@copilotkit/aimock 1.29.0 → 1.30.0

package/dist/messages.js

@@ -1,75 +1,92 @@
-import { extractOverrides, flattenHeaders, generateMessageId, generateToolUseId, getContext, getTestId, isContentWithToolCallsResponse, isErrorResponse, isTextResponse, isToolCallResponse, resolveResponse, resolveStrictMode, strictOverrideField } from "./helpers.js";
-import { matchFixture } from "./router.js";
+import { extractOverrides, flattenHeaders, generateMessageId, generateToolUseId, getContext, getTestId, isContentWithToolCallsResponse, isErrorResponse, isTextResponse, isToolCallResponse, resolveReasoningArtifactsForModel, resolveReasoningForModel, resolveResponse, resolveStrictMode, strictNoMatchLogLine, strictNoMatchMessage, strictOverrideField } from "./helpers.js";
+import { matchFixtureDiagnostic } from "./router.js";
 import { calculateDelay, delay, writeErrorResponse } from "./sse-writer.js";
 import { createInterruptionSignal } from "./interruption.js";
 import { applyChaos } from "./chaos.js";
 import { proxyAndRecord } from "./recorder.js";
 
 //#region src/messages.ts
+/**
+* Non-empty placeholder signature written into emitted `thinking` blocks.
+*
+* The real Anthropic signature is a cryptographic value aimock cannot
+* reproduce, but extended-thinking invariant (b) requires a non-empty
+* `signature` on the leading thinking block of a tool-loop continuation turn.
+* Emitting "" would make a record→replay round-trip of an aimock thinking turn
+* self-trip that invariant under strict mode. A non-empty placeholder keeps
+* round-trips green; the invariant only checks for non-emptiness, not value.
+*/
+const PLACEHOLDER_SIGNATURE = "aimock-placeholder-signature";
 function extractClaudeTextContent(content) {
 	if (typeof content === "string") return content;
-	return content.filter((b) => b.type === "text").map((b) => b.text ?? "").join("");
+	if (!Array.isArray(content)) return "";
+	return content.filter((b) => b != null && typeof b === "object" && b.type === "text").map((b) => b.text ?? "").join("");
 }
 function claudeToCompletionRequest(req) {
 	const messages = [];
 	if (req.system) {
-		const systemText = typeof req.system === "string" ? req.system : req.system.filter((b) => b.type === "text").map((b) => b.text ?? "").join("");
+		const systemText = typeof req.system === "string" ? req.system : Array.isArray(req.system) ? req.system.filter((b) => b != null && typeof b === "object" && b.type === "text").map((b) => b.text ?? "").join("") : "";
 		if (systemText) messages.push({
 			role: "system",
 			content: systemText
 		});
 	}
-	for (const msg of req.messages) if (msg.role === "user") {
-		if (typeof msg.content !== "string" && Array.isArray(msg.content)) {
-			const toolResults = msg.content.filter((b) => b.type === "tool_result");
-			const textBlocks = msg.content.filter((b) => b.type === "text");
-			if (toolResults.length > 0) {
-				for (const tr of toolResults) {
-					const resultContent = typeof tr.content === "string" ? tr.content : Array.isArray(tr.content) ? tr.content.filter((b) => b.type === "text").map((b) => b.text ?? "").join("") : "";
-					messages.push({
-						role: "tool",
-						content: resultContent,
-						tool_call_id: tr.tool_use_id
+	const reqMessages = Array.isArray(req.messages) ? req.messages : [];
+	for (const msg of reqMessages) {
+		if (!msg || typeof msg !== "object") continue;
+		if (msg.role === "user") {
+			if (typeof msg.content !== "string" && Array.isArray(msg.content)) {
+				const blocks = msg.content.filter((b) => b != null && typeof b === "object");
+				const toolResults = blocks.filter((b) => b.type === "tool_result");
+				const textBlocks = blocks.filter((b) => b.type === "text");
+				if (toolResults.length > 0) {
+					for (const tr of toolResults) {
