Merge pull request #1662 from dgageot/multi-turn

Multi turn (cagent exec|run|eval)

Author: dgageot

cmd/root/exec.go

@@ -11,17 +11,16 @@ func newExecCmd() *cobra.Command {
 	var flags runExecFlags
 
 	cmd := &cobra.Command{
-		Use:   "exec <agent-file>|<registry-ref>",
+		Use:   "exec <agent-file>|<registry-ref> <message>...",
 		Short: "Execute an agent",
-		Long:  "Execute an agent (Single user message / No TUI)",
-		Example: `  cagent exec ./agent.yaml
-  cagent exec ./team.yaml --agent root
-  cagent exec ./echo.yaml "INSTRUCTIONS"
+		Long:  "Execute an agent with one or more user messages (multi-turn, No TUI)",
+		Example: `  cagent exec ./agent.yaml "What is Go?"
+  cagent exec ./team.yaml --agent root "First question" "Follow-up question"
   echo "INSTRUCTIONS" | cagent exec ./echo.yaml -
   cagent exec ./agent.yaml "question" --record  # Records to auto-generated file`,
 		GroupID:           "core",
 		ValidArgsFunction: completeRunExec,
-		Args:              cobra.RangeArgs(1, 2),
+		Args:              cobra.MinimumNArgs(2),
 		RunE:              flags.runExecCommand,
 	}
 

cmd/root/run.go

@@ -58,19 +58,20 @@ func newRunCmd() *cobra.Command {
 	var flags runExecFlags
 
 	cmd := &cobra.Command{
-		Use:   "run [<agent-file>|<registry-ref>] [message|-]",
+		Use:   "run [<agent-file>|<registry-ref>] [message]...",
 		Short: "Run an agent",
 		Long:  "Run an agent with the specified configuration and prompt",
 		Example: `  cagent run ./agent.yaml
   cagent run ./team.yaml --agent root
   cagent run # built-in default agent
   cagent run coder # built-in coding agent
   cagent run ./echo.yaml "INSTRUCTIONS"
+  cagent run ./echo.yaml "First question" "Follow-up question"
   echo "INSTRUCTIONS" | cagent run ./echo.yaml -
   cagent run ./agent.yaml --record  # Records session to auto-generated file`,
 		GroupID:           "core",
 		ValidArgsFunction: completeRunExec,
-		Args:              cobra.RangeArgs(0, 2),
+		Args:              cobra.ArbitraryArgs,
 		RunE:              flags.runRunCommand,
 	}
 
@@ -419,20 +420,16 @@ func (f *runExecFlags) createLocalRuntimeAndSession(ctx context.Context, loadRes
 }
 
 func (f *runExecFlags) handleExecMode(ctx context.Context, out *cli.Printer, rt runtime.Runtime, sess *session.Session, args []string) error {
-	execArgs := []string{"exec"}
-	if len(args) == 2 {
-		execArgs = append(execArgs, args[1])
-	} else {
-		execArgs = append(execArgs, "Please proceed.")
-	}
+	// args[0] is the agent file; args[1:] are user messages for multi-turn conversation
+	userMessages := args[1:]
 
 	err := cli.Run(ctx, out, cli.Config{
 		AppName:        AppName,
 		AttachmentPath: f.attachmentPath,
 		HideToolCalls:  f.hideToolCalls,
 		OutputJSON:     f.outputJSON,
 		AutoApprove:    f.autoApprove,
-	}, rt, sess, execArgs)
+	}, rt, sess, userMessages)
 	var cliErr cli.RuntimeError
 	if errors.As(err, &cliErr) {
 		return RuntimeError{Err: cliErr.Err}
@@ -467,6 +464,9 @@ func (f *runExecFlags) handleRunMode(ctx context.Context, rt runtime.Runtime, se
 	if firstMessage != nil {
 		opts = append(opts, app.WithFirstMessage(*firstMessage))
 	}
+	if len(args) > 2 {
+		opts = append(opts, app.WithQueuedMessages(args[2:]))
+	}
 	if f.attachmentPath != "" {
 		opts = append(opts, app.WithFirstMessageAttachment(f.attachmentPath))
 	}

pkg/app/app.go

@@ -32,6 +32,7 @@ type App struct {
 	session                *session.Session
 	firstMessage           *string
 	firstMessageAttach     string
+	queuedMessages         []string
 	events                 chan tea.Msg
 	throttleDuration       time.Duration
 	cancel                 context.CancelFunc
@@ -65,6 +66,15 @@ func WithExitAfterFirstResponse() Opt {
 	}
 }
 
