feat(tools): stream shell_run output live into the TUI

internal/tools/shell.go

@@ -1,6 +1,7 @@
 package tools
 
 import (
+	"bytes"
 	"context"
 	"encoding/json"
 	"errors"
@@ -8,6 +9,7 @@ import (
 	"os"
 	"os/exec"
 	"strings"
+	"sync"
 
 	"agentu/pkg/llm"
 )
@@ -46,6 +48,10 @@ type shellRunArgs struct {
 }
 
 func (t *ShellRunTool) Execute(ctx context.Context, raw json.RawMessage) (string, error) {
+	return t.ExecuteStream(ctx, raw, nil)
+}
+
+func (t *ShellRunTool) ExecuteStream(ctx context.Context, raw json.RawMessage, emit func(string) error) (string, error) {
 	var args shellRunArgs
 	if err := json.Unmarshal(raw, &args); err != nil {
 		return "", fmt.Errorf("parse arguments: %w", err)
@@ -66,16 +72,45 @@ func (t *ShellRunTool) Execute(ctx context.Context, raw json.RawMessage) (string
 	cmd := exec.CommandContext(ctx, shell, "-lc", args.Command)
 	cmd.Dir = workdir
 
-	output, err := cmd.CombinedOutput()
-	text := FormatResult(string(output))
+	var mu sync.Mutex
+	var combined bytes.Buffer
+	writer := &shellStreamWriter{mu: &mu, buf: &combined, emit: emit}
+	cmd.Stdout = writer
+	cmd.Stderr = writer
+
+	runErr := cmd.Run()
+	text := FormatResult(combined.String())
 	if errors.Is(ctx.Err(), context.DeadlineExceeded) {
 		return text, fmt.Errorf("command timed out: %w", ctx.Err())
 	}
-	if err != nil {
+	if runErr != nil {
 		if text != "" {
-			return text, fmt.Errorf("command failed: %w", err)
+			return text, fmt.Errorf("command failed: %w", runErr)
 		}
-		return "", fmt.Errorf("command failed: %w", err)
+		return "", fmt.Errorf("command failed: %w", runErr)
 	}
 	return text, nil
 }
+
+// shellStreamWriter captures command output for the final result while
+// forwarding every write to the streaming callback as it happens.
+type shellStreamWriter struct {
+	mu   *sync.Mutex
+	buf  *bytes.Buffer
+	emit func(string) error
+}
+
+func (w *shellStreamWriter) Write(p []byte) (int, error) {
+	w.mu.Lock()
+	n, writeErr := w.buf.Write(p)
+	w.mu.Unlock()
+	if writeErr != nil {
+		return n, writeErr
+	}
+	if w.emit != nil {
+		if err := w.emit(string(p)); err != nil {
+			return 0, err
+		}
+	}
+	return len(p), nil
+}

internal/tools/shell_test.go

@@ -0,0 +1,71 @@
+package tools
+
+import (
+	"context"
+	"encoding/json"
+	"strings"
+	"sync"
+	"testing"
+)
+
+func TestShellRunStreamsOutputWhileRunning(t *testing.T) {
+	tool := NewShellRunTool(t.TempDir())
+	var mu sync.Mutex
+	var chunks []string
+	output, err := tool.ExecuteStream(context.Background(), json.RawMessage(`{"command":"printf 'a\\n'; sleep 0.05; printf 'b\\n'"}`), func(chunk string) error {
+		mu.Lock()
+		chunks = append(chunks, chunk)
+		mu.Unlock()
+		return nil
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(output, "a") || !strings.Contains(output, "b") {
+		t.Fatalf("output = %q", output)
+	}
+	mu.Lock()
+	joined := strings.Join(chunks, "")
+	mu.Unlock()
+	if len(chunks) == 0 {
+		t.Fatal("expected at least one streamed chunk")
+	}
+	if !strings.Contains(joined, "a") || !strings.Contains(joined, "b") {
+		t.Fatalf("streamed chunks = %q", joined)
+	}
+}
+
+func TestShellRunExecuteCollectsFullOutput(t *testing.T) {
+	tool := NewShellRunTool(t.TempDir())
+	output, err := tool.Execute(context.Background(), json.RawMessage(`{"command":"printf 'hello world'"}`))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if strings.TrimSpace(output) != "hello world" {
+		t.Fatalf("output = %q", output)
+	}
+}
+
+func TestShellRunStreamsStderrAndStdout(t *testing.T) {
+	tool := NewShellRunTool(t.TempDir())
+	var mu sync.Mutex
+	var chunks []string
+	output, err := tool.ExecuteStream(context.Background(), json.RawMessage(`{"command":"echo out; echo err >&2"}`), func(chunk string) error {
+		mu.Lock()
+		chunks = append(chunks, chunk)
+		mu.Unlock()
+		return nil
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(output, "out") || !strings.Contains(output, "err") {
+		t.Fatalf("output = %q", output)
+	}
+	mu.Lock()
+	joined := strings.Join(chunks, "")
+	mu.Unlock()
+	if !strings.Contains(joined, "out") || !strings.Contains(joined, "err") {
+		t.Fatalf("streamed chunks = %q", joined)
+	}
+}

internal/tools/tools.go

@@ -19,6 +19,15 @@ type Tool interface {
 	Execute(ctx context.Context, args json.RawMessage) (string, error)
 }
 
+// StreamingTool is optionally implemented by tools that can report partial
+// output while they execute. emit receives chunks of the tool's output as
+// they are produced; the returned string is still the complete result that
+// gets sent back to the model.
+type StreamingTool interface {
+	Tool
+	ExecuteStream(ctx context.Context, args json.RawMessage, emit func(string) error) (string, error)
+}
+
 // MutatingTool is optionally implemented by tools that modify external state.
 // It is used by the agent to determine whether tool calls can run in parallel.
 type MutatingTool interface {