From 26e5efbbdf6739eafa782cb98225e3fc2fbea99c Mon Sep 17 00:00:00 2001
From: Luis Pater <webmaster@idotorg.org>
Date: Sun, 4 Oct 2026 21:00:20 +0800
Subject: [PATCH] fix(claude): reserve cache breakpoint for thread continuation

- Decrement `maxBlocks` when a `thread` field is present in payload to reserve a cache breakpoint for Anthropic thread continuation

Closes: #6338
---
 .../executor/claude_executor_cloaking.go      |  6 ++++++
 .../runtime/executor/claude_executor_test.go  | 21 +++++++++++++++++++
 2 files changed, 27 insertions(+)

diff --git a/internal/runtime/executor/claude_executor_cloaking.go b/internal/runtime/executor/claude_executor_cloaking.go
index e482119ce..14bcd5b3d 100644
--- a/internal/runtime/executor/claude_executor_cloaking.go
+++ b/internal/runtime/executor/claude_executor_cloaking.go
@@ -1876,6 +1876,12 @@ func enforceCacheControlLimit(payload []byte, maxBlocks int) []byte {
 		return payload
 	}
 
+	thread := gjson.GetBytes(payload, "thread")
+	if thread.Exists() && thread.Type != gjson.Null && maxBlocks > 0 {
+		// Anthropic reserves one cache breakpoint for thread continuation.
+		maxBlocks--
+	}
+
 	total := countCacheControls(payload)
 	if total <= maxBlocks {
 		return payload
diff --git a/internal/runtime/executor/claude_executor_test.go b/internal/runtime/executor/claude_executor_test.go
index 666a63ab3..da361df6b 100644
--- a/internal/runtime/executor/claude_executor_test.go
+++ b/internal/runtime/executor/claude_executor_test.go
@@ -3240,6 +3240,27 @@ func TestEnforceCacheControlLimit_PreservesKeyOrderWhenModified(t *testing.T) {
 	}
 }
 
+func TestEnforceCacheControlLimit_ReservesThreadMarker(t *testing.T) {
+	payload := []byte(`{
+		"thread": {"type":"create"},
+		"tools": [{"name":"t1","cache_control":{"type":"ephemeral"}}],
+		"system": [
+			{"type":"text","text":"s1","cache_control":{"type":"ephemeral"}},
+			{"type":"text","text":"s2","cache_control":{"type":"ephemeral"}}
+		],
+		"messages": [{"role":"user","content":[
+			{"type":"text","text":"u1","cache_control":{"type":"ephemeral"}},
+			{"type":"text","text":"u2","cache_control":{"type":"ephemeral"}}
+		]}]
+	}`)
+
+	out := enforceCacheControlLimit(payload, 4)
+
+	if got := countCacheControls(out); got != 3 {
+		t.Fatalf("cache_control count = %d, want 3 when thread is present", got)
+	}
+}
+
 func TestEnforceCacheControlLimit_ToolOnlyPayloadStillRespectsLimit(t *testing.T) {
 	payload := []byte(`{
 		"tools": [