feat(core): persist v2 session context epochs (#30789)

packages/llm/src/protocols/anthropic-messages.ts

@@ -351,38 +351,29 @@ const endsInServerToolUse = (message: LLMRequest["messages"][number]) => {
   return message.role === "assistant" && last?.type === "tool-call" && last.providerExecuted === true
 }
 
-const endsInLocalToolUse = (message: LLMRequest["messages"][number]) => {
-  const last = message.content.at(-1)
-  return message.role === "assistant" && last?.type === "tool-call" && last.providerExecuted !== true
-}
-
-const validateNativeSystemUpdate = Effect.fn("AnthropicMessages.validateNativeSystemUpdate")(function* (
-  messages: LLMRequest["messages"],
-  index: number,
-) {
+const canUseNativeSystemUpdate = (messages: LLMRequest["messages"], index: number) => {
   const previous = messages[index - 1]
   const next = messages[index + 1]
-  if (!previous)
-    return yield* invalid(
-      "Anthropic Messages chronological system updates cannot be the first message; use LLMRequest.system",
-    )
-  if (previous.role === "system")
-    return yield* invalid("Anthropic Messages chronological system updates cannot be consecutive")
-  if (endsInLocalToolUse(previous))
-    return yield* invalid(
-      "Anthropic Messages chronological system updates cannot appear between a local tool call and its tool result",
-    )
-  if (previous.role !== "user" && previous.role !== "tool" && !endsInServerToolUse(previous))
-    return yield* invalid(
-      "Anthropic Messages chronological system updates must follow a user message, tool result, or assistant server tool use",
-    )
-  if (next?.role === "system")
-    return yield* invalid("Anthropic Messages chronological system updates cannot be consecutive")
-  if (next && next.role !== "assistant")
-    return yield* invalid(
-      "Anthropic Messages chronological system updates must end the messages array or immediately precede an assistant message",
-    )
-})
+  return (
+    previous !== undefined &&
+    previous.role !== "system" &&
+    (previous.role === "user" || previous.role === "tool" || endsInServerToolUse(previous)) &&
+    next?.role !== "system" &&
+    (next === undefined || next.role === "assistant")
+  )
+}
+
+const splitsLocalToolResults = (messages: LLMRequest["messages"], index: number) => {
+  const pending = new Set<string>()
+  for (const message of messages.slice(0, index)) {
+    for (const part of message.content) {
+      if (message.role === "assistant" && part.type === "tool-call" && part.providerExecuted !== true)
+        pending.add(part.id)
+      if (message.role === "tool" && part.type === "tool-result") pending.delete(part.id)
+    }
+  }
+  return pending.size > 0
+}
 
 const lowerNativeSystemUpdate = Effect.fn("AnthropicMessages.lowerNativeSystemUpdate")(function* (
   message: LLMRequest["messages"][number],
@@ -407,8 +398,9 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
 
   for (const [index, message] of request.messages.entries()) {
     if (message.role === "system") {
-      if (supportsNativeSystemUpdates(request)) {
-        yield* validateNativeSystemUpdate(request.messages, index)
+      if (splitsLocalToolResults(request.messages, index))
+        return yield* invalid("Anthropic Messages system updates cannot split a local tool call from its tool result")
+      if (supportsNativeSystemUpdates(request) && canUseNativeSystemUpdate(request.messages, index)) {
         messages.push(yield* lowerNativeSystemUpdate(message, breakpoints))
         continue
       }

packages/llm/test/provider/anthropic-messages.test.ts

@@ -125,26 +125,65 @@ describe("Anthropic Messages route", () => {
     }),
   )
 