+						const resultContent = typeof tr.content === "string" ? tr.content : Array.isArray(tr.content) ? tr.content.filter((b) => b != null && typeof b === "object" && b.type === "text").map((b) => b.text ?? "").join("") : "";
+						messages.push({
+							role: "tool",
+							content: resultContent,
+							tool_call_id: tr.tool_use_id
+						});
+					}
+					if (textBlocks.length > 0) messages.push({
+						role: "user",
+						content: textBlocks.map((b) => b.text ?? "").join("")
 					});
+					continue;
 				}
-				if (textBlocks.length > 0) messages.push({
-					role: "user",
-					content: textBlocks.map((b) => b.text ?? "").join("")
-				});
-				continue;
 			}
-		}
-		messages.push({
-			role: "user",
-			content: extractClaudeTextContent(msg.content)
-		});
-	} else if (msg.role === "assistant") if (typeof msg.content === "string") messages.push({
-		role: "assistant",
-		content: msg.content
-	});
-	else if (Array.isArray(msg.content)) {
-		const toolUseBlocks = msg.content.filter((b) => b.type === "tool_use");
-		const textContent = extractClaudeTextContent(msg.content);
-		if (toolUseBlocks.length > 0) messages.push({
+			messages.push({
+				role: "user",
+				content: extractClaudeTextContent(msg.content)
+			});
+		} else if (msg.role === "assistant") if (typeof msg.content === "string") messages.push({
 			role: "assistant",
-			content: textContent || null,
-			tool_calls: toolUseBlocks.map((b) => ({
-				id: b.id ?? generateToolUseId(),
-				type: "function",
-				function: {
-					name: b.name ?? "",
-					arguments: typeof b.input === "string" ? b.input : JSON.stringify(b.input ?? {})
-				}
-			}))
+			content: msg.content
 		});
-		else messages.push({
+		else if (Array.isArray(msg.content)) {
+			const toolUseBlocks = msg.content.filter((b) => b != null && typeof b === "object" && b.type === "tool_use");
+			const textContent = extractClaudeTextContent(msg.content);
+			if (toolUseBlocks.length > 0) messages.push({
+				role: "assistant",
+				content: textContent || null,
+				tool_calls: toolUseBlocks.map((b) => ({
+					id: b.id ?? generateToolUseId(),
+					type: "function",
+					function: {
+						name: b.name ?? "",
+						arguments: typeof b.input === "string" ? b.input : JSON.stringify(b.input ?? {})
+					}
+				}))
+			});
+			else messages.push({
+				role: "assistant",
+				content: textContent || null
+			});
+		} else messages.push({
 			role: "assistant",
-			content: textContent || null
+			content: null
 		});
-	} else messages.push({
-		role: "assistant",
-		content: null
-	});
+	}
 	let tools;
 	if (req.tools && req.tools.length > 0) tools = req.tools.map((t) => ({
 		type: "function",
@@ -89,6 +106,88 @@ function claudeToCompletionRequest(req) {
 		_endpointType: "chat"
 	};
 }
+/** True iff `req.thinking` is an object with `type === "enabled"`. */
+function isThinkingEnabled(req) {
+	const t = req.thinking;
+	return typeof t === "object" && t !== null && t.type === "enabled";
+}
+/**
+* True iff the user turn at `req.messages[userIndex]` carries a `tool_result`
+* referencing one of `toolUseIds` — i.e. it answers the preceding assistant
+* turn's `tool_use` block(s), making that assistant turn a tool-loop
+* continuation point.
+*/
+function userTurnAnswersToolUse(msg, toolUseIds) {
+	if (!msg || msg.role !== "user" || !Array.isArray(msg.content)) return false;
+	return msg.content.some((b) => b != null && typeof b === "object" && b.type === "tool_result" && typeof b.tool_use_id === "string" && toolUseIds.has(b.tool_use_id));
+}
+/**
+* Validate the Anthropic extended-thinking request invariants on tool-loop
+* continuation turns. Pure: returns the first detected `ThinkingViolation` or
+* `null` when thinking is disabled or every in-scope turn is well-formed.
