...(system ? { system } : {}),
-                    messages: msgs,
-                    ...(tools.length ? { tools: this.buildTools(tools, !Array.isArray(system), strategy) } : {}),
-                }, extensions);
+                const raw = await this.createMessage(plan.params, plan.transport);
                 this.circuit.recordSuccess();
-                let content = "";
-                const toolCalls = [];
-                for (const block of resp.content) {
-                    if (block.type === "text")
-                        content += block.text;
-                    else if (block.type === "tool_use") {
-                        const tc = normalizeToolCall(block.id, block.name, block.input);
-                        if (tc)
-                            toolCalls.push(tc);
-                    }
+                const decoded = this.adapter.decodeComplete(raw, { input });
+                if (decoded.replay?.native_blocks) {
+                    this.rememberNativeBlocks(decoded.message, decoded.replay.native_blocks);
                 }
-                const message = { role: "assistant", content, tokenCount: resp.usage.output_tokens, toolCalls };
-                this.rememberNativeBlocks(message, resp.content);
-                return message;
+                return decoded.message;
             }
-            catch (err) {
-                lastErr = err;
+            catch (error) {
+                lastErr = error;
                 this.circuit.recordFailure();
-                if (i < this.maxRetries - 1)
-                    await new Promise(r => setTimeout(r, this.baseDelay * 2 ** i));
+                if (attempt < this.maxRetries - 1) {
+                    await new Promise(resolve => setTimeout(resolve, this.baseDelay * 2 ** attempt));
+                }
             }
         }
-        throw lastErr;
+        throw classifyProviderError(provider, lastErr);
+    }
+    /** Native measurement belongs to the verified official endpoint, not the wire protocol. */
+    async countTokens(context, tools, extensions) {
+        const enabled = this.resolvedRuntime
+            ? this.resolvedRuntime.effectiveCapabilities.nativeTokenCounting.state === "supported"
+            : this.providerName() === "anthropic" && this.directNativeTokenCounting;
+        if (!enabled) {
+            throw new Error(`Native token counting is unavailable on ${this.providerName()} Anthropic-compatible endpoint`);
+        }
+        const { plan } = this.buildPlan(context, tools, extensions);
+        const { model, system, messages, tools: requestTools } = plan.params;
+        const response = await this.client.messages.countTokens({
+            model,
+            ...(system ? { system } : {}),
+            messages,
+            ...(requestTools ? { tools: requestTools } : {}),
+        });
+        return {
+            inputTokens: response.input_tokens,
+            source: { kind: "native", provider: "anthropic" },
+            confidence: "exact",
+        };
     }
     async *stream(context, tools, extensions, _state, signal) {
-        const strategy = resolveCacheBreakpointStrategy(extensions);
-        const system = this.buildSystem(context, strategy);
-        const msgs = this.buildMessages(context, strategy);
-        assertCacheBudget(system, tools.length);
-        const requestExtensions = this.requestExtensions(extensions);
-        const builtTools = tools.length ? this.buildTools(tools, !Array.isArray(system), strategy) : undefined;
-        // I1: capture which slots will carry cache_control so cache_read_input_tokens can be
-        // attributed pro-rata when the response arrives. Honest annotation: Anthropic returns one
-        // scalar (no per-slot breakdown), so this is an estimate, not authoritative.
-        const slotBp = countCacheControlSlots(system, builtTools, msgs);
-        const toolBlocks = {};
-        const nativeBlocks = {};
-        let finalText = "";
-        const finalToolCalls = [];
-        const stream = this.streamMessage({
-            ...requestExtensions,
-            model: this.model,
-            max_tokens: typeof extensions?.max_tokens === "number" ? extensions.max_tokens : 8096,
-            ...(system ? { system } : {}),
-            messages: msgs,
-            ...(builtTools ? { tools: builtTools } : {}),
-        }, extensions, signal);
-        let uncachedInput = 0;
-        let cacheReadTokens = 0;
-        let cacheCreationTokens = 0;
-        let outputTokens = 0;
-        for await (const evt of stream) {
-            if (evt.type === "message_start" || evt.type === "message_delta") {
-                const usage = evt.usage ?? evt.message?.usage;
-                if (usage) {
-                    // input + cache counts are cumulative and pinned at message_start; a
-                    // later message_delta may omit them (null), so Math.max keeps the
-                    // running totals from being clobbered back to zero.
