Files
kanban/backend/chat_ws_test.go
egutierrez ce49fdf9ff feat(chat): MCP server + WebSocket streaming, replace XML actions
- Backend: kanban binary gana subcomando `kanban mcp` que actua como MCP
  server via stdio. Tools = mismo set que executeTool (14). El subprocess
  llama de vuelta al backend via /api/tool/{name} con token interno.
- Backend: nuevo endpoint POST /api/tool/{name} (auth: X-Internal-Token).
- Backend: chat.go refactor — POST /api/chat reemplazado por GET
  /api/chat/ws (WebSocket). Lanza claude -p con --output-format stream-json
  --verbose --mcp-config y reenvia eventos (delta/tool_use/tool_result/
  result/done/error) como mensajes JSON al cliente.
- Backend: usa funciones nuevas del registry claude_stream_go_core (spawn
  + parser NDJSON) y mcp_server_stdio_go_infra (JSON-RPC stdio).
- Frontend: streamChat sobre WebSocket. ChatPanel renderiza deltas en
  vivo, chips para tool_use, badges teal/red para tool_result.
- Borrado: extractActions, actionsBlockMarker, XML system prompt.
- Tests: 7 nuevos en backend (chat_ws_test.go + endpoint /api/tool).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-09 14:54:38 +02:00

297 lines
8.8 KiB
Go

package main
import (
"context"
"encoding/json"
"net/http"
"net/http/httptest"
"net/url"
"os"
"path/filepath"
"strings"
"testing"
"time"
"nhooyr.io/websocket"
)
// fakeClaudeScript writes a bash script that emits NDJSON stream-json events
// to stdout and exits 0. Returns the absolute path of the script.
func fakeClaudeScript(t *testing.T, payload string) string {
t.Helper()
if _, err := os.Stat("/bin/bash"); err != nil {
t.Skip("/bin/bash not available")
}
dir := t.TempDir()
path := filepath.Join(dir, "claude")
body := "#!/bin/bash\nset -e\ncat <<'__EOF__'\n" + payload + "\n__EOF__\n"
if err := os.WriteFile(path, []byte(body), 0o755); err != nil {
t.Fatalf("write fake claude: %v", err)
}
return path
}
// chatWSTestServer wires the WebSocket chat handler in front of a test DB.
func chatWSTestServer(t *testing.T) (*httptest.Server, *DB, string) {
t.Helper()
db := setupTestDB(t)
dir := t.TempDir()
logger := newChatLogger(filepath.Join(dir, "chat.log"))
token := generateInternalToken()
srv := httptest.NewServer(handleChatWS(db, dir, logger, token))
t.Cleanup(srv.Close)
return srv, db, token
}
func dialChatWS(t *testing.T, srv *httptest.Server) *websocket.Conn {
t.Helper()
u, _ := url.Parse(srv.URL)
wsURL := "ws://" + u.Host
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
c, _, err := websocket.Dial(ctx, wsURL, nil)
if err != nil {
t.Fatalf("dial %s: %v", wsURL, err)
}
return c
}
func readWSEvent(t *testing.T, conn *websocket.Conn) wsEvent {
t.Helper()
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
_, data, err := conn.Read(ctx)
if err != nil {
t.Fatalf("read: %v", err)
}
var ev wsEvent
if err := json.Unmarshal(data, &ev); err != nil {
t.Fatalf("unmarshal %q: %v", string(data), err)
}
return ev
}
func sendInitial(t *testing.T, conn *websocket.Conn, msgs []chatMessage) {
t.Helper()
body, _ := json.Marshal(chatRequest{Messages: msgs})
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
if err := conn.Write(ctx, websocket.MessageText, body); err != nil {
t.Fatalf("write: %v", err)
}
}
// --- WS streaming tests ---------------------------------------------------