+*
+* Scope: an assistant turn is in-scope only when it (a) is array content
+* carrying at least one `tool_use` block and (b) is followed by a matching
+* `tool_result` on the next user turn. Text-only / `end_turn` turns, string or
+* empty turns, and trailing unanswered `tool_use` turns are all exempt.
+*
+* Scope detection assumes well-formed Anthropic transcripts: `tool_use` ids are
+* unique, and a tool_use turn is answered by the immediately-following user
+* turn's `tool_result` (adjacency). Malformed shapes — non-adjacent answers,
+* idless `tool_use` blocks, or a `tool_result` separated from its `tool_use` by
+* intervening turns — are intentionally treated as out-of-scope (and therefore
+* not 400'd) rather than validated, since the real Anthropic API would never
+* have produced them.
+*/
+function validateThinkingInvariants(req) {
+	if (!isThinkingEnabled(req)) return null;
+	const messages = Array.isArray(req.messages) ? req.messages : [];
+	for (let i = 0; i < messages.length; i++) {
+		const msg = messages[i];
+		if (!msg || typeof msg !== "object") continue;
+		if (msg.role !== "assistant" || !Array.isArray(msg.content) || msg.content.length === 0) continue;
+		const toolUseIds = /* @__PURE__ */ new Set();
+		for (const b of msg.content) {
+			if (!b || typeof b !== "object") continue;
+			if (b.type === "tool_use" && typeof b.id === "string") toolUseIds.add(b.id);
+		}
+		if (toolUseIds.size === 0) continue;
+		if (!userTurnAnswersToolUse(messages[i + 1], toolUseIds)) continue;
+		const first = msg.content[0];
+		if (!first || typeof first !== "object") return {
+			kind: "missing_thinking_first",
+			messageIndex: i,
+			observedFirstBlockType: void 0
+		};
+		if (first.type !== "thinking" && first.type !== "redacted_thinking") return {
+			kind: "missing_thinking_first",
+			messageIndex: i,
+			observedFirstBlockType: first.type
+		};
+		if (first.type === "thinking") {
+			if (typeof first.signature !== "string" || first.signature.length === 0) return {
+				kind: "missing_signature",
+				messageIndex: i
+			};
+		}
+		if (first.type === "redacted_thinking") {
+			if (typeof first.data !== "string" || first.data.length === 0) return {
+				kind: "dropped_redacted_thinking",
+				messageIndex: i
+			};
+		}
+	}
+	return null;
+}
+/** Render the Anthropic-shaped 400 error message for a thinking violation. */
+function thinkingViolationMessage(v) {
+	const prefix = `messages.${v.messageIndex}.content.0`;
+	switch (v.kind) {
+		case "missing_thinking_first": return `${prefix}: when \`thinking\` is enabled, a tool-loop continuation assistant turn must begin with a \`thinking\` block; got \`${v.observedFirstBlockType ?? "unknown"}\`.`;
+		case "missing_signature": return `${prefix}: the leading \`thinking\` block is missing a non-empty \`signature\`.`;
+		case "dropped_redacted_thinking": return `${prefix}: the leading \`redacted_thinking\` block must preserve its \`data\`.`;
+	}
+}
 function claudeStopReason(finishReason, defaultReason) {
 	if (!finishReason) return defaultReason;
 	if (finishReason === "stop") return "end_turn";
@@ -106,9 +205,44 @@ function claudeUsage(overrides) {
 		output_tokens: overrides.usage.output_tokens ?? overrides.usage.completion_tokens ?? 0
 	};
 }
-function buildClaudeTextStreamEvents(content, model, chunkSize, reasoning, overrides) {
+/**
+* Emit faithful Anthropic `redacted_thinking` content-block events at the head
+* of a streamed turn — one `content_block_start` (carrying the opaque `data`)
+* and `content_block_stop` per recorded block, with NO deltas (real Anthropic
+* delivers the encrypted reasoning entirely on the start event). Returns the
+* next free block index. When `redactedThinking` is empty/absent this is a
+* no-op, so no events are added.