-                    uncachedInput = Math.max(uncachedInput, usage.input_tokens ?? 0);
-                    cacheReadTokens = Math.max(cacheReadTokens, usage.cache_read_input_tokens ?? 0);
-                    cacheCreationTokens = Math.max(cacheCreationTokens, usage.cache_creation_input_tokens ?? 0);
-                    outputTokens = Math.max(outputTokens, usage.output_tokens ?? 0);
-                    // inputTokens is the FULL prompt size (uncached + cache read + cache
-                    // write). The kernel reads it as the authoritative prompt size for
-                    // context-pressure/compaction — excluding cached tokens would make a
-                    // cache-heavy turn look tiny and suppress compaction until a 413.
-                    const inputTokens = uncachedInput + cacheReadTokens + cacheCreationTokens;
-                    const bySlot = estimateCacheReadBySlot(cacheReadTokens, slotBp);
-                    // stop_reason is only present on message_delta (the closing frame). `max_tokens` drives
-                    // the kernel's output-cap recovery; other reasons (end_turn/tool_use) are informational.
-                    const stopReason = evt.delta?.stop_reason;
-                    yield {
-                        type: "usage",
-                        totalTokens: inputTokens + outputTokens,
-                        inputTokens,
-                        outputTokens,
-                        cacheReadInputTokens: cacheReadTokens,
-                        cacheCreationInputTokens: cacheCreationTokens,
-                        ...(bySlot ? { cacheReadInputTokensBySlot: bySlot } : {}),
-                        ...(stopReason ? { stopReason } : {}),
-                    };
-                }
-            }
-            else if (evt.type === "content_block_start") {
-                nativeBlocks[evt.index] = { ...evt.content_block };
-                if (evt.content_block.type === "tool_use") {
-                    toolBlocks[evt.index] = { id: evt.content_block.id, name: evt.content_block.name, argsBuf: "" };
-                }
-            }
-            else if (evt.type === "content_block_delta") {
-                const d = evt.delta;
-                if (d.type === "text_delta") {
-                    finalText += d.text;
-                    nativeBlocks[evt.index] = { ...nativeBlocks[evt.index], text: String(nativeBlocks[evt.index]?.text ?? "") + d.text };
-                    yield { type: "text_delta", delta: d.text };
-                }
-                else if (d.type === "thinking_delta") {
-                    nativeBlocks[evt.index] = { ...nativeBlocks[evt.index], thinking: String(nativeBlocks[evt.index]?.thinking ?? "") + d.thinking };
-                    yield { type: "thinking_delta", delta: d.thinking };
-                }
-                else if (d.type === "signature_delta") {
-                    nativeBlocks[evt.index] = { ...nativeBlocks[evt.index], signature: String(nativeBlocks[evt.index]?.signature ?? "") + d.signature };
-                }
-                else if (d.type === "input_json_delta" && toolBlocks[evt.index]) {
-                    toolBlocks[evt.index].argsBuf += d.partial_json;
-                }
+        const provider = this.providerName();
+        try {
+            const { input, plan } = this.buildPlan(context, tools, extensions);
+            const state = this.adapter.createStreamState({ input });
+            for await (const chunk of this.streamMessage(plan.params, plan.transport, signal)) {
+                for (const event of this.adapter.pushStreamChunk(chunk, state).events)
+                    yield event;
             }
-            else if (evt.type === "content_block_stop" && toolBlocks[evt.index] !== undefined) {
-                const tb = toolBlocks[evt.index];
-                delete toolBlocks[evt.index];
-                let args = {};
-                try {
-                    args = JSON.parse(tb.argsBuf || "{}");
-                }
-                catch {
-                    args = {};
-                }
-                nativeBlocks[evt.index] = { ...nativeBlocks[evt.index], input: args };
-                finalToolCalls.push({ id: tb.id, name: tb.name, arguments: JSON.stringify(args) });
-                yield { type: "tool_call", id: tb.id, name: tb.name, arguments: args };
+            const final = this.adapter.finishStream(state);
+            for (const event of final.events)
+                yield event;
+            if (final.replay?.native_blocks) {
+                this.rememberNativeBlocks({ content: state.finalText, toolCalls: state.finalToolCalls }, final.replay.native_blocks);
             }
         }
-        this.rememberNativeBlocks({ content: finalText, toolCalls: finalToolCalls }, Object.keys(nativeBlocks).map(Number).sort((a, b) => a - b).map(index => nativeBlocks[index]));
-    }
-    requestExtensions(extensions) {
-        return omitExtensionKeys(extensions, ["model", "messages", "system", "tools", "max_tokens", "stream"]);
-    }
-    hasBetas(extensions) {
-        const betas = extensions?.betas;
-        return Array.isArray(betas) && betas.length > 0;
+        catch (error) {
+            throw classifyProviderError(provider, error);
+        }
     }
-    createMessage(params, extensions) {
-        return this.hasBetas(extensions)
+    createMessage(params, transport) {
+        return (transport === "beta"
             ? this.client.beta.messages.create(params)
-            : this.client.messages.create(params);
+            : this.client.messages.create(params));
     }
-    streamMessage(params, extensions, signal) {
-        // #2-B-ii: forward the abort signal as a request option so a preempt cancels the HTTP request.