+// WithQueuedMessages sets messages to be queued after the first message is sent.
+// These messages will be delivered to the TUI as SendMsg events, which the
+// chat page will queue and process sequentially after each agent response.
+func WithQueuedMessages(msgs []string) Opt {
+	return func(a *App) {
+		a.queuedMessages = msgs
+	}
+}
+
 // WithTitleGenerator sets the title generator for local title generation.
 // If not set, title generation will be handled by the runtime (for remote) or skipped.
 func WithTitleGenerator(gen *sessiontitle.Generator) Opt {
@@ -123,21 +133,36 @@ func (a *App) SendFirstMessage() tea.Cmd {
 		return nil
 	}
 
-	return func() tea.Msg {
-		// Use the shared PrepareUserMessage function for consistent attachment handling
-		userMsg := cli.PrepareUserMessage(context.Background(), a.runtime, *a.firstMessage, a.firstMessageAttach)
+	cmds := []tea.Cmd{
+		func() tea.Msg {
+			// Use the shared PrepareUserMessage function for consistent attachment handling
+			userMsg := cli.PrepareUserMessage(context.Background(), a.runtime, *a.firstMessage, a.firstMessageAttach)
 
-		// If the message has multi-content (attachments), we need to handle it specially
-		if len(userMsg.Message.MultiContent) > 0 {
-			return messages.SendAttachmentMsg{
-				Content: userMsg,
+			// If the message has multi-content (attachments), we need to handle it specially
+			if len(userMsg.Message.MultiContent) > 0 {
+				return messages.SendAttachmentMsg{
+					Content: userMsg,
+				}
 			}
-		}
 
-		return messages.SendMsg{
-			Content: userMsg.Message.Content,
-		}
+			return messages.SendMsg{
+				Content: userMsg.Message.Content,
+			}
+		},
+	}
+
+	// Queue additional messages to be sent after the first one.
+	// The TUI's message queue will hold them until the agent finishes
+	// processing the previous message.
+	for _, msg := range a.queuedMessages {
+		cmds = append(cmds, func() tea.Msg {
+			return messages.SendMsg{
+				Content: msg,
+			}
+		})
 	}
+
+	return tea.Sequence(cmds...)
 }
 
 // CurrentAgentCommands returns the commands for the active agent

pkg/cli/runner.go

@@ -40,8 +40,10 @@ type Config struct {
 	OutputJSON     bool
 }
 
-// Run executes an agent in non-TUI mode, handling user input and runtime events
-func Run(ctx context.Context, out *Printer, cfg Config, rt runtime.Runtime, sess *session.Session, args []string) error {
+// Run executes an agent in non-TUI mode, handling user input and runtime events.
+// userMessages contains the user messages to send. If a single message is "-",
+// input is read from stdin. If empty, an interactive prompt loop is started.
+func Run(ctx context.Context, out *Printer, cfg Config, rt runtime.Runtime, sess *session.Session, userMessages []string) error {
 	// Create a cancellable context for this agentic loop and wire Ctrl+C to cancel it
 	ctx, cancel := context.WithCancel(ctx)
 	defer cancel()
@@ -193,22 +195,26 @@ func Run(ctx context.Context, out *Printer, cfg Config, rt runtime.Runtime, sess
 		return nil
 	}
 
-	if len(args) == 2 {
-		if args[1] == "-" {
-			buf, err := io.ReadAll(os.Stdin)
-			if err != nil {
-				return fmt.Errorf("failed to read from stdin: %w", err)
-			}
+	switch {
+	case len(userMessages) == 1 && userMessages[0] == "-":
+		// Single "-" argument: read from stdin
+		buf, err := io.ReadAll(os.Stdin)
+		if err != nil {
+			return fmt.Errorf("failed to read from stdin: %w", err)
+		}
 