func TestChatWS_StreamsTextDelta(t *testing.T) {
payload := `{"type":"system","subtype":"init","session_id":"s1","model":"test"}
{"type":"assistant","message":{"role":"assistant","content":[{"type":"text","text":"Hola "}]}}
{"type":"assistant","message":{"role":"assistant","content":[{"type":"text","text":"mundo"}]}}
{"type":"result","subtype":"success","is_error":false,"result":"Hola mundo","stop_reason":"end_turn"}`
t.Setenv("KANBAN_CLAUDE_BIN", fakeClaudeScript(t, payload))
srv, _, _ := chatWSTestServer(t)
conn := dialChatWS(t, srv)
defer conn.Close(websocket.StatusNormalClosure, "")
sendInitial(t, conn, []chatMessage{{Role: "user", Content: "saluda"}})
var deltas []string
var sawResult, sawDone bool
for i := 0; i < 12 && !sawDone; i++ {
ev := readWSEvent(t, conn)
switch ev.Type {
case "delta":
deltas = append(deltas, ev.Text)
case "result":
sawResult = true
case "done":
sawDone = true
case "error":
t.Fatalf("unexpected error event: %s", ev.Error)
}
}
if !sawDone {
t.Fatalf("never received done event")
}
if !sawResult {
t.Fatalf("never received result event")
}
if got := strings.Join(deltas, ""); got != "Hola mundo" {
t.Fatalf("expected 'Hola mundo' from deltas, got %q", got)
}
}
func TestChatWS_StreamsToolUseAndResult(t *testing.T) {
payload := `{"type":"system","subtype":"init"}
{"type":"assistant","message":{"role":"assistant","content":[{"type":"tool_use","id":"toolu_1","name":"mcp__kanban__create_column","input":{"name":"Backlog"}}]}}
{"type":"user","message":{"role":"user","content":[{"type":"tool_result","tool_use_id":"toolu_1","content":"{\"ok\":true,\"result\":{\"id\":\"col_x\"}}","is_error":false}]}}
{"type":"assistant","message":{"role":"assistant","content":[{"type":"text","text":"Listo"}]}}
{"type":"result","subtype":"success","is_error":false,"result":"Listo","stop_reason":"end_turn"}`
t.Setenv("KANBAN_CLAUDE_BIN", fakeClaudeScript(t, payload))
srv, _, _ := chatWSTestServer(t)
conn := dialChatWS(t, srv)
defer conn.Close(websocket.StatusNormalClosure, "")
sendInitial(t, conn, []chatMessage{{Role: "user", Content: "crea Backlog"}})
var sawToolUse, sawToolResult, sawDelta, sawDone bool
var doneEv wsEvent
for i := 0; i < 16 && !sawDone; i++ {
ev := readWSEvent(t, conn)
switch ev.Type {
case "tool_use":
sawToolUse = true
if ev.Tool != "create_column" {
t.Errorf("tool name not stripped: %q", ev.Tool)
}
if !strings.Contains(string(ev.Input), "Backlog") {
t.Errorf("input missing Backlog: %s", ev.Input)
}
case "tool_result":
sawToolResult = true
if ev.IsError {
t.Errorf("tool_result is_error true")
}
case "delta":
sawDelta = true
case "done":
sawDone = true
doneEv = ev
case "error":
t.Fatalf("unexpected error: %s", ev.Error)
}
}
if !sawToolUse || !sawToolResult || !sawDelta || !sawDone {
t.Fatalf("missing events: tool_use=%v tool_result=%v delta=%v done=%v",
sawToolUse, sawToolResult, sawDelta, sawDone)
}
if !doneEv.BoardChanged {
t.Errorf("expected board_changed=true (create_column is a mutator)")
}
}
func TestChatWS_RejectsEmptyMessages(t *testing.T) {
t.Setenv("KANBAN_CLAUDE_BIN", fakeClaudeScript(t,
`{"type":"result","subtype":"success","is_error":false,"result":""}`))
srv, _, _ := chatWSTestServer(t)
conn := dialChatWS(t, srv)