+*
+* Leading placement keeps the turn round-trip-safe for the extended-thinking
+* leading-block invariant (a thinking / redacted_thinking block must come
+* first). It does NOT reproduce real Anthropic's stream order: aimock joins all
+* plaintext reasoning into one block and emits all redacted blocks ahead of it,
+* whereas real Anthropic interleaves `thinking` and `redacted_thinking` in the
+* order they occurred. The interleaving between the two is not preserved.
+*/
+function pushRedactedThinkingStreamEvents(events, startIndex, redactedThinking) {
+	let blockIndex = startIndex;
+	for (const data of redactedThinking ?? []) {
+		events.push({
+			type: "content_block_start",
+			index: blockIndex,
+			content_block: {
+				type: "redacted_thinking",
+				data
+			}
+		});
+		events.push({
+			type: "content_block_stop",
+			index: blockIndex
+		});
+		blockIndex++;
+	}
+	return blockIndex;
+}
+function buildClaudeTextStreamEvents(content, model, chunkSize, reasoning, overrides, reasoningSignature, redactedThinking) {
 	const msgId = overrides?.id ?? generateMessageId();
 	const effectiveModel = overrides?.model ?? model;
+	const signature = reasoningSignature ?? PLACEHOLDER_SIGNATURE;
 	const events = [];
 	events.push({
 		type: "message_start",
@@ -124,6 +258,7 @@ function buildClaudeTextStreamEvents(content, model, chunkSize, reasoning, overr
 		}
 	});
 	let blockIndex = 0;
+	blockIndex = pushRedactedThinkingStreamEvents(events, blockIndex, redactedThinking);
 	if (reasoning) {
 		events.push({
 			type: "content_block_start",
@@ -150,7 +285,7 @@ function buildClaudeTextStreamEvents(content, model, chunkSize, reasoning, overr
 			index: blockIndex,
 			delta: {
 				type: "signature_delta",
-				signature: ""
+				signature
 			}
 		});
 		events.push({
@@ -193,9 +328,10 @@ function buildClaudeTextStreamEvents(content, model, chunkSize, reasoning, overr
 	events.push({ type: "message_stop" });
 	return events;
 }
-function buildClaudeToolCallStreamEvents(toolCalls, model, chunkSize, logger, overrides) {
+function buildClaudeToolCallStreamEvents(toolCalls, model, chunkSize, logger, reasoning, overrides, reasoningSignature, redactedThinking) {
 	const msgId = overrides?.id ?? generateMessageId();
 	const effectiveModel = overrides?.model ?? model;
+	const signature = reasoningSignature ?? PLACEHOLDER_SIGNATURE;
 	const events = [];
 	events.push({
 		type: "message_start",
@@ -210,8 +346,44 @@ function buildClaudeToolCallStreamEvents(toolCalls, model, chunkSize, logger, ov
 			usage: claudeUsage(overrides)
 		}
 	});
-	for (let idx = 0; idx < toolCalls.length; idx++) {
-		const tc = toolCalls[idx];
+	let blockIndex = 0;
+	blockIndex = pushRedactedThinkingStreamEvents(events, blockIndex, redactedThinking);
+	if (reasoning) {
+		events.push({
+			type: "content_block_start",
+			index: blockIndex,
+			content_block: {
+				type: "thinking",
+				thinking: "",
+				signature: ""
+			}
+		});
+		for (let i = 0; i < reasoning.length; i += chunkSize) {
+			const slice = reasoning.slice(i, i + chunkSize);
+			events.push({
+				type: "content_block_delta",
+				index: blockIndex,
+				delta: {
+					type: "thinking_delta",
+					thinking: slice
+				}
+			});
+		}
+		events.push({
+			type: "content_block_delta",
+			index: blockIndex,
+			delta: {
+				type: "signature_delta",
+				signature
+			}
+		});
+		events.push({
+			type: "content_block_stop",
+			index: blockIndex
+		});