-			if err := oneLoop(string(buf), os.Stdin); err != nil {
-				return err
-			}
-		} else {
-			if err := oneLoop(args[1], os.Stdin); err != nil {
+		if err := oneLoop(string(buf), os.Stdin); err != nil {
+			return err
+		}
+	case len(userMessages) > 0:
+		// One or more messages: multi-turn conversation
+		for _, msg := range userMessages {
+			if err := oneLoop(msg, os.Stdin); err != nil {
 				return err
 			}
 		}
-	} else {
+	default:
+		// No messages: interactive prompt loop
 		out.PrintWelcomeMessage(cfg.AppName)
 		firstQuestion := true
 		for {

pkg/evaluation/eval.go

@@ -292,7 +292,7 @@ func (r *Runner) runSingleEval(ctx context.Context, evalSess *InputSession) (Res
 	result := Result{
 		InputPath:         evalSess.SourcePath,
 		Title:             evalSess.Title,
-		Question:          getFirstUserMessage(evalSess.Session),
+		Question:          strings.Join(getUserMessages(evalSess.Session), "\n"),
 		SizeExpected:      evals.Size,
 		RelevanceExpected: float64(len(evals.Relevance)),
 	}
@@ -309,7 +309,7 @@ func (r *Runner) runSingleEval(ctx context.Context, evalSess *InputSession) (Res
 		return result, fmt.Errorf("building eval image: %w", err)
 	}
 
-	events, err := r.runCagentInContainer(ctx, imageID, result.Question)
+	events, err := r.runCagentInContainer(ctx, imageID, getUserMessages(evalSess.Session))
 	if err != nil {
 		return result, fmt.Errorf("running cagent in container: %w", err)
 	}
@@ -322,7 +322,7 @@ func (r *Runner) runSingleEval(ctx context.Context, evalSess *InputSession) (Res
 	result.Size = getResponseSize(result.Response)
 
 	// Build session from events for database storage
-	result.Session = SessionFromEvents(events, evalSess.Title, result.Question)
+	result.Session = SessionFromEvents(events, evalSess.Title, getUserMessages(evalSess.Session))
 	result.Session.Evals = evals
 
 	if len(expectedToolCalls) > 0 || len(actualToolCalls) > 0 {
@@ -346,7 +346,7 @@ func (r *Runner) runSingleEval(ctx context.Context, evalSess *InputSession) (Res
 	return result, nil
 }
 
-func (r *Runner) runCagentInContainer(ctx context.Context, imageID, question string) ([]map[string]any, error) {
+func (r *Runner) runCagentInContainer(ctx context.Context, imageID string, questions []string) ([]map[string]any, error) {
 	agentDir := r.agentSource.ParentDir()
 	agentFile := filepath.Base(r.agentSource.Name())
 	containerName := fmt.Sprintf("cagent-eval-%d", uuid.New().ID())
@@ -396,7 +396,8 @@ func (r *Runner) runCagentInContainer(ctx context.Context, imageID, question str
 		}
 	}
 
-	args = append(args, imageID, "/configs/"+agentFile, question)
+	args = append(args, imageID, "/configs/"+agentFile)
+	args = append(args, questions...)
 
 	cmd := exec.CommandContext(ctx, "docker", args...)
 	cmd.Env = append(env, os.Environ()...)

pkg/evaluation/save.go

@@ -55,17 +55,28 @@ func SaveRunSessions(ctx context.Context, run *EvalRun, outputDir string) (strin
 // SessionFromEvents reconstructs a session from raw container output events.
 // This parses the JSON events emitted by cagent --json and builds a session
 // with the conversation history.
-func SessionFromEvents(events []map[string]any, title, question string) *session.Session {
+func SessionFromEvents(events []map[string]any, title string, questions []string) *session.Session {
 	sess := session.New(
 		session.WithTitle(title),
 		session.WithToolsApproved(true),
 	)
 
-	// Add the user question as the first message
-	if question != "" {
-		sess.AddMessage(session.UserMessage(question))
+	// Add user questions as initial messages.
+	// For multi-turn evals, these are interleaved with agent responses
+	// as they appear in the event stream. The first question is added
+	// upfront; subsequent questions are inserted when a stream_stopped
+	// event indicates the agent finished processing the previous turn.
+	questionIdx := 0
+	addNextQuestion := func() {
+		if questionIdx < len(questions) {
+			sess.AddMessage(session.UserMessage(questions[questionIdx]))
+			questionIdx++
+		}
 	}
 