-        const opts = signal ? { signal } : undefined;
-        return (this.hasBetas(extensions)
-            ? this.client.beta.messages.stream(params, opts)
-            : this.client.messages.stream(params, opts));
+    streamMessage(params, transport, signal) {
+        const options = signal ? { signal } : undefined;
+        return (transport === "beta"
+            ? this.client.beta.messages.stream(params, options)
+            : this.client.messages.stream(params, options));
     }
+    // White-box test seams. Request construction itself belongs to the adapter.
     buildSystem(context, strategy) {
-        // B3 note: the system shape is content-driven — 0 blocks (string), 1 block
-        // (stable only), or 2 blocks (stable + knowledge). The first turn `systemKnowledge`
-        // appears, the block count rises 1→2, which is a one-time prompt-cache invalidation
-        // (the knowledge prefix didn't exist to cache before). It is byte-stable thereafter;
-        // dynamic per-turn knowledge belongs in the uncached tail, not this block. An empty
-        // knowledge string is intentionally never emitted (the API rejects empty text blocks).
-        if (!context.systemStable && !context.systemKnowledge) {
-            return context.systemText || undefined;
-        }
-        // System cache_control is emitted under "default" and "system-only". Other strategies
-        // keep the text-block structure for protocol parity but omit cache_control.
-        const emitOnSystemBlocks = strategy === "default" || strategy === "system-only";
-        const cc = { type: "ephemeral" };
-        const blocks = [];
-        if (context.systemStable) {
-            blocks.push({ type: "text", text: context.systemStable, ...(emitOnSystemBlocks ? { cache_control: cc } : {}) });
-        }
-        if (context.systemKnowledge) {
-            blocks.push({ type: "text", text: context.systemKnowledge, ...(emitOnSystemBlocks ? { cache_control: cc } : {}) });
-        }
-        return blocks.length ? blocks : undefined;
+        return this.buildPlan(context, [], { cacheBreakpointStrategy: strategy }).plan.params.system;
     }
     buildMessages(context, strategy) {
-        const msgs = toAnthropicMessages(context.turns, message => this.nativeAssistantBlocks.get(assistantReplayKey(message)));
-        // Cache breakpoints anchor on the stable history; the volatile State turn is
-        // appended AFTER them as the uncached tail (so the history prefix re-reads
-        // across turns). On un-rebuilt bindings stateTurn is absent and the state is
-        // already inside `turns` — rendered as-is above. `frozenPrefixLen` (P1-E) pins
-        // the deep breakpoint at the compaction boundary; absent ⇒ rolling-pair fallback.
-        applyMessageCacheControl(msgs, context.frozenPrefixLen, strategy);
-        if (context.stateTurn) {
-            // Render through toAnthropicMessages so assistant tool_use blocks and
-            // tool-role tool_result parts are serialized correctly — toAnthropicContent
-            // only handles contentParts/content and would silently drop toolCalls.