+		blockIndex++;
+	}
+	for (const tc of toolCalls) {
 		const toolUseId = tc.id || generateToolUseId();
 		let argsObj;
 		try {
@@ -223,7 +395,7 @@ function buildClaudeToolCallStreamEvents(toolCalls, model, chunkSize, logger, ov
 		const argsJson = JSON.stringify(argsObj);
 		events.push({
 			type: "content_block_start",
-			index: idx,
+			index: blockIndex,
 			content_block: {
 				type: "tool_use",
 				id: toolUseId,
@@ -235,7 +407,7 @@ function buildClaudeToolCallStreamEvents(toolCalls, model, chunkSize, logger, ov
 			const slice = argsJson.slice(i, i + chunkSize);
 			events.push({
 				type: "content_block_delta",
-				index: idx,
+				index: blockIndex,
 				delta: {
 					type: "input_json_delta",
 					partial_json: slice
@@ -244,8 +416,9 @@ function buildClaudeToolCallStreamEvents(toolCalls, model, chunkSize, logger, ov
 		}
 		events.push({
 			type: "content_block_stop",
-			index: idx
+			index: blockIndex
 		});
+		blockIndex++;
 	}
 	events.push({
 		type: "message_delta",
@@ -258,12 +431,31 @@ function buildClaudeToolCallStreamEvents(toolCalls, model, chunkSize, logger, ov
 	events.push({ type: "message_stop" });
 	return events;
 }
-function buildClaudeTextResponse(content, model, reasoning, overrides) {
+/**
+* Push faithful Anthropic `redacted_thinking` content blocks (each a
+* `{ type: "redacted_thinking", data }`) onto a non-streaming content array, in
+* recorded order. No-op when `redactedThinking` is empty/absent, so no blocks
+* are added.
+*
+* Leading placement keeps the turn round-trip-safe for the extended-thinking
+* leading-block invariant. It does NOT reproduce real Anthropic's block order:
+* aimock joins all plaintext reasoning into one block and emits all redacted
+* blocks ahead of it, whereas real Anthropic interleaves `thinking` and
+* `redacted_thinking` in occurrence order. The interleaving is not preserved.
+*/
+function pushRedactedThinkingBlocks(contentBlocks, redactedThinking) {
+	for (const data of redactedThinking ?? []) contentBlocks.push({
+		type: "redacted_thinking",
+		data
+	});
+}
+function buildClaudeTextResponse(content, model, reasoning, overrides, reasoningSignature, redactedThinking) {
 	const contentBlocks = [];
+	pushRedactedThinkingBlocks(contentBlocks, redactedThinking);
 	if (reasoning) contentBlocks.push({
 		type: "thinking",
 		thinking: reasoning,
-		signature: ""
+		signature: reasoningSignature ?? PLACEHOLDER_SIGNATURE
 	});
 	contentBlocks.push({
 		type: "text",
@@ -280,35 +472,44 @@ function buildClaudeTextResponse(content, model, reasoning, overrides) {
 		usage: claudeUsage(overrides)
 	};
 }
-function buildClaudeToolCallResponse(toolCalls, model, logger, overrides) {
+function buildClaudeToolCallResponse(toolCalls, model, logger, reasoning, overrides, reasoningSignature, redactedThinking) {
+	const contentBlocks = [];
+	pushRedactedThinkingBlocks(contentBlocks, redactedThinking);
+	if (reasoning) contentBlocks.push({
+		type: "thinking",
+		thinking: reasoning,
+		signature: reasoningSignature ?? PLACEHOLDER_SIGNATURE
+	});
+	for (const tc of toolCalls) {
+		let argsObj;
+		try {
+			argsObj = JSON.parse(tc.arguments || "{}");
+		} catch {
+			logger.warn(`Malformed JSON in fixture tool call arguments for "${tc.name}": ${tc.arguments}`);
+			argsObj = {};
+		}
+		contentBlocks.push({
+			type: "tool_use",
+			id: tc.id || generateToolUseId(),
+			name: tc.name,
+			input: argsObj
+		});
+	}
 	return {
