diff --git a/Taskfile.yml b/Taskfile.yml
index 5dad354..d9b162c 100644
--- a/Taskfile.yml
+++ b/Taskfile.yml
@@ -95,7 +95,7 @@ tasks:
       PATH: '{{.GOPATH}}/bin:{{.PATH}}'
     cmds:
       - task: install:lint
-      - golangci-lint run
+      - golangci-lint run ./...
 
   lint:fix:
     desc: Run golangci-lint and fix issues
@@ -103,7 +103,7 @@ tasks:
       PATH: '{{.GOPATH}}/bin:{{.PATH}}'
     cmds:
       - task: install:lint
-      - golangci-lint run --fix
+      - golangci-lint run ./... --fix
 
   install:lint:
     desc: Install golangci-lint if missing
diff --git a/cmd/root.go b/cmd/root.go
index cd54371..72ee9eb 100644
--- a/cmd/root.go
+++ b/cmd/root.go
@@ -166,7 +166,7 @@ func runRoot(_ *cobra.Command, args []string) error {
 				return nil
 			},
 		}
-		return interactiveRun(Version, cmds, cfg, client, shellInfo)
+		return interactiveRun(Version, cmds, cfg, client, shellInfo, explainFlag)
 	}
 
 	// Single-shot mode
diff --git a/cmd/root_test.go b/cmd/root_test.go
index 320b7e7..4fb6c50 100644
--- a/cmd/root_test.go
+++ b/cmd/root_test.go
@@ -101,7 +101,7 @@ func TestRunRoot_InteractiveMode_ConfigRunShow(t *testing.T) {
 	t.Cleanup(func() { interactiveRun = origRun })
 
 	var capturedCmds interactive.BuiltinCommands
-	interactiveRun = func(_ string, cmds interactive.BuiltinCommands, _ *config.Config, _ llm.Client, _ shell.Info) error {
+	interactiveRun = func(_ string, cmds interactive.BuiltinCommands, _ *config.Config, _ llm.Client, _ shell.Info, _ bool) error {
 		capturedCmds = cmds
 		return nil
 	}
@@ -160,7 +160,7 @@ func TestRunRoot_InteractiveMode_MemoryRun(t *testing.T) {
 	t.Cleanup(func() { interactiveRun = origRun })
 
 	var capturedCmds interactive.BuiltinCommands
-	interactiveRun = func(_ string, cmds interactive.BuiltinCommands, _ *config.Config, _ llm.Client, _ shell.Info) error {
+	interactiveRun = func(_ string, cmds interactive.BuiltinCommands, _ *config.Config, _ llm.Client, _ shell.Info, _ bool) error {
 		capturedCmds = cmds
 		return nil
 	}
@@ -219,7 +219,7 @@ func TestRunRoot_InteractiveMode_StatusDoctorSetModel(t *testing.T) {
 	t.Cleanup(func() { interactiveRun = origRun })
 
 	var capturedCmds interactive.BuiltinCommands
-	interactiveRun = func(_ string, cmds interactive.BuiltinCommands, _ *config.Config, _ llm.Client, _ shell.Info) error {
+	interactiveRun = func(_ string, cmds interactive.BuiltinCommands, _ *config.Config, _ llm.Client, _ shell.Info, _ bool) error {
 		capturedCmds = cmds
 		return nil
 	}
@@ -253,7 +253,7 @@ func TestRunRoot_InteractiveError(t *testing.T) {
 	origRun := interactiveRun
 	t.Cleanup(func() { interactiveRun = origRun })
 
-	interactiveRun = func(_ string, _ interactive.BuiltinCommands, _ *config.Config, _ llm.Client, _ shell.Info) error {
+	interactiveRun = func(_ string, _ interactive.BuiltinCommands, _ *config.Config, _ llm.Client, _ shell.Info, _ bool) error {
 		return errors.New("interactive failed")
 	}
 