+	// Add the first question
+	addNextQuestion()
+
 	// Track current assistant message being built
 	var currentContent strings.Builder
 	var currentReasoningContent strings.Builder
@@ -225,6 +236,9 @@ func SessionFromEvents(events []map[string]any, title, question string) *session
 		case "stream_stopped":
 			// Flush final assistant message
 			flushAssistantMessage()
+
+			// In multi-turn evals, add the next user question after each turn
+			addNextQuestion()
 		}
 	}
 

pkg/evaluation/save_test.go

@@ -341,7 +341,7 @@ func TestSessionFromEvents(t *testing.T) {
 		t.Run(tt.name, func(t *testing.T) {
 			t.Parallel()
 
-			sess := SessionFromEvents(tt.events, tt.title, tt.question)
+			sess := SessionFromEvents(tt.events, tt.title, []string{tt.question})
 
 			assert.Equal(t, tt.title, sess.Title)
 			assert.Len(t, sess.Messages, tt.wantMessages)
@@ -377,7 +377,7 @@ func TestSessionFromEventsTokenUsage(t *testing.T) {
 		{"type": "stream_stopped"},
 	}
 
-	sess := SessionFromEvents(events, "test", "question")
+	sess := SessionFromEvents(events, "test", []string{"question"})
 
 	assert.Equal(t, int64(100), sess.InputTokens)
 	assert.Equal(t, int64(50), sess.OutputTokens)
@@ -461,7 +461,7 @@ func TestSessionFromEventsWithToolDefinitions(t *testing.T) {
 		{"type": "stream_stopped"},
 	}
 
-	sess := SessionFromEvents(events, "test", "read the file")
+	sess := SessionFromEvents(events, "test", []string{"read the file"})
 
 	// Find the assistant message with tool calls
 	var assistantMsg *session.Message
@@ -498,7 +498,7 @@ func TestSessionFromEventsWithReasoningContent(t *testing.T) {
 		{"type": "stream_stopped"},
 	}
 
-	sess := SessionFromEvents(events, "test", "complex question")
+	sess := SessionFromEvents(events, "test", []string{"complex question"})
 
 	// Find the assistant message
 	var assistantMsg *session.Message
@@ -537,7 +537,7 @@ func TestSessionFromEventsWithPerMessageUsage(t *testing.T) {
 		{"type": "stream_stopped"},
 	}
 
-	sess := SessionFromEvents(events, "test", "hi")
+	sess := SessionFromEvents(events, "test", []string{"hi"})
 
 	// Check session-level usage
 	assert.Equal(t, int64(100), sess.InputTokens)
@@ -571,7 +571,7 @@ func TestSessionFromEventsWithError(t *testing.T) {
 		{"type": "stream_stopped"},
 	}
 
-	sess := SessionFromEvents(events, "test", "do something")
+	sess := SessionFromEvents(events, "test", []string{"do something"})
 
 	// Should have: user message, assistant message, error message
 	assert.Len(t, sess.Messages, 3)
@@ -593,7 +593,7 @@ func TestSessionFromEventsWithSessionTitle(t *testing.T) {
 	}
 
 	// Start with a default title
-	sess := SessionFromEvents(events, "default-title", "hi")
+	sess := SessionFromEvents(events, "default-title", []string{"hi"})
 
 	// Title should be updated from the event
 	assert.Equal(t, "Auto-generated title", sess.Title)

pkg/evaluation/types.go

@@ -122,13 +122,14 @@ type Config struct {
 
 // Session helper functions
 
-func getFirstUserMessage(sess *session.Session) string {
+func getUserMessages(sess *session.Session) []string {
+	var messages []string
 	for _, msg := range sess.GetAllMessages() {
 		if msg.Message.Role == "user" {
-			return msg.Message.Content
+			messages = append(messages, msg.Message.Content)
 		}
 	}
-	return ""
+	return messages
 }
 
 func extractToolCalls(items []session.Item) []string {