defer conn.Close(websocket.StatusNormalClosure, "")
sendInitial(t, conn, []chatMessage{})
ev := readWSEvent(t, conn)
if ev.Type != "error" {
t.Fatalf("expected error event, got %+v", ev)
}
if !strings.Contains(ev.Error, "messages required") {
t.Fatalf("unexpected error: %s", ev.Error)
}
}
func TestChatWS_PropagatesClaudeFailure(t *testing.T) {
dir := t.TempDir()
bin := filepath.Join(dir, "claude")
body := "#!/bin/bash\necho 'broken' >&2\nexit 7\n"
if err := os.WriteFile(bin, []byte(body), 0o755); err != nil {
t.Fatalf("write: %v", err)
}
t.Setenv("KANBAN_CLAUDE_BIN", bin)
srv, _, _ := chatWSTestServer(t)
conn := dialChatWS(t, srv)
defer conn.Close(websocket.StatusNormalClosure, "")
sendInitial(t, conn, []chatMessage{{Role: "user", Content: "hola"}})
deadline := time.Now().Add(5 * time.Second)
for time.Now().Before(deadline) {
ev := readWSEvent(t, conn)
switch ev.Type {
case "error":
if !strings.Contains(ev.Error, "claude exit") {
t.Fatalf("expected claude exit error, got: %s", ev.Error)
}
return
case "done":
t.Fatalf("done received before error")
}
}
t.Fatalf("never received error event")
}
// --- /api/tool internal endpoint tests ------------------------------------
func internalToolServer(t *testing.T) (*httptest.Server, *DB, string) {
t.Helper()
db := setupTestDB(t)
logger := newChatLogger(filepath.Join(t.TempDir(), "log"))
token := generateInternalToken()
mux := http.NewServeMux()
mux.Handle("POST /api/tool/{name}", handleInternalTool(db, token, logger))
srv := httptest.NewServer(mux)
t.Cleanup(srv.Close)
return srv, db, token
}
func TestInternalTool_CreateColumnRoundtrip(t *testing.T) {
srv, db, token := internalToolServer(t)
req, _ := http.NewRequest("POST", srv.URL+"/api/tool/create_column", strings.NewReader(`{"name":"Backlog"}`))
req.Header.Set("Content-Type", "application/json")
req.Header.Set(internalTokenHeader, token)
resp, err := srv.Client().Do(req)
if err != nil {
t.Fatalf("do: %v", err)
}
defer resp.Body.Close()
if resp.StatusCode != 200 {
t.Fatalf("status %d", resp.StatusCode)
}
var tr ToolResult
if err := json.NewDecoder(resp.Body).Decode(&tr); err != nil {
t.Fatalf("decode: %v", err)
}
if !tr.OK {
t.Fatalf("create_column failed: %s", tr.Error)
}
cols, err := db.ListColumns()
if err != nil {
t.Fatalf("list: %v", err)
}
if len(cols) != 1 || cols[0].Name != "Backlog" {
t.Fatalf("expected 1 col Backlog, got %+v", cols)
}
}
func TestInternalTool_RejectsMissingToken(t *testing.T) {
srv, _, _ := internalToolServer(t)
req, _ := http.NewRequest("POST", srv.URL+"/api/tool/create_column", strings.NewReader(`{"name":"X"}`))
req.Header.Set("Content-Type", "application/json")
resp, err := srv.Client().Do(req)
if err != nil {
t.Fatalf("do: %v", err)
}
defer resp.Body.Close()
if resp.StatusCode != 401 {
t.Fatalf("expected 401, got %d", resp.StatusCode)
}
}
func TestInternalTool_UnknownTool(t *testing.T) {
srv, _, token := internalToolServer(t)
req, _ := http.NewRequest("POST", srv.URL+"/api/tool/no_such", strings.NewReader(`{}`))
req.Header.Set("Content-Type", "application/json")
req.Header.Set(internalTokenHeader, token)
resp, err := srv.Client().Do(req)
if err != nil {
t.Fatalf("do: %v", err)
}
defer resp.Body.Close()
if resp.StatusCode != 404 {
t.Fatalf("expected 404, got %d", resp.StatusCode)
}
}