@@ -267,6 +267,46 @@ func TestRunRoot_InteractiveError(t *testing.T) {
 	}
 }
 
+func TestRunRoot_InteractiveMode_ExplainFlag(t *testing.T) {
+	tempHome(t)
+	cfg := config.DefaultConfig()
+	cfg.Provider.Default = config.ProviderOpenAI
+	cfg.Provider.OpenAI.APIKey = "sk-test-interactive"
+	if err := config.Save(cfg); err != nil {
+		t.Fatalf("Save: %v", err)
+	}
+
+	origRun := interactiveRun
+	origExplain := explainFlag
+	t.Cleanup(func() {
+		interactiveRun = origRun
+		explainFlag = origExplain
+	})
+
+	var capturedExplain bool
+	interactiveRun = func(_ string, _ interactive.BuiltinCommands, _ *config.Config, _ llm.Client, _ shell.Info, explain bool) error {
+		capturedExplain = explain
+		return nil
+	}
+
+	debugFlag = ""
+	explainFlag = false
+	if err := runRoot(nil, nil); err != nil {
+		t.Fatalf("runRoot() error: %v", err)
+	}
+	if capturedExplain {
+		t.Fatal("explain = true, want false (default)")
+	}
+
+	explainFlag = true
+	if err := runRoot(nil, nil); err != nil {
+		t.Fatalf("runRoot() error: %v", err)
+	}
+	if !capturedExplain {
+		t.Fatal("explain = false, want true")
+	}
+}
+
 func TestRunRoot_DebugFileError(t *testing.T) {
 	tempHome(t)
 	cfg := config.DefaultConfig()
diff --git a/internal/executor/executor.go b/internal/executor/executor.go
index 514f026..a7722fb 100644
--- a/internal/executor/executor.go
+++ b/internal/executor/executor.go
@@ -44,7 +44,7 @@ func Run(commands []llm.Command, cfg *config.Config, shellInfo shell.Info, expla
 		dimColor.Printf(" %d%% certainty\n", cmd.Certainty)
 
 		if explain && cmd.Explanation != "" {
-			dimColor.Printf("  💡 %s\n", cmd.Explanation)
+			dimColor.Fprintf(os.Stdout, "  💡 %s\n", cmd.Explanation)
 		}
 
 		if ShouldConfirm(cmd, cfg) {
diff --git a/internal/executor/executor_test.go b/internal/executor/executor_test.go
index adeba5c..ae8b60d 100644
--- a/internal/executor/executor_test.go
+++ b/internal/executor/executor_test.go
@@ -131,12 +131,13 @@ func TestRunExecutesSafeCommandWithoutConfirmation(t *testing.T) {
 
 func TestRunWithExplainTrue(t *testing.T) {
 	cfg := defaultSafetyCfg()
+	const explanationText = "Unique explanation string that only appears in the explanation output."
 	cmds := []llm.Command{{
-		Command:     "echo explain-test",
+		Command:     "echo hello",
 		Description: "echo with explanation",
 		Risk:        "safe",
 		Certainty:   100,
-		Explanation: "Prints the text 'explain-test' to stdout.",
+		Explanation: explanationText,
 	}}
 
 	// Capture stdout to verify explanation is printed
@@ -160,7 +161,7 @@ func TestRunWithExplainTrue(t *testing.T) {
 		t.Fatalf("Run() error = %v, want nil", runErr)
 	}
 	output := buf.String()
-	if !strings.Contains(output, "explain-test") {
+	if !strings.Contains(output, explanationText) {
 		t.Errorf("output missing explanation text:\n%s", output)
 	}
 }
diff --git a/internal/interactive/repl.go b/internal/interactive/repl.go
index 7e149ba..c3ae54f 100644
--- a/internal/interactive/repl.go
+++ b/internal/interactive/repl.go
@@ -26,10 +26,8 @@ type replLineReader interface {
 var (
 	replConfigDir         = config.Dir
 	replNewReadline       = func(cfg *readline.Config) (replLineReader, error) { return readline.NewEx(cfg) }
-	replBuildSystemPrompt = func(osName, shellName, shellVersion, cwd string, toolsEnabled bool) string {
-		return llm.BuildSystemPrompt(osName, shellName, shellVersion, cwd, false, toolsEnabled)
-	}
-	replHandleResponse = handleResponse
+	replBuildSystemPrompt = llm.BuildSystemPrompt
+	replHandleResponse    = handleResponse
 )
 
 // BuiltinCommands holds handlers for built-in REPL commands so the interactive
@@ -45,7 +43,7 @@ type BuiltinCommands struct {
 	MemoryRun func(args []string) error
 }
 
