Merge pull request #2491 from mpfo0106/feature/claude-code-safe-alignment-sentinels

test(claude): add compatibility sentinels and centralize builtin fallback handling

internal/runtime/executor/claude_executor.go

@@ -951,12 +951,9 @@ func applyClaudeToolPrefix(body []byte, prefix string) []byte {
 		return body
 	}
 
-	// Collect built-in tool names (those with a non-empty "type" field) so we can
-	// skip them consistently in both tools and message history.
-	builtinTools := map[string]bool{}
-	for _, name := range []string{"web_search", "code_execution", "text_editor", "computer"} {
-		builtinTools[name] = true
-	}
+	// Collect built-in tool names from the authoritative fallback seed list and
+	// augment it with any typed built-ins present in the current request body.
+	builtinTools := helps.AugmentClaudeBuiltinToolRegistry(body, nil)
 
 	if tools := gjson.GetBytes(body, "tools"); tools.Exists() && tools.IsArray() {
 		tools.ForEach(func(index, tool gjson.Result) bool {

internal/runtime/executor/claude_executor_test.go

@@ -739,6 +739,35 @@ func TestApplyClaudeToolPrefix_ToolChoiceBuiltin(t *testing.T) {
 	}
 }
 
+func TestApplyClaudeToolPrefix_KnownFallbackBuiltinsRemainUnprefixed(t *testing.T) {
+	for _, builtin := range []string{"web_search", "code_execution", "text_editor", "computer"} {
+		t.Run(builtin, func(t *testing.T) {
+			input := []byte(fmt.Sprintf(`{
+				"tools":[{"name":"Read"}],
+				"tool_choice":{"type":"tool","name":%q},
+				"messages":[{"role":"assistant","content":[{"type":"tool_use","name":%q,"id":"toolu_1","input":{}},{"type":"tool_reference","tool_name":%q},{"type":"tool_result","tool_use_id":"toolu_1","content":[{"type":"tool_reference","tool_name":%q}]}]}]
+			}`, builtin, builtin, builtin, builtin))
+			out := applyClaudeToolPrefix(input, "proxy_")
+
+			if got := gjson.GetBytes(out, "tool_choice.name").String(); got != builtin {
+				t.Fatalf("tool_choice.name = %q, want %q", got, builtin)
+			}
+			if got := gjson.GetBytes(out, "messages.0.content.0.name").String(); got != builtin {
+				t.Fatalf("messages.0.content.0.name = %q, want %q", got, builtin)
+			}
+			if got := gjson.GetBytes(out, "messages.0.content.1.tool_name").String(); got != builtin {
+				t.Fatalf("messages.0.content.1.tool_name = %q, want %q", got, builtin)
+			}
+			if got := gjson.GetBytes(out, "messages.0.content.2.content.0.tool_name").String(); got != builtin {
+				t.Fatalf("messages.0.content.2.content.0.tool_name = %q, want %q", got, builtin)
+			}
+			if got := gjson.GetBytes(out, "tools.0.name").String(); got != "proxy_Read" {
+				t.Fatalf("tools.0.name = %q, want %q", got, "proxy_Read")
+			}
+		})
+	}
+}
+
 func TestStripClaudeToolPrefixFromResponse(t *testing.T) {
 	input := []byte(`{"content":[{"type":"tool_use","name":"proxy_alpha","id":"t1","input":{}},{"type":"tool_use","name":"bravo","id":"t2","input":{}}]}`)
 	out := stripClaudeToolPrefixFromResponse(input, "proxy_")

internal/runtime/executor/helps/claude_builtin_tools.go

@@ -0,0 +1,38 @@
+package helps
+
+import "github.com/tidwall/gjson"
+
+var defaultClaudeBuiltinToolNames = []string{
+	"web_search",
+	"code_execution",
+	"text_editor",
+	"computer",
+}
+
+func newClaudeBuiltinToolRegistry() map[string]bool {
+	registry := make(map[string]bool, len(defaultClaudeBuiltinToolNames))
+	for _, name := range defaultClaudeBuiltinToolNames {
+		registry[name] = true
+	}
+	return registry
+}
+
+func AugmentClaudeBuiltinToolRegistry(body []byte, registry map[string]bool) map[string]bool {
+	if registry == nil {
+		registry = newClaudeBuiltinToolRegistry()
+	}
+	tools := gjson.GetBytes(body, "tools")
+	if !tools.Exists() || !tools.IsArray() {
+		return registry
+	}
+	tools.ForEach(func(_, tool gjson.Result) bool {
+		if tool.Get("type").String() == "" {
+			return true
+		}
+		if name := tool.Get("name").String(); name != "" {
+			registry[name] = true
+		}
+		return true
+	})
+	return registry
+}

internal/runtime/executor/helps/claude_builtin_tools_test.go

@@ -0,0 +1,32 @@
+package helps
+
+import "testing"
+
+func TestClaudeBuiltinToolRegistry_DefaultSeedFallback(t *testing.T) {
+	registry := AugmentClaudeBuiltinToolRegistry(nil, nil)
+	for _, name := range defaultClaudeBuiltinToolNames {
+		if !registry[name] {
+			t.Fatalf("default builtin %q missing from fallback registry", name)
+		}
+	}
+}
+
+func TestClaudeBuiltinToolRegistry_AugmentsTypedBuiltinsFromBody(t *testing.T) {
+	registry := AugmentClaudeBuiltinToolRegistry([]byte(`{
+		"tools": [
+			{"type": "web_search_20250305", "name": "web_search"},
+			{"type": "custom_builtin_20250401", "name": "special_builtin"},
+			{"name": "Read"}
+		]
+	}`), nil)
+
+	if !registry["web_search"] {
+		t.Fatal("expected default typed builtin web_search in registry")
+	}
+	if !registry["special_builtin"] {
+		t.Fatal("expected typed builtin from body to be added to registry")
+	}
+	if registry["Read"] {
+		t.Fatal("expected untyped custom tool to stay out of builtin registry")
+	}
+}