Merge pull request #1834 from DragonFSKY/fix/sse-streaming-accept-encoding

fix(claude): extend gzip fix to SSE success path and header-absent compression (#1763)

internal/runtime/executor/claude_executor.go

@@ -187,17 +187,15 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	}
 	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
-		// Decompress error responses (e.g. gzip-compressed 400 errors from Anthropic API).
-		errBody := httpResp.Body
-		if ce := httpResp.Header.Get("Content-Encoding"); ce != "" {
-			var decErr error
-			errBody, decErr = decodeResponseBody(httpResp.Body, ce)
-			if decErr != nil {
-				recordAPIResponseError(ctx, e.cfg, decErr)
-				msg := fmt.Sprintf("failed to decode error response body (encoding=%s): %v", ce, decErr)
-				logWithRequestID(ctx).Warn(msg)
-				return resp, statusErr{code: httpResp.StatusCode, msg: msg}
-			}
+		// Decompress error responses — pass the Content-Encoding value (may be empty)
+		// and let decodeResponseBody handle both header-declared and magic-byte-detected
+		// compression.  This keeps error-path behaviour consistent with the success path.
+		errBody, decErr := decodeResponseBody(httpResp.Body, httpResp.Header.Get("Content-Encoding"))
+		if decErr != nil {
+			recordAPIResponseError(ctx, e.cfg, decErr)
+			msg := fmt.Sprintf("failed to decode error response body: %v", decErr)
+			logWithRequestID(ctx).Warn(msg)
+			return resp, statusErr{code: httpResp.StatusCode, msg: msg}
 		}
 		b, readErr := io.ReadAll(errBody)
 		if readErr != nil {
@@ -352,17 +350,15 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	}
 	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
-		// Decompress error responses (e.g. gzip-compressed 400 errors from Anthropic API).
-		errBody := httpResp.Body
-		if ce := httpResp.Header.Get("Content-Encoding"); ce != "" {
-			var decErr error
-			errBody, decErr = decodeResponseBody(httpResp.Body, ce)
-			if decErr != nil {
-				recordAPIResponseError(ctx, e.cfg, decErr)
-				msg := fmt.Sprintf("failed to decode error response body (encoding=%s): %v", ce, decErr)
-				logWithRequestID(ctx).Warn(msg)
-				return nil, statusErr{code: httpResp.StatusCode, msg: msg}
-			}
+		// Decompress error responses — pass the Content-Encoding value (may be empty)
+		// and let decodeResponseBody handle both header-declared and magic-byte-detected
+		// compression.  This keeps error-path behaviour consistent with the success path.
+		errBody, decErr := decodeResponseBody(httpResp.Body, httpResp.Header.Get("Content-Encoding"))
+		if decErr != nil {
+			recordAPIResponseError(ctx, e.cfg, decErr)
+			msg := fmt.Sprintf("failed to decode error response body: %v", decErr)
+			logWithRequestID(ctx).Warn(msg)
+			return nil, statusErr{code: httpResp.StatusCode, msg: msg}
 		}
 		b, readErr := io.ReadAll(errBody)
 		if readErr != nil {
@@ -521,17 +517,15 @@ func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	}
 	recordAPIResponseMetadata(ctx, e.cfg, resp.StatusCode, resp.Header.Clone())
 	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
-		// Decompress error responses (e.g. gzip-compressed 400 errors from Anthropic API).
-		errBody := resp.Body
-		if ce := resp.Header.Get("Content-Encoding"); ce != "" {
-			var decErr error
-			errBody, decErr = decodeResponseBody(resp.Body, ce)
-			if decErr != nil {
-				recordAPIResponseError(ctx, e.cfg, decErr)
-				msg := fmt.Sprintf("failed to decode error response body (encoding=%s): %v", ce, decErr)
-				logWithRequestID(ctx).Warn(msg)
-				return cliproxyexecutor.Response{}, statusErr{code: resp.StatusCode, msg: msg}
-			}
+		// Decompress error responses — pass the Content-Encoding value (may be empty)
+		// and let decodeResponseBody handle both header-declared and magic-byte-detected
+		// compression.  This keeps error-path behaviour consistent with the success path.
