fix: surface agent stderr, auto-retry restart-killed tasks, handle stale sandboxes

#1 - Diagnostics: tailFile() reads last 20 lines of subprocess stderr and
appends to error message when claude/gemini exits non-zero. Previously all
exit-1 failures were opaque; now the error_msg carries the actual subprocess
output.

#4 - Restart recovery: RecoverStaleRunning() now re-queues tasks after
marking them FAILED, so tasks killed by a server restart automatically
retry on the next boot rather than staying permanently FAILED.

#2 - Stale sandbox: If a resume execution's preserved SandboxDir no longer
exists (e.g. /tmp purge after reboot), clone a fresh sandbox instead of
failing immediately with "no such file or directory".

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

1 files changed, 86 insertions, 0 deletions

diff --git a/internal/executor/claude_test.go b/internal/executor/claude_test.go
index 7ab0802..9bb873f 100644
--- a/internal/executor/claude_test.go
+++ b/internal/executor/claude_test.go
@@ -3,6 +3,7 @@ package executor
 import (
 	"context"
 	"errors"
+	"fmt"
 	"io"
 	"log/slog"
 	"os"
@@ -579,6 +580,63 @@ func TestClaudeRunner_Run_ResumeUsesStoredSandboxDir(t *testing.T) {
 	}
 }
 
+func TestClaudeRunner_Run_StaleSandboxDir_ClonesAfresh(t *testing.T) {
+	logDir := t.TempDir()
+	projectDir := t.TempDir()
+	initGitRepo(t, projectDir)
+
+	cwdFile := filepath.Join(logDir, "cwd.txt")
+	scriptPath := filepath.Join(t.TempDir(), "fake-claude.sh")
+	script := "#!/bin/sh\nprintf '%s' \"$PWD\" > " + cwdFile + "\n"
+	if err := os.WriteFile(scriptPath, []byte(script), 0755); err != nil {
+		t.Fatalf("write script: %v", err)
+	}
+
+	r := &ClaudeRunner{
+		BinaryPath: scriptPath,
+		Logger:     slog.New(slog.NewTextHandler(io.Discard, nil)),
+		LogDir:     logDir,
+	}
+	tk := &task.Task{
+		Agent: task.AgentConfig{
+			Type:         "claude",
+			ProjectDir:   projectDir,
+			SkipPlanning: true,
+		},
+	}
+	// Point to a sandbox that no longer exists (e.g. /tmp was purged).
+	staleSandbox := filepath.Join(t.TempDir(), "gone")
+	e := &storage.Execution{
+		ID:              "resume-exec-2",
+		TaskID:          "task-2",
+		ResumeSessionID: "session-abc",
+		ResumeAnswer:    "ok",
+		SandboxDir:      staleSandbox,
+	}
+
+	if err := r.Run(context.Background(), tk, e); err != nil {
+		t.Fatalf("Run with stale sandbox: %v", err)
+	}
+
+	got, err := os.ReadFile(cwdFile)
+	if err != nil {
+		t.Fatalf("cwd file not written: %v", err)
+	}
+	// Should have run in a fresh sandbox (not the stale path, not the raw projectDir).
+	// The sandbox is removed after teardown, so we only check what it wasn't.
+	cwd := string(got)
+	if cwd == staleSandbox {
+		t.Error("ran in stale sandbox dir that doesn't exist")
+	}
+	if cwd == projectDir {
+		t.Error("ran directly in project_dir; expected a fresh sandbox clone")
+	}
+	// cwd should look like a claudomator sandbox path.
+	if !strings.Contains(cwd, "claudomator-sandbox-") {
+		t.Errorf("expected sandbox path, got %q", cwd)
+	}
+}
+
 func TestIsCompletionReport(t *testing.T) {
 	tests := []struct {
 		name     string
@@ -621,6 +679,34 @@ func TestIsCompletionReport(t *testing.T) {
 	}
 }
 
+func TestTailFile_ReturnsLastNLines(t *testing.T) {
+	f, err := os.CreateTemp("", "tailfile-*")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer os.Remove(f.Name())
+	for i := 1; i <= 30; i++ {
+		fmt.Fprintf(f, "line %d\n", i)
+	}
+	f.Close()
+
+	got := tailFile(f.Name(), 5)
+	lines := strings.Split(got, "\n")
+	if len(lines) != 5 {
+		t.Fatalf("want 5 lines, got %d: %q", len(lines), got)
+	}
+	if lines[0] != "line 26" || lines[4] != "line 30" {
+		t.Errorf("want lines 26-30, got: %q", got)
+	}
+}
+
+func TestTailFile_MissingFile_ReturnsEmpty(t *testing.T) {
+	got := tailFile("/nonexistent/path/file.log", 10)
+	if got != "" {
+		t.Errorf("want empty string for missing file, got %q", got)
+	}
+}
+
 func TestGitSafe_PrependsSafeDirectory(t *testing.T) {
 	got := gitSafe("-C", "/some/path", "status")
 	want := []string{"-c", "safe.directory=*", "-C", "/some/path", "status"}