 		id: overrides?.id ?? generateMessageId(),
 		type: "message",
 		role: overrides?.role ?? "assistant",
-		content: toolCalls.map((tc) => {
-			let argsObj;
-			try {
-				argsObj = JSON.parse(tc.arguments || "{}");
-			} catch {
-				logger.warn(`Malformed JSON in fixture tool call arguments for "${tc.name}": ${tc.arguments}`);
-				argsObj = {};
-			}
-			return {
-				type: "tool_use",
-				id: tc.id || generateToolUseId(),
-				name: tc.name,
-				input: argsObj
-			};
-		}),
+		content: contentBlocks,
 		model: overrides?.model ?? model,
 		stop_reason: claudeStopReason(overrides?.finishReason, "tool_use"),
 		stop_sequence: null,
 		usage: claudeUsage(overrides)
 	};
 }
-function buildClaudeContentWithToolCallsStreamEvents(content, toolCalls, model, chunkSize, logger, reasoning, overrides) {
+function buildClaudeContentWithToolCallsStreamEvents(content, toolCalls, model, chunkSize, logger, reasoning, overrides, reasoningSignature, redactedThinking) {
 	const msgId = overrides?.id ?? generateMessageId();
 	const effectiveModel = overrides?.model ?? model;
+	const signature = reasoningSignature ?? PLACEHOLDER_SIGNATURE;
 	const events = [];
 	events.push({
 		type: "message_start",
@@ -324,6 +525,7 @@ function buildClaudeContentWithToolCallsStreamEvents(content, toolCalls, model,
 		}
 	});
 	let blockIndex = 0;
+	blockIndex = pushRedactedThinkingStreamEvents(events, blockIndex, redactedThinking);
 	if (reasoning) {
 		events.push({
 			type: "content_block_start",
@@ -350,7 +552,7 @@ function buildClaudeContentWithToolCallsStreamEvents(content, toolCalls, model,
 			index: blockIndex,
 			delta: {
 				type: "signature_delta",
-				signature: ""
+				signature
 			}
 		});
 		events.push({
@@ -431,12 +633,13 @@ function buildClaudeContentWithToolCallsStreamEvents(content, toolCalls, model,
 	events.push({ type: "message_stop" });
 	return events;
 }
-function buildClaudeContentWithToolCallsResponse(content, toolCalls, model, logger, reasoning, overrides) {
+function buildClaudeContentWithToolCallsResponse(content, toolCalls, model, logger, reasoning, overrides, reasoningSignature, redactedThinking) {
 	const contentBlocks = [];
+	pushRedactedThinkingBlocks(contentBlocks, redactedThinking);
 	if (reasoning) contentBlocks.push({
 		type: "thinking",
 		thinking: reasoning,
-		signature: ""
+		signature: reasoningSignature ?? PLACEHOLDER_SIGNATURE
 	});
 	contentBlocks.push({
 		type: "text",
@@ -517,10 +720,38 @@ async function handleMessages(req, res, raw, fixtures, journal, defaults, setCor
 		} }));
 		return;
 	}
+	const thinkingViolation = validateThinkingInvariants(claudeReq);
+	if (thinkingViolation) {
+		const effectiveStrict = resolveStrictMode(defaults.strict, req.headers);
+		const violationMessage = thinkingViolationMessage(thinkingViolation);
+		if (effectiveStrict) {
+			logger.error(`THINKING: ${violationMessage}`);
+			journal.add({
+				method: req.method ?? "POST",
+				path: req.url ?? "/v1/messages",
+				headers: flattenHeaders(req.headers),
+				body: null,
+				response: {
+					status: 400,
+					fixture: null,
+					...strictOverrideField(defaults.strict, req.headers)
+				}
+			});
+			writeErrorResponse(res, 400, JSON.stringify({
+				type: "error",
+				error: {
+					type: "invalid_request_error",
+					message: violationMessage
+				}
+			}));
+			return;
+		}
+		logger.warn(`THINKING: ${violationMessage} (strict off — replaying anyway)`);
+	}