+		errBody, decErr := decodeResponseBody(resp.Body, resp.Header.Get("Content-Encoding"))
+		if decErr != nil {
+			recordAPIResponseError(ctx, e.cfg, decErr)
+			msg := fmt.Sprintf("failed to decode error response body: %v", decErr)
+			logWithRequestID(ctx).Warn(msg)
+			return cliproxyexecutor.Response{}, statusErr{code: resp.StatusCode, msg: msg}
 		}
 		b, readErr := io.ReadAll(errBody)
 		if readErr != nil {
@@ -662,12 +656,61 @@ func (c *compositeReadCloser) Close() error {
 	return firstErr
 }
 
+// peekableBody wraps a bufio.Reader around the original ReadCloser so that
+// magic bytes can be inspected without consuming them from the stream.
+type peekableBody struct {
+	*bufio.Reader
+	closer io.Closer
+}
+
+func (p *peekableBody) Close() error {
+	return p.closer.Close()
+}
+
 func decodeResponseBody(body io.ReadCloser, contentEncoding string) (io.ReadCloser, error) {
 	if body == nil {
 		return nil, fmt.Errorf("response body is nil")
 	}
 	if contentEncoding == "" {
-		return body, nil
+		// No Content-Encoding header.  Attempt best-effort magic-byte detection to
+		// handle misbehaving upstreams that compress without setting the header.
+		// Only gzip (1f 8b) and zstd (28 b5 2f fd) have reliable magic sequences;
+		// br and deflate have none and are left as-is.
+		// The bufio wrapper preserves unread bytes so callers always see the full
+		// stream regardless of whether decompression was applied.
+		pb := &peekableBody{Reader: bufio.NewReader(body), closer: body}
+		magic, peekErr := pb.Peek(4)
+		if peekErr == nil || (peekErr == io.EOF && len(magic) >= 2) {
+			switch {
+			case len(magic) >= 2 && magic[0] == 0x1f && magic[1] == 0x8b:
+				gzipReader, gzErr := gzip.NewReader(pb)
+				if gzErr != nil {
+					_ = pb.Close()
+					return nil, fmt.Errorf("magic-byte gzip: failed to create reader: %w", gzErr)
+				}
+				return &compositeReadCloser{
+					Reader: gzipReader,
+					closers: []func() error{
+						gzipReader.Close,
+						pb.Close,
+					},
+				}, nil
+			case len(magic) >= 4 && magic[0] == 0x28 && magic[1] == 0xb5 && magic[2] == 0x2f && magic[3] == 0xfd:
+				decoder, zdErr := zstd.NewReader(pb)
+				if zdErr != nil {
+					_ = pb.Close()
+					return nil, fmt.Errorf("magic-byte zstd: failed to create reader: %w", zdErr)
+				}
+				return &compositeReadCloser{
+					Reader: decoder,
+					closers: []func() error{
+						func() error { decoder.Close(); return nil },
+						pb.Close,
+					},
+				}, nil
+			}
+		}
+		return pb, nil
 	}
 	encodings := strings.Split(contentEncoding, ",")
 	for _, raw := range encodings {
@@ -844,11 +887,15 @@ func applyClaudeHeaders(r *http.Request, auth *cliproxyauth.Auth, apiKey string,
 		r.Header.Set("User-Agent", hdrDefault(hd.UserAgent, "claude-cli/2.1.63 (external, cli)"))
 	}
 	r.Header.Set("Connection", "keep-alive")
-	r.Header.Set("Accept-Encoding", "gzip, deflate, br, zstd")
 	if stream {
 		r.Header.Set("Accept", "text/event-stream")
+		// SSE streams must not be compressed: the downstream scanner reads
+		// line-delimited text and cannot parse compressed bytes.  Using
+		// "identity" tells the upstream to send an uncompressed stream.
+		r.Header.Set("Accept-Encoding", "identity")
 	} else {
 		r.Header.Set("Accept", "application/json")
+		r.Header.Set("Accept-Encoding", "gzip, deflate, br, zstd")
 	}
 	// Keep OS/Arch mapping dynamic (not configurable).
 	// They intentionally continue to derive from runtime.GOOS/runtime.GOARCH.
@@ -857,6 +904,12 @@ func applyClaudeHeaders(r *http.Request, auth *cliproxyauth.Auth, apiKey string,
 		attrs = auth.Attributes
 	}
 	util.ApplyCustomHeadersFromAttrs(r, attrs)
+	// Re-enforce Accept-Encoding: identity after ApplyCustomHeadersFromAttrs, which
+	// may override it with a user-configured value.  Compressed SSE breaks the line
+	// scanner regardless of user preference, so this is non-negotiable for streams.
