@clinebot/llms 0.0.10 → 0.0.11

package/src/providers/transform/anthropic-format.ts

@@ -32,12 +32,24 @@ export function convertToAnthropicMessages(
 	messages: Message[],
 	enableCaching = false,
 ): AnthropicMessage[] {
+	const userMessageIndices = messages.reduce<number[]>(
+		(indices, message, index) => {
+			if (message.role === "user") {
+				indices.push(index);
+			}
+			return indices;
+		},
+		[],
+	);
+	const cacheableMessageIndices = enableCaching
+		? new Set(userMessageIndices.slice(-2))
+		: new Set<number>();
 	const result: AnthropicMessage[] = [];
 
-	for (const message of messages) {
+	for (const [index, message] of messages.entries()) {
 		const converted = convertMessage(
 			message,
-			enableCaching && messages.indexOf(message) === messages.length - 1,
+			cacheableMessageIndices.has(index),
 		);
 		if (converted) {
 			result.push(converted);

package/src/providers/transform/format-conversion.test.ts

@@ -285,10 +285,33 @@ describe("format conversion", () => {
 		];
 
 		const anthropic = convertToAnthropicMessages(messages, true) as any[];
+		expect(anthropic[0].content[0].cache_control).toEqual({
+			type: "ephemeral",
+		});
 		expect(anthropic[1].content[0].type).toBe("thinking");
 		expect(anthropic[1].content[0].signature).toBe("anthropic-sig");
 	});
 
+	it("applies anthropic cache markers to the last two user messages", () => {
+		const messages: Message[] = [
+			{ role: "user", content: "first prompt" },
+			{ role: "assistant", content: "intermediate response" },
+			{ role: "user", content: "second prompt" },
+			{ role: "assistant", content: "another response" },
+			{ role: "user", content: "third prompt" },
+		];
+
+		const anthropic = convertToAnthropicMessages(messages, true) as any[];
+
+		expect(anthropic[0].content[0].cache_control).toBeUndefined();
+		expect(anthropic[2].content[0].cache_control).toEqual({
+			type: "ephemeral",
+		});
+		expect(anthropic[4].content[0].cache_control).toEqual({
+			type: "ephemeral",
+		});
+	});
+
 	it("normalizes array-shaped tool_use input for anthropic replay", () => {
 		const messages: Message[] = [
 			{ role: "user", content: "run these" },

package/src/providers/types/stream.ts

@@ -55,7 +55,7 @@ export interface ApiStreamReasoningChunk {
  */
 export interface ApiStreamUsageChunk {
 	type: "usage";
-	/** Number of input tokens (excluding cached) */
+	/** Total number of input tokens reported by the provider */
 	inputTokens: number;
 	/** Number of output tokens */
 	outputTokens: number;