 	const completionReq = claudeToCompletionRequest(claudeReq);
 	completionReq._context = getContext(req);
 	const testId = getTestId(req);
-	const fixture = matchFixture(fixtures, completionReq, journal.getFixtureMatchCountsForTest(testId), defaults.requestTransform);
+	const { fixture, skippedBySequenceOrTurn } = matchFixtureDiagnostic(fixtures, completionReq, journal.getFixtureMatchCountsForTest(testId), defaults.requestTransform);
 	if (fixture) {
 		journal.incrementFixtureMatchCount(fixture, fixtures, testId);
 		logger.debug(`Fixture matched: ${JSON.stringify(fixture.match).slice(0, 120)}`);
@@ -538,8 +769,8 @@ async function handleMessages(req, res, raw, fixtures, journal, defaults, setCor
 	if (!fixture) {
 		if (resolveStrictMode(defaults.strict, req.headers)) {
 			const strictStatus = 503;
-			const strictMessage = "Strict mode: no fixture matched";
-			logger.error(`STRICT: No fixture matched for ${req.method ?? "POST"} ${req.url ?? "/v1/messages"}`);
+			const strictMessage = strictNoMatchMessage(skippedBySequenceOrTurn);
+			logger.error(strictNoMatchLogLine(req.method ?? "POST", req.url ?? "/v1/messages", skippedBySequenceOrTurn));
 			journal.add({
 				method: req.method ?? "POST",
 				path: req.url ?? "/v1/messages",
@@ -620,6 +851,9 @@ async function handleMessages(req, res, raw, fixtures, journal, defaults, setCor
 	if (isContentWithToolCallsResponse(response)) {
 		if (response.webSearches?.length) logger.warn("webSearches in fixture response are not supported for Claude Messages API — ignoring");
 		const overrides = extractOverrides(response);
+		const effectiveStrict = resolveStrictMode(defaults.strict, req.headers);
+		const effReasoning = resolveReasoningForModel(response.reasoning, completionReq.model, effectiveStrict, defaults.logger);
+		const { reasoningSignature: effReasoningSignature, redactedThinking: effRedactedThinking } = resolveReasoningArtifactsForModel(response.reasoningSignature, response.redactedThinking, completionReq.model, effectiveStrict, defaults.logger);
 		const journalEntry = journal.add({
 			method: req.method ?? "POST",
 			path: req.url ?? "/v1/messages",
@@ -631,11 +865,11 @@ async function handleMessages(req, res, raw, fixtures, journal, defaults, setCor
 			}
 		});
 		if (claudeReq.stream !== true) {
-			const body = buildClaudeContentWithToolCallsResponse(response.content, response.toolCalls, completionReq.model, logger, response.reasoning, overrides);
+			const body = buildClaudeContentWithToolCallsResponse(response.content, response.toolCalls, completionReq.model, logger, effReasoning, overrides, effReasoningSignature, effRedactedThinking);
 			res.writeHead(200, { "Content-Type": "application/json" });
 			res.end(JSON.stringify(body));
 		} else {
-			const events = buildClaudeContentWithToolCallsStreamEvents(response.content, response.toolCalls, completionReq.model, chunkSize, logger, response.reasoning, overrides);
+			const events = buildClaudeContentWithToolCallsStreamEvents(response.content, response.toolCalls, completionReq.model, chunkSize, logger, effReasoning, overrides, effReasoningSignature, effRedactedThinking);
 			const interruption = createInterruptionSignal(fixture);
 			if (!await writeClaudeSSEStream(res, events, {
 				latency,
@@ -656,6 +890,9 @@ async function handleMessages(req, res, raw, fixtures, journal, defaults, setCor
 	if (isTextResponse(response)) {
 		if (response.webSearches?.length) logger.warn("webSearches in fixture response are not supported for Claude Messages API — ignoring");