+	if stream {
+		r.Header.Set("Accept-Encoding", "identity")
+	}
 }
 
 func claudeCreds(a *cliproxyauth.Auth) (apiKey, baseURL string) {

internal/runtime/executor/claude_executor_test.go

@@ -2,6 +2,7 @@ package executor
 
 import (
 	"bytes"
+	"compress/gzip"
 	"context"
 	"io"
 	"net/http"
@@ -9,6 +10,7 @@ import (
 	"strings"
 	"testing"
 
+	"github.com/klauspost/compress/zstd"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
 	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
@@ -583,3 +585,385 @@ func testClaudeExecutorInvalidCompressedErrorBody(
 		t.Fatalf("expected status code 400, got: %v", err)
 	}
 }
+
+// TestClaudeExecutor_ExecuteStream_SetsIdentityAcceptEncoding verifies that streaming
+// requests use Accept-Encoding: identity so the upstream cannot respond with a
+// compressed SSE body that would silently break the line scanner.
+func TestClaudeExecutor_ExecuteStream_SetsIdentityAcceptEncoding(t *testing.T) {
+	var gotEncoding, gotAccept string
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		gotEncoding = r.Header.Get("Accept-Encoding")
+		gotAccept = r.Header.Get("Accept")
+		w.Header().Set("Content-Type", "text/event-stream")
+		_, _ = w.Write([]byte("data: {\"type\":\"message_stop\"}\n\n"))
+	}))
+	defer server.Close()
+
+	executor := NewClaudeExecutor(&config.Config{})
+	auth := &cliproxyauth.Auth{Attributes: map[string]string{
+		"api_key":  "key-123",
+		"base_url": server.URL,
+	}}
+	payload := []byte(`{"messages":[{"role":"user","content":[{"type":"text","text":"hi"}]}]}`)
+
+	result, err := executor.ExecuteStream(context.Background(), auth, cliproxyexecutor.Request{
+		Model:   "claude-3-5-sonnet-20241022",
+		Payload: payload,
+	}, cliproxyexecutor.Options{
+		SourceFormat: sdktranslator.FromString("claude"),
+	})
+	if err != nil {
+		t.Fatalf("ExecuteStream error: %v", err)
+	}
+	for chunk := range result.Chunks {
+		if chunk.Err != nil {
+			t.Fatalf("unexpected chunk error: %v", chunk.Err)
+		}
+	}
+
+	if gotEncoding != "identity" {
+		t.Errorf("Accept-Encoding = %q, want %q", gotEncoding, "identity")
+	}
+	if gotAccept != "text/event-stream" {
+		t.Errorf("Accept = %q, want %q", gotAccept, "text/event-stream")
+	}
+}
+
+// TestClaudeExecutor_Execute_SetsCompressedAcceptEncoding verifies that non-streaming
+// requests keep the full accept-encoding to allow response compression (which
+// decodeResponseBody handles correctly).
+func TestClaudeExecutor_Execute_SetsCompressedAcceptEncoding(t *testing.T) {
+	var gotEncoding, gotAccept string
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		gotEncoding = r.Header.Get("Accept-Encoding")
+		gotAccept = r.Header.Get("Accept")
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"id":"msg_1","type":"message","model":"claude-3-5-sonnet-20241022","role":"assistant","content":[{"type":"text","text":"hi"}],"usage":{"input_tokens":1,"output_tokens":1}}`))
+	}))
+	defer server.Close()
+
+	executor := NewClaudeExecutor(&config.Config{})
+	auth := &cliproxyauth.Auth{Attributes: map[string]string{
+		"api_key":  "key-123",
+		"base_url": server.URL,
+	}}
+	payload := []byte(`{"messages":[{"role":"user","content":[{"type":"text","text":"hi"}]}]}`)
+
+	_, err := executor.Execute(context.Background(), auth, cliproxyexecutor.Request{
+		Model:   "claude-3-5-sonnet-20241022",
+		Payload: payload,
+	}, cliproxyexecutor.Options{
+		SourceFormat: sdktranslator.FromString("claude"),
+	})
+	if err != nil {
+		t.Fatalf("Execute error: %v", err)
+	}
+
+	if gotEncoding != "gzip, deflate, br, zstd" {
+		t.Errorf("Accept-Encoding = %q, want %q", gotEncoding, "gzip, deflate, br, zstd")
+	}
+	if gotAccept != "application/json" {
+		t.Errorf("Accept = %q, want %q", gotAccept, "application/json")
+	}
+}
+
+// TestClaudeExecutor_ExecuteStream_GzipSuccessBodyDecoded verifies that a streaming
+// HTTP 200 response with Content-Encoding: gzip is correctly decompressed before
+// the line scanner runs, so SSE chunks are not silently dropped.