-func Run(version string, cmds BuiltinCommands, cfg *config.Config, client llm.Client, shellInfo shell.Info) error {
+func Run(version string, cmds BuiltinCommands, cfg *config.Config, client llm.Client, shellInfo shell.Info, explain bool) error {
 	configDir, err := replConfigDir()
 	if err != nil {
 		return err
@@ -63,7 +61,7 @@ func Run(version string, cmds BuiltinCommands, cfg *config.Config, client llm.Cl
 	fmt.Printf("AI CLI %s — interactive mode. Type 'help' for commands or 'exit' to quit.\n", version)
 
 	toolsEnabled := cfg.Safety.ToolCalling != config.ToolCallingNever
-	systemPrompt := replBuildSystemPrompt(shellInfo.OS, shellInfo.Shell, shellInfo.Version, "", toolsEnabled)
+	systemPrompt := replBuildSystemPrompt(shellInfo.OS, shellInfo.Shell, shellInfo.Version, "", explain, toolsEnabled)
 
 	for {
 		line, err := rl.Readline()
@@ -140,7 +138,7 @@ func Run(version string, cmds BuiltinCommands, cfg *config.Config, client llm.Cl
 				color.Red("Error: %v", err)
 				continue
 			}
-			if err := replHandleResponse(resp, cfg, shellInfo); err != nil {
+			if err := replHandleResponse(resp, cfg, shellInfo, explain); err != nil {
 				color.Red("Error: %v", err)
 			}
 		}
@@ -167,10 +165,10 @@ func printHelp() {
 	fmt.Println()
 }
 
-func handleResponse(resp *llm.Response, cfg *config.Config, shellInfo shell.Info) error {
+func handleResponse(resp *llm.Response, cfg *config.Config, shellInfo shell.Info, explain bool) error {
 	switch resp.Type {
 	case "commands":
-		return executor.Run(resp.Commands, cfg, shellInfo, false)
+		return executor.Run(resp.Commands, cfg, shellInfo, explain)
 	case "config":
 		return applyConfig(resp, cfg)
 	default:
diff --git a/internal/interactive/repl_run_test.go b/internal/interactive/repl_run_test.go
index 5bed3f8..c614525 100644
--- a/internal/interactive/repl_run_test.go
+++ b/internal/interactive/repl_run_test.go
@@ -17,25 +17,22 @@ import (
 )
 
 type fakeClient struct {
-	chat func(systemPrompt, userMessage string) (*llm.Response, error)
+	chat         func(systemPrompt, userMessage string) (*llm.Response, error)
+	chatMessages func(messages []llm.Message) (*llm.Response, error)
 }
 
 func (f fakeClient) Chat(systemPrompt, userMessage string) (*llm.Response, error) {
+	if f.chat == nil {
+		return nil, errors.New("fakeClient.Chat not implemented")
+	}
 	return f.chat(systemPrompt, userMessage)
 }
 
 func (f fakeClient) ChatMessages(messages []llm.Message) (*llm.Response, error) {
-	// Extract system and user message for compatibility with existing tests
-	var system, user string
-	for _, m := range messages {
-		switch m.Role {
-		case "system":
-			system = m.Content
-		case "user":
-			user = m.Content
-		}
+	if f.chatMessages == nil {
+		return nil, errors.New("fakeClient.ChatMessages not implemented")
 	}
-	return f.chat(system, user)
+	return f.chatMessages(messages)
 }
 
 type fakeReadlineStep struct {
@@ -107,7 +104,7 @@ func TestRun_ConfigDirError(t *testing.T) {
 	err := Run("dev", BuiltinCommands{}, testCfg(t), fakeClient{chat: func(string, string) (*llm.Response, error) {
 		t.Fatal("client.Chat should not be called")
 		return nil, nil
-	}}, shell.Info{})
+	}}, shell.Info{}, false)
 	if err == nil {
 		t.Fatal("Run() error = nil, want error")
 	}
@@ -126,7 +123,7 @@ func TestRun_ReadlineInitError(t *testing.T) {
 	err := Run("dev", BuiltinCommands{}, testCfg(t), fakeClient{chat: func(string, string) (*llm.Response, error) {
 		t.Fatal("client.Chat should not be called")
 		return nil, nil
-	}}, shell.Info{})
+	}}, shell.Info{}, false)
 	if err == nil {
 		t.Fatal("Run() error = nil, want error")
 	}
@@ -155,7 +152,7 @@ func TestRun_ExitOnEOFAndInterrupt(t *testing.T) {
 				err := Run("dev", BuiltinCommands{}, testCfg(t), fakeClient{chat: func(string, string) (*llm.Response, error) {
 					t.Fatal("client.Chat should not be called")
 					return nil, nil
-				}}, shell.Info{})
+				}}, shell.Info{}, false)
 				if err != nil {
 					t.Fatalf("Run() error = %v, want nil", err)
 				}
@@ -180,7 +177,7 @@ func TestRun_ReadlineUnexpectedError(t *testing.T) {
 	err := Run("dev", BuiltinCommands{}, testCfg(t), fakeClient{chat: func(string, string) (*llm.Response, error) {
 		t.Fatal("client.Chat should not be called")
 		return nil, nil
-	}}, shell.Info{})
+	}}, shell.Info{}, false)
 	if err == nil {
 		t.Fatal("Run() error = nil, want error")
 	}
@@ -212,13 +209,13 @@ func TestRun_DispatchesBuiltinsAndLLM(t *testing.T) {
 	replNewReadline = func(*readline.Config) (replLineReader, error) { return rl, nil }
 