-            const stateMsgs = toAnthropicMessages([context.stateTurn], message => this.nativeAssistantBlocks.get(assistantReplayKey(message)));
-            msgs.push(...stateMsgs);
-        }
-        if (msgs.length === 0) {
-            msgs.push({ role: "user", content: "Proceed." });
-        }
-        return msgs;
+        return this.buildPlan(context, [], { cacheBreakpointStrategy: strategy }).plan.params.messages;
     }
     rememberNativeBlocks(message, blocks) {
         if (!blocks.length)
             return;
-        if (!message.toolCalls?.length && !blocks.some(b => b.type === "thinking"))
+        if (!message.toolCalls?.length && !blocks.some(block => block.type === "thinking"))
             return;
         this.nativeAssistantBlocks.set(assistantReplayKey(message), blocks);
     }
 }
-/** Recognised cache-breakpoint strategy values; any other input (incl. undefined) falls to `"default"`. */
-const CACHE_BREAKPOINT_STRATEGIES = new Set([
-    "default", "tools-only", "system-only", "frozen-prefix", "none",
-]);
-/** Pull `cacheBreakpointStrategy` from per-call extensions; unrecognised values → `"default"`. */
-function resolveCacheBreakpointStrategy(extensions) {
-    const raw = extensions?.cacheBreakpointStrategy;
-    if (typeof raw === "string" && CACHE_BREAKPOINT_STRATEGIES.has(raw)) {
-        return raw;
-    }
-    return "default";
-}
-/**
- * I1: count which slots of the outgoing request carry a `cache_control` breakpoint. Used to
- * pro-rata-attribute the response's `cache_read_input_tokens` (a single scalar with no per-slot
- * breakdown) across the slots that contributed to the cache hit. Returns whether each slot has
- * any breakpoint — not the actual count, since pro-rata only needs the contributing slot set.
- */
-function countCacheControlSlots(system, builtTools, msgs) {
-    const sysBp = Array.isArray(system) && system.some(b => b?.cache_control != null);
-    const toolBp = !!builtTools && builtTools.some(t => t?.cache_control != null);
-    let msgBp = false;
-    for (const m of msgs) {
-        if (Array.isArray(m.content)) {
-            if (m.content.some(b => b?.cache_control != null)) {
-                msgBp = true;
-                break;
-            }
-        }
-    }
-    return { system: sysBp, tools: toolBp, messages: msgBp };
-}
-/**
- * I1: split the response's `cache_read_input_tokens` evenly across the slots that carried a
- * cache_control breakpoint on the request. Returns undefined when there's no cache read or no
- * contributing slot — in those cases the consumer is better off seeing the field absent than
- * seeing all zeros. The remainder (if the total doesn't divide evenly) lands on the first
- * contributing slot to keep the sum exact.
- */
-function estimateCacheReadBySlot(cacheRead, slotBp) {
-    if (cacheRead <= 0)
-        return undefined;
-    const count = (slotBp.system ? 1 : 0) + (slotBp.tools ? 1 : 0) + (slotBp.messages ? 1 : 0);
-    if (count === 0)
-        return undefined;
-    const share = Math.floor(cacheRead / count);
-    const remainder = cacheRead - share * count;
-    const out = {};
-    let firstDone = false;
-    const give = () => {
-        if (!firstDone) {
-            firstDone = true;
-            return share + remainder;
-        }
-        return share;
+function compatibilityCapabilities() {
+    const unknown = { state: "unknown", evidence: [] };
+    const unsupported = { state: "unsupported", evidence: ["protocol"] };
+    return {
+        inputModalities: {
+            text: unknown,
+            image: unknown,
+            audio: unsupported,
+            video: unsupported,
+            file: unsupported,
+        },
+        outputModalities: {
+            text: unknown,
+            image: unsupported,
+            audio: unsupported,
+            embedding: unsupported,
+        },
+        tools: unknown,
+        reasoning: unknown,
+        parallelToolCalls: unknown,
+        structuredOutput: unsupported,
+        promptCaching: unknown,
+        nativeTokenCounting: unknown,
+        mediaForms: {
+            imageUrl: unknown,
+            imageBase64: unknown,
+            fileId: unsupported,
+            audioUrl: unsupported,
+            audioBase64: unsupported,
+        },
     };
-    if (slotBp.system)
-        out.system = give();
-    if (slotBp.tools)
-        out.tools = give();
-    if (slotBp.messages)
-        out.messages = give();
-    return out;
-}
-/** Anthropic accepts at most this many cache_control breakpoints per request. */
-const MAX_CACHE_BREAKPOINTS = 4;
-/**
- * Number of rolling cache breakpoints to spend on the message history. Anthropic
- * allows 4 cache_control breakpoints total; the static system/tools prefix
- * consumes up to 2 (systemStable + systemKnowledge), leaving 2 for the history.
- */
-const MESSAGE_CACHE_BREAKPOINTS = 2;
-/**
- * Regression guard: fail loudly if the static (system + tools) breakpoints plus
- * the rolling message budget could exceed Anthropic's hard limit, instead of
- * letting the API reject the request with an opaque 400. Uses the worst-case
- * message count (`MESSAGE_CACHE_BREAKPOINTS`), so it can only fire if a future
- * change adds a system partition or raises the message budget.