+func TestClaudeExecutor_ExecuteStream_GzipSuccessBodyDecoded(t *testing.T) {
+	var buf bytes.Buffer
+	gz := gzip.NewWriter(&buf)
+	_, _ = gz.Write([]byte("data: {\"type\":\"message_stop\"}\n"))
+	_ = gz.Close()
+	compressedBody := buf.Bytes()
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/event-stream")
+		w.Header().Set("Content-Encoding", "gzip")
+		_, _ = w.Write(compressedBody)
+	}))
+	defer server.Close()
+
+	executor := NewClaudeExecutor(&config.Config{})
+	auth := &cliproxyauth.Auth{Attributes: map[string]string{
+		"api_key":  "key-123",
+		"base_url": server.URL,
+	}}
+	payload := []byte(`{"messages":[{"role":"user","content":[{"type":"text","text":"hi"}]}]}`)
+
+	result, err := executor.ExecuteStream(context.Background(), auth, cliproxyexecutor.Request{
+		Model:   "claude-3-5-sonnet-20241022",
+		Payload: payload,
+	}, cliproxyexecutor.Options{
+		SourceFormat: sdktranslator.FromString("claude"),
+	})
+	if err != nil {
+		t.Fatalf("ExecuteStream error: %v", err)
+	}
+
+	var combined strings.Builder
+	for chunk := range result.Chunks {
+		if chunk.Err != nil {
+			t.Fatalf("chunk error: %v", chunk.Err)
+		}
+		combined.Write(chunk.Payload)
+	}
+
+	if combined.Len() == 0 {
+		t.Fatal("expected at least one chunk from gzip-encoded SSE body, got none (body was not decompressed)")
+	}
+	if !strings.Contains(combined.String(), "message_stop") {
+		t.Errorf("expected SSE content in chunks, got: %q", combined.String())
+	}
+}
+
+// TestDecodeResponseBody_MagicByteGzipNoHeader verifies that decodeResponseBody
+// detects gzip-compressed content via magic bytes even when Content-Encoding is absent.
+func TestDecodeResponseBody_MagicByteGzipNoHeader(t *testing.T) {
+	const plaintext = "data: {\"type\":\"message_stop\"}\n"
+
+	var buf bytes.Buffer
+	gz := gzip.NewWriter(&buf)
+	_, _ = gz.Write([]byte(plaintext))
+	_ = gz.Close()
+
+	rc := io.NopCloser(&buf)
+	decoded, err := decodeResponseBody(rc, "")
+	if err != nil {
+		t.Fatalf("decodeResponseBody error: %v", err)
+	}
+	defer decoded.Close()
+
+	got, err := io.ReadAll(decoded)
+	if err != nil {
+		t.Fatalf("ReadAll error: %v", err)
+	}
+	if string(got) != plaintext {
+		t.Errorf("decoded = %q, want %q", got, plaintext)
+	}
+}
+
+// TestDecodeResponseBody_PlainTextNoHeader verifies that decodeResponseBody returns
+// plain text untouched when Content-Encoding is absent and no magic bytes match.
+func TestDecodeResponseBody_PlainTextNoHeader(t *testing.T) {
+	const plaintext = "data: {\"type\":\"message_stop\"}\n"
+	rc := io.NopCloser(strings.NewReader(plaintext))
+	decoded, err := decodeResponseBody(rc, "")
+	if err != nil {
+		t.Fatalf("decodeResponseBody error: %v", err)
+	}
+	defer decoded.Close()
+
+	got, err := io.ReadAll(decoded)
+	if err != nil {
+		t.Fatalf("ReadAll error: %v", err)
+	}
+	if string(got) != plaintext {
+		t.Errorf("decoded = %q, want %q", got, plaintext)
+	}
+}
+
+// TestClaudeExecutor_ExecuteStream_GzipNoContentEncodingHeader verifies the full
+// pipeline: when the upstream returns a gzip-compressed SSE body WITHOUT setting
+// Content-Encoding (a misbehaving upstream), the magic-byte sniff in
+// decodeResponseBody still decompresses it, so chunks reach the caller.