-  it.effect("rejects invalid native chronological system update placement", () =>
+  it.effect("falls back for unsupported native chronological system update placement", () =>
     Effect.gen(function* () {
-      const placementError = (messages: Parameters<typeof LLM.request>[0]["messages"]) =>
-        LLMClient.prepare(LLM.request({ model: opus48, messages, cache: "none" })).pipe(Effect.flip)
+      expect(
+        (yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+          LLM.request({
+            model: opus48,
+            messages: [Message.assistant("Plain."), Message.system("After plain assistant.")],
+            cache: "none",
+          }),
+        )).body.messages,
+      ).toEqual([
+        { role: "assistant", content: [{ type: "text", text: "Plain." }] },
+        {
+          role: "user",
+          content: [{ type: "text", text: "<system-update>\nAfter plain assistant.\n</system-update>" }],
+        },
+      ])
+      expect(
+        (yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+          LLM.request({ model: opus48, messages: [Message.system("First.")], cache: "none" }),
+        )).body.messages,
+      ).toEqual([{ role: "user", content: [{ type: "text", text: "<system-update>\nFirst.\n</system-update>" }] }])
+      expect(
+        (yield* LLMClient.prepare<AnthropicMessages.AnthropicMessagesBody>(
+          LLM.request({
+            model: opus48,
+            messages: [Message.user("Before."), Message.system("One."), Message.system("Two.")],
+            cache: "none",
+          }),
+        )).body.messages,
+      ).toEqual([
+        {
+          role: "user",
+          content: [
+            { type: "text", text: "Before." },
+            { type: "text", text: "<system-update>\nOne.\n</system-update>" },
+            { type: "text", text: "<system-update>\nTwo.\n</system-update>" },
+          ],
+        },
+      ])
+    }),
+  )
 
-      expect((yield* placementError([Message.system("First.")])).message).toContain("cannot be the first message")
-      expect(
-        (yield* placementError([Message.user("Before."), Message.system("One."), Message.system("Two.")])).message,
-      ).toContain("cannot be consecutive")
-      expect(
-        (yield* placementError([Message.assistant("Plain."), Message.system("After plain assistant.")])).message,
-      ).toContain("must follow a user message, tool result, or assistant server tool use")
-      expect(
-        (yield* placementError([
-          Message.user("Use the tool."),
-          Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
-          Message.system("Too early."),
-          Message.tool({ id: "call_1", name: "lookup", result: "Done." }),
-        ])).message,
-      ).toContain("cannot appear between a local tool call and its tool result")
+  it.effect("rejects a system update between a local tool call and its result", () =>
+    Effect.gen(function* () {
+      const error = yield* LLMClient.prepare(
+        LLM.request({
+          model: opus48,
+          messages: [
+            Message.user("Use the tool."),
+            Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
+            Message.system("Too early."),
+            Message.tool({ id: "call_1", name: "lookup", result: "Done." }),
+          ],
+          cache: "none",
+        }),
+      ).pipe(Effect.flip)
+
+      expect(error.message).toContain("system updates cannot split a local tool call from its tool result")
     }),
   )
 

packages/llm/test/provider/golden.recorded.test.ts

@@ -123,7 +123,11 @@ describeRecordedGoldenScenarios([
     prefix: "gemini",
     model: gemini,
     requires: ["GOOGLE_GENERATIVE_AI_API_KEY"],
-    scenarios: [{ id: "text", maxTokens: 80 }, "tool-call", { id: "image", maxTokens: 160 }],
+    scenarios: [
+      { id: "text", maxTokens: 80 },
+      "tool-call",
+      { id: "image", maxTokens: 160 },
+    ],
   },
   {
     name: "xAI Grok 3 Mini",

packages/llm/test/recorded-scenarios.ts

@@ -1,5 +1,5 @@
 import { expect } from "bun:test"
-import { Effect, Schema, Stream } from "effect"
+import { Effect, Schema } from "effect"
 import {
   LLM,
   LLMEvent,

packages/llm/test/schema.test.ts

@@ -2,7 +2,15 @@ import { describe, expect, test } from "bun:test"
 import { Schema } from "effect"
 import * as OpenAIChat from "../src/protocols/openai-chat"
 import * as OpenAIResponses from "../src/protocols/openai-responses"
-import { ContentPart, LLMEvent, LLMRequest, Model, ModelID, ProviderID, Usage } from "../src/schema"
+import {
+  ContentPart,
+  LLMEvent,
+  LLMRequest,
+  Model,
+  ModelID,
+  ProviderID,
+  Usage,
+} from "../src/schema"
 import { ProviderShared } from "../src/protocols/shared"
 
 const model = new Model({
@@ -43,17 +51,6 @@ describe("llm schema", () => {
     expect(decoded.model.route.id).toBe("openai-responses")
   })
 
-  test("decodes chronological system messages", () => {
-    const decoded = decodeLLMRequest({
-      model,
-      system: [],
-      messages: [{ role: "system", content: [{ type: "text", text: "Operator update." }] }],
-      tools: [],
-    })
-
-    expect(decoded.messages[0]).toMatchObject({ role: "system", content: [{ type: "text", text: "Operator update." }] })
-  })
-
   test("rejects invalid event type", () => {
     expect(() => decodeLLMEvent({ type: "bogus" })).toThrow()
   })