fix(kiro): improve direct session cache reuse

Reshape Kiro direct requests so resumed client sessions reuse Kiro's
cache-affinity fields instead of starting unrelated CodeWhisperer
conversations.

- keep conversationState.conversationId stable when the client sends an
  explicit session id (x-session-id, session_id, conversation_id, Claude
  Code session metadata)
- add a stable conversationState.agentContinuationId per Kiro session
- send conversationState.agentTaskType: "vibe" and agentMode: "vibe",
  matching the normal Kiro CLI/KAS chat path
- move Kiro thinking instructions into Kiro-compatible systemPrompt /
  additionalModelRequestFields instead of generic top-level thinking
- keep volatile timestamp context out of the top-level systemPrompt; it
  remains only in user content fallback
- suppress additionalModelRequestFields for legacy 4.5-era Claude/Kiro
  models that reject it, while defaulting future Claude/Kiro model ids
  to supported
- preserve Kiro meteringEvent credit usage internally for accounting
  without leaking provider-specific fields into OpenAI-compatible usage
- prevent unrelated headerless Kiro requests from sharing one
  connection-wide continuation
- cap/evict continuation sessions so long-running processes do not grow
  the continuation map unbounded
- treat generated headerless Kiro sessions as one-shot so they do not
  evict real explicit-session continuations
- keep credit-only Kiro metering valid for internal persistence when
  token metrics are unavailable

tests/translator/claude-kiro-direct.test.js

@@ -6,8 +6,8 @@ import "./registerAll.js";
 import { translateRequest, translateResponse } from "../../open-sse/translator/index.js";
 import { FORMATS } from "../../open-sse/translator/formats.js";
 
-const C2K = (body) =>
-  translateRequest(FORMATS.CLAUDE, FORMATS.KIRO, "claude-sonnet-4.5", body, true, null, "kiro");
+const C2K = (body, credentials = null, model = "claude-sonnet-4.5") =>
+  translateRequest(FORMATS.CLAUDE, FORMATS.KIRO, model, body, true, credentials, "kiro");
 
 describe("Claude → Kiro (direct route)", () => {
   it("produces a Kiro conversationState payload", () => {
@@ -16,6 +16,27 @@ describe("Claude → Kiro (direct route)", () => {
     expect(out.conversationState.currentMessage.userInputMessage.content).toContain("hello");
   });
 
+  it("keeps conversationId stable from client session headers and replays frozen msg0", () => {
+    const credentials = {
+      rawHeaders: { "x-session-id": "hermes-session-123-claude-replay" },
+      connectionId: "kiro-account-1",
+    };
+    const first = C2K({ messages: [{ role: "user", content: "first" }] }, credentials);
+    const second = C2K({ messages: [{ role: "user", content: "second" }] }, credentials);
+
+    expect(first.conversationState.conversationId).toBe("hermes-session-123-claude-replay");
+    expect(second.conversationState.conversationId).toBe("hermes-session-123-claude-replay");
+    expect(first.conversationState.agentContinuationId).toBeTruthy();
+    expect(second.conversationState.agentContinuationId).toBe(first.conversationState.agentContinuationId);
+    expect(first.conversationState.agentTaskType).toBe("vibe");
+    expect(second.conversationState.history[0].userInputMessage.content).toBe(
+      first.conversationState.currentMessage.userInputMessage.content
+    );
+    expect(second.conversationState.history[0].userInputMessage.modelId).toBe("claude-sonnet-4.5");
+    expect(second.conversationState.currentMessage.userInputMessage.content).toContain("Current time");
+    expect(second.conversationState.currentMessage.userInputMessage.content).toContain("second");
+  });
+
   it("guard 1: with no tools, a dangling tool_result is flattened to text (no structured ref)", () => {
     // Client omitted `tools` but kept a tool_result after compaction.
     const out = C2K({
@@ -60,20 +81,60 @@ describe("Claude → Kiro (direct route)", () => {
       null,
       "kiro"
     );
-    expect(out.conversationState.currentMessage.userInputMessage.content).toContain(
+    expect(out.systemPrompt).toContain(
       "<thinking_mode>enabled</thinking_mode>"
     );
+    expect(out.agentMode).toBe("vibe");
   });
 
-  it("maps output_config.effort high to Kiro max_thinking_length 24576", () => {
+  it("does not send additionalModelRequestFields for Kiro models without effort support", () => {
     const out = C2K({
       output_config: { effort: "high" },
       messages: [{ role: "user", content: "think with adaptive effort" }],
     });
 
-    expect(out.conversationState.currentMessage.userInputMessage.content).toContain(
-      "<max_thinking_length>24576</max_thinking_length>"
-    );
+    expect(out.additionalModelRequestFields).toBeUndefined();
+    expect(out.thinking).toBeUndefined();
+    expect(out.systemPrompt).toContain("<max_thinking_length>24576</max_thinking_length>");
+  });
+
+  it("maps output_config.effort high to Kiro CLI-style additionalModelRequestFields for effort models", () => {
+    const out = C2K({
+      output_config: { effort: "high" },
+      messages: [{ role: "user", content: "think with adaptive effort" }],
+    }, null, "claude-sonnet-5");
+
+    expect(out.additionalModelRequestFields).toEqual({
+      thinking: { type: "adaptive", display: "summarized" },
+      output_config: { effort: "high" },
+    });
+    expect(out.thinking).toBeUndefined();
+    expect(out.systemPrompt).toContain("<max_thinking_length>24576</max_thinking_length>");
+  });
+
+  it("sends Claude system as top-level systemPrompt and keeps a user-content fallback", () => {
+    const out = C2K({
+      system: "system-only instruction",
+      messages: [{ role: "user", content: "hello" }],
+    });
+
+    expect(out.systemPrompt).toContain("system-only instruction");
+    expect(out.conversationState.currentMessage.userInputMessage.content).toContain("system-only instruction");
+  });
+
+  it("keeps top-level systemPrompt stable across turns", () => {
+    const first = C2K({
+      system: "stable instruction",
+      messages: [{ role: "user", content: "first" }],
+    });
+    const second = C2K({
+      system: "stable instruction",
+      messages: [{ role: "user", content: "second" }],
+    });
+
+    expect(first.systemPrompt).toBe(second.systemPrompt);
+    expect(first.systemPrompt).not.toContain("Current time");
+    expect(first.conversationState.currentMessage.userInputMessage.content).toContain("Current time");
   });
 });