+func TestClaudeExecutor_ExecuteStream_GzipNoContentEncodingHeader(t *testing.T) {
+	var buf bytes.Buffer
+	gz := gzip.NewWriter(&buf)
+	_, _ = gz.Write([]byte("data: {\"type\":\"message_stop\"}\n"))
+	_ = gz.Close()
+	compressedBody := buf.Bytes()
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/event-stream")
+		// Intentionally omit Content-Encoding to simulate misbehaving upstream.
+		_, _ = w.Write(compressedBody)
+	}))
+	defer server.Close()
+
+	executor := NewClaudeExecutor(&config.Config{})
+	auth := &cliproxyauth.Auth{Attributes: map[string]string{
+		"api_key":  "key-123",
+		"base_url": server.URL,
+	}}
+	payload := []byte(`{"messages":[{"role":"user","content":[{"type":"text","text":"hi"}]}]}`)
+
+	result, err := executor.ExecuteStream(context.Background(), auth, cliproxyexecutor.Request{
+		Model:   "claude-3-5-sonnet-20241022",
+		Payload: payload,
+	}, cliproxyexecutor.Options{
+		SourceFormat: sdktranslator.FromString("claude"),
+	})
+	if err != nil {
+		t.Fatalf("ExecuteStream error: %v", err)
+	}
+
+	var combined strings.Builder
+	for chunk := range result.Chunks {
+		if chunk.Err != nil {
+			t.Fatalf("chunk error: %v", chunk.Err)
+		}
+		combined.Write(chunk.Payload)
+	}
+
+	if combined.Len() == 0 {
+		t.Fatal("expected chunks from gzip body without Content-Encoding header, got none (magic-byte sniff failed)")
+	}
+	if !strings.Contains(combined.String(), "message_stop") {
+		t.Errorf("unexpected chunk content: %q", combined.String())
+	}
+}
+
+// TestClaudeExecutor_ExecuteStream_AcceptEncodingOverrideCannotBypassIdentity verifies
+// that injecting Accept-Encoding via auth.Attributes cannot override the stream
+// path's enforced identity encoding.
+func TestClaudeExecutor_ExecuteStream_AcceptEncodingOverrideCannotBypassIdentity(t *testing.T) {
+	var gotEncoding string
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		gotEncoding = r.Header.Get("Accept-Encoding")
+		w.Header().Set("Content-Type", "text/event-stream")
+		_, _ = w.Write([]byte("data: {\"type\":\"message_stop\"}\n\n"))
+	}))
+	defer server.Close()
+
+	executor := NewClaudeExecutor(&config.Config{})
+	// Inject Accept-Encoding via the custom header attribute mechanism.
+	auth := &cliproxyauth.Auth{Attributes: map[string]string{
+		"api_key":             "key-123",
+		"base_url":            server.URL,
+		"header:Accept-Encoding": "gzip, deflate, br, zstd",
+	}}
+	payload := []byte(`{"messages":[{"role":"user","content":[{"type":"text","text":"hi"}]}]}`)
+
+	result, err := executor.ExecuteStream(context.Background(), auth, cliproxyexecutor.Request{
+		Model:   "claude-3-5-sonnet-20241022",
+		Payload: payload,
+	}, cliproxyexecutor.Options{
+		SourceFormat: sdktranslator.FromString("claude"),
+	})
+	if err != nil {
+		t.Fatalf("ExecuteStream error: %v", err)
+	}
+	for chunk := range result.Chunks {
+		if chunk.Err != nil {
+			t.Fatalf("unexpected chunk error: %v", chunk.Err)
+		}
+	}
+
+	if gotEncoding != "identity" {
+		t.Errorf("Accept-Encoding = %q; stream path must enforce identity regardless of auth.Attributes override", gotEncoding)
+	}
+}
+
+// TestDecodeResponseBody_MagicByteZstdNoHeader verifies that decodeResponseBody
+// detects zstd-compressed content via magic bytes (28 b5 2f fd) even when
+// Content-Encoding is absent.