- */
-function assertCacheBudget(system, toolCount) {
-    const systemBreakpoints = Array.isArray(system) ? system.length : 0;
-    const toolBreakpoints = toolCount > 0 && !Array.isArray(system) ? 1 : 0;
-    const worstCase = systemBreakpoints + toolBreakpoints + MESSAGE_CACHE_BREAKPOINTS;
-    if (worstCase > MAX_CACHE_BREAKPOINTS) {
-        throw new Error(`Anthropic cache_control budget exceeded: ${systemBreakpoints} system + ${toolBreakpoints} tool + ${MESSAGE_CACHE_BREAKPOINTS} message > ${MAX_CACHE_BREAKPOINTS}`);
-    }
-}
-/**
- * Place the (≤2) message-history cache breakpoints. The final message always gets
- * one — it writes the current full prefix for the next turn to read. The second is
- * placed by one of two strategies:
- *
- *   • **Deep anchor (P1-E)** — when `frozenPrefixLen` marks a distinct frozen prefix
- *     (the compaction boundary), pin the second breakpoint there. It is byte-stable
- *     across turns, so `[0..frozen]` is re-read cheaply every turn and is immune to
- *     the 20-block lookback miss that strikes heavy tool turns (>20 blocks/turn); the
- *     tail breakpoint then writes only the incremental `[frozen..tail]`.
- *   • **Rolling fallback** — otherwise (older binding / no compaction yet / whole
- *     render hot), roll the second breakpoint to the nearest preceding user turn, the
- *     previous turn's read anchor (Anthropic's 20-block lookback bridges light turns).
- *
- * Without any of this the cached prefix stops at the end of `system` and every turn
- * re-bills the entire tool-result history at full price (~quadratic cumulative cost).
- * cache_control attaches to the last content block of each target, promoting a bare
- * string body to a text block.
- */
-function applyMessageCacheControl(msgs, frozenPrefixLen, strategy) {
-    if (!msgs.length)
-        return;
-    // Message-level cache_control is emitted under "default" and "frozen-prefix" only.
-    // "tools-only", "system-only", and "none" skip the history entirely.
-    if (strategy === "tools-only" || strategy === "system-only" || strategy === "none")
-        return;
-    const targets = new Set([msgs.length - 1]);
-    if (typeof frozenPrefixLen === "number" && frozenPrefixLen >= 1 && frozenPrefixLen < msgs.length) {
-        // Deep anchor at the frozen-prefix boundary (last frozen turn). Fixed between compactions.
-        targets.add(frozenPrefixLen - 1);
-    }
-    else if (strategy === "default") {
-        // Rolling fallback is part of the default strategy only — `"frozen-prefix"` deliberately
-        // skips it so a verify can isolate the deep-anchor contribution from the rolling pair.
-        for (let i = msgs.length - 2; i >= 0 && targets.size < MESSAGE_CACHE_BREAKPOINTS; i--) {
-            if (msgs[i].role === "user")
-                targets.add(i);
-        }
-    }
-    for (const idx of targets)
-        markLastBlockCacheable(msgs[idx]);
-}
-/** Attach an ephemeral cache breakpoint to a message's final content block. */
-function markLastBlockCacheable(msg) {
-    const cache_control = { type: "ephemeral" };
-    if (typeof msg.content === "string") {
-        if (!msg.content)
-            return; // don't synthesize an empty (API-rejected) text block
-        msg.content = [{ type: "text", text: msg.content, cache_control }];
-        return;
-    }
-    if (Array.isArray(msg.content) && msg.content.length) {
-        const last = msg.content[msg.content.length - 1];
-        last.cache_control = cache_control;
-    }
-}
-/**
- * Reconstruct Anthropic assistant content blocks from a neutral transcript when
- * no provider replay was persisted. Only meaningful for tool-use turns: a plain
- * text turn needs no native blocks to replay.
- */
-function reconstructAnthropicBlocks(message) {
-    const toolCalls = message.toolCalls ?? [];
-    if (!toolCalls.length)
-        return [];
-    const blocks = [];
-    if (message.content)
-        blocks.push({ type: "text", text: message.content });
-    for (const tc of toolCalls) {
-        let input = {};
-        try {
-            input = JSON.parse(tc.arguments || "{}");
-        }
-        catch {
-            input = {};
-        }
-        blocks.push({ type: "tool_use", id: tc.id, name: tc.name, input });
-    }
-    return blocks;
 }