 		const overrides = extractOverrides(response);
+		const effectiveStrict = resolveStrictMode(defaults.strict, req.headers);
+		const effReasoning = resolveReasoningForModel(response.reasoning, completionReq.model, effectiveStrict, defaults.logger);
+		const { reasoningSignature: effReasoningSignature, redactedThinking: effRedactedThinking } = resolveReasoningArtifactsForModel(response.reasoningSignature, response.redactedThinking, completionReq.model, effectiveStrict, defaults.logger);
 		const journalEntry = journal.add({
 			method: req.method ?? "POST",
 			path: req.url ?? "/v1/messages",
@@ -667,11 +904,11 @@ async function handleMessages(req, res, raw, fixtures, journal, defaults, setCor
 			}
 		});
 		if (claudeReq.stream !== true) {
-			const body = buildClaudeTextResponse(response.content, completionReq.model, response.reasoning, overrides);
+			const body = buildClaudeTextResponse(response.content, completionReq.model, effReasoning, overrides, effReasoningSignature, effRedactedThinking);
 			res.writeHead(200, { "Content-Type": "application/json" });
 			res.end(JSON.stringify(body));
 		} else {
-			const events = buildClaudeTextStreamEvents(response.content, completionReq.model, chunkSize, response.reasoning, overrides);
+			const events = buildClaudeTextStreamEvents(response.content, completionReq.model, chunkSize, effReasoning, overrides, effReasoningSignature, effRedactedThinking);
 			const interruption = createInterruptionSignal(fixture);
 			if (!await writeClaudeSSEStream(res, events, {
 				latency,
@@ -692,6 +929,9 @@ async function handleMessages(req, res, raw, fixtures, journal, defaults, setCor
 	if (isToolCallResponse(response)) {
 		if (response.webSearches?.length) logger.warn("webSearches in fixture response are not supported for Claude Messages API — ignoring");
 		const overrides = extractOverrides(response);
+		const effectiveStrict = resolveStrictMode(defaults.strict, req.headers);
+		const effReasoning = resolveReasoningForModel(response.reasoning, completionReq.model, effectiveStrict, defaults.logger);
+		const { reasoningSignature: effReasoningSignature, redactedThinking: effRedactedThinking } = resolveReasoningArtifactsForModel(response.reasoningSignature, response.redactedThinking, completionReq.model, effectiveStrict, defaults.logger);
 		const journalEntry = journal.add({
 			method: req.method ?? "POST",
 			path: req.url ?? "/v1/messages",
@@ -703,11 +943,11 @@ async function handleMessages(req, res, raw, fixtures, journal, defaults, setCor
 			}
 		});
 		if (claudeReq.stream !== true) {
-			const body = buildClaudeToolCallResponse(response.toolCalls, completionReq.model, logger, overrides);
+			const body = buildClaudeToolCallResponse(response.toolCalls, completionReq.model, logger, effReasoning, overrides, effReasoningSignature, effRedactedThinking);
 			res.writeHead(200, { "Content-Type": "application/json" });
 			res.end(JSON.stringify(body));
 		} else {
-			const events = buildClaudeToolCallStreamEvents(response.toolCalls, completionReq.model, chunkSize, logger, overrides);
+			const events = buildClaudeToolCallStreamEvents(response.toolCalls, completionReq.model, chunkSize, logger, effReasoning, overrides, effReasoningSignature, effRedactedThinking);
 			const interruption = createInterruptionSignal(fixture);
 			if (!await writeClaudeSSEStream(res, events, {
 				latency,