+func TestDecodeResponseBody_MagicByteZstdNoHeader(t *testing.T) {
+	const plaintext = "data: {\"type\":\"message_stop\"}\n"
+
+	var buf bytes.Buffer
+	enc, err := zstd.NewWriter(&buf)
+	if err != nil {
+		t.Fatalf("zstd.NewWriter: %v", err)
+	}
+	_, _ = enc.Write([]byte(plaintext))
+	_ = enc.Close()
+
+	rc := io.NopCloser(&buf)
+	decoded, err := decodeResponseBody(rc, "")
+	if err != nil {
+		t.Fatalf("decodeResponseBody error: %v", err)
+	}
+	defer decoded.Close()
+
+	got, err := io.ReadAll(decoded)
+	if err != nil {
+		t.Fatalf("ReadAll error: %v", err)
+	}
+	if string(got) != plaintext {
+		t.Errorf("decoded = %q, want %q", got, plaintext)
+	}
+}
+
+// TestClaudeExecutor_Execute_GzipErrorBodyNoContentEncodingHeader verifies that the
+// error path (4xx) correctly decompresses a gzip body even when the upstream omits
+// the Content-Encoding header.  This closes the gap left by PR #1771, which only
+// fixed header-declared compression on the error path.
+func TestClaudeExecutor_Execute_GzipErrorBodyNoContentEncodingHeader(t *testing.T) {
+	const errJSON = `{"type":"error","error":{"type":"invalid_request_error","message":"test error"}}`
+
+	var buf bytes.Buffer
+	gz := gzip.NewWriter(&buf)
+	_, _ = gz.Write([]byte(errJSON))
+	_ = gz.Close()
+	compressedBody := buf.Bytes()
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		// Intentionally omit Content-Encoding to simulate misbehaving upstream.
+		w.WriteHeader(http.StatusBadRequest)
+		_, _ = w.Write(compressedBody)
+	}))
+	defer server.Close()
+
+	executor := NewClaudeExecutor(&config.Config{})
+	auth := &cliproxyauth.Auth{Attributes: map[string]string{
+		"api_key":  "key-123",
+		"base_url": server.URL,
+	}}
+	payload := []byte(`{"messages":[{"role":"user","content":[{"type":"text","text":"hi"}]}]}`)
+
+	_, err := executor.Execute(context.Background(), auth, cliproxyexecutor.Request{
+		Model:   "claude-3-5-sonnet-20241022",
+		Payload: payload,
+	}, cliproxyexecutor.Options{
+		SourceFormat: sdktranslator.FromString("claude"),
+	})
+	if err == nil {
+		t.Fatal("expected an error for 400 response, got nil")
+	}
+	if !strings.Contains(err.Error(), "test error") {
+		t.Errorf("error message should contain decompressed JSON, got: %q", err.Error())
+	}
+}
+
+// TestClaudeExecutor_ExecuteStream_GzipErrorBodyNoContentEncodingHeader verifies
+// the same for the streaming executor: 4xx gzip body without Content-Encoding is
+// decoded and the error message is readable.
+func TestClaudeExecutor_ExecuteStream_GzipErrorBodyNoContentEncodingHeader(t *testing.T) {
+	const errJSON = `{"type":"error","error":{"type":"invalid_request_error","message":"stream test error"}}`
+
+	var buf bytes.Buffer
+	gz := gzip.NewWriter(&buf)
+	_, _ = gz.Write([]byte(errJSON))
+	_ = gz.Close()
+	compressedBody := buf.Bytes()
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		// Intentionally omit Content-Encoding to simulate misbehaving upstream.
+		w.WriteHeader(http.StatusBadRequest)
+		_, _ = w.Write(compressedBody)
+	}))
+	defer server.Close()
+
+	executor := NewClaudeExecutor(&config.Config{})
+	auth := &cliproxyauth.Auth{Attributes: map[string]string{
+		"api_key":  "key-123",
+		"base_url": server.URL,
+	}}
+	payload := []byte(`{"messages":[{"role":"user","content":[{"type":"text","text":"hi"}]}]}`)
+
+	_, err := executor.ExecuteStream(context.Background(), auth, cliproxyexecutor.Request{
+		Model:   "claude-3-5-sonnet-20241022",
+		Payload: payload,
+	}, cliproxyexecutor.Options{
+		SourceFormat: sdktranslator.FromString("claude"),
+	})
+	if err == nil {
+		t.Fatal("expected an error for 400 response, got nil")
+	}
+	if !strings.Contains(err.Error(), "stream test error") {
+		t.Errorf("error message should contain decompressed JSON, got: %q", err.Error())
+	}
+}