 	var promptArgs []string
-	replBuildSystemPrompt = func(osName, shellName, shellVersion, cwd string, _ bool) string {
+	replBuildSystemPrompt = func(osName, shellName, shellVersion, cwd string, _, _ bool) string {
 		promptArgs = []string{osName, shellName, shellVersion, cwd}
 		return "system-prompt"
 	}
 
 	var handled []*llm.Response
-	replHandleResponse = func(resp *llm.Response, _ *config.Config, info shell.Info) error {
+	replHandleResponse = func(resp *llm.Response, _ *config.Config, info shell.Info, _ bool) error {
 		if info.Shell != "powershell" {
 			t.Fatalf("shellInfo not passed through: %+v", info)
 		}
@@ -260,7 +257,7 @@ func TestRun_DispatchesBuiltinsAndLLM(t *testing.T) {
 		OS:      "windows/amd64",
 		Shell:   "powershell",
 		Version: "7.5.0",
-	})
+	}, false)
 	if err != nil {
 		t.Fatalf("Run() error = %v", err)
 	}
@@ -305,10 +302,10 @@ func TestRun_ContinuesAfterBuiltinLLMAndHandleErrors(t *testing.T) {
 		},
 	}
 	replNewReadline = func(*readline.Config) (replLineReader, error) { return rl, nil }
-	replBuildSystemPrompt = func(_, _, _, _ string, _ bool) string { return "sys" }
+	replBuildSystemPrompt = func(_, _, _, _ string, _ bool, _ bool) string { return "sys" }
 
 	handleCalls := 0
-	replHandleResponse = func(_ *llm.Response, _ *config.Config, _ shell.Info) error {
+	replHandleResponse = func(_ *llm.Response, _ *config.Config, _ shell.Info, _ bool) error {
 		handleCalls++
 		return errors.New("handle failed")
 	}
@@ -332,7 +329,7 @@ func TestRun_ContinuesAfterBuiltinLLMAndHandleErrors(t *testing.T) {
 	}
 
 	out := captureOutput(t, func() {
-		if err := Run("dev", cmds, testCfg(t), client, shell.Info{}); err != nil {
+		if err := Run("dev", cmds, testCfg(t), client, shell.Info{}, false); err != nil {
 			t.Fatalf("Run() error = %v, want nil", err)
 		}
 	})
@@ -350,3 +347,42 @@ func TestRun_ContinuesAfterBuiltinLLMAndHandleErrors(t *testing.T) {
 		t.Fatalf("output missing Bye!\n%s", out)
 	}
 }
+
+func TestRun_ExplainFlagPropagated(t *testing.T) {
+	stubReplHooks(t)
+	replConfigDir = func() (string, error) { return t.TempDir(), nil }
+	rl := &fakeReadline{
+		steps: []fakeReadlineStep{
+			{line: "do something"},
+			{line: "exit"},
+		},
+	}
+	replNewReadline = func(*readline.Config) (replLineReader, error) { return rl, nil }
+
+	var capturedPromptExplain bool
+	replBuildSystemPrompt = func(_, _, _, _ string, explain, _ bool) string {
+		capturedPromptExplain = explain
+		return "sys"
+	}
+
+	var capturedHandleExplain bool
+	replHandleResponse = func(_ *llm.Response, _ *config.Config, _ shell.Info, explain bool) error {
+		capturedHandleExplain = explain
+		return nil
+	}
+
+	client := fakeClient{chat: func(_, _ string) (*llm.Response, error) {
+		return &llm.Response{Type: "commands"}, nil
+	}}
+
+	if err := Run("dev", BuiltinCommands{}, testCfg(t), client, shell.Info{}, true); err != nil {
+		t.Fatalf("Run() error = %v", err)
+	}
+
+	if !capturedPromptExplain {
+		t.Fatal("replBuildSystemPrompt: explain = false, want true")
+	}
+	if !capturedHandleExplain {
+		t.Fatal("replHandleResponse: explain = false, want true")
+	}
+}
diff --git a/internal/interactive/repl_test.go b/internal/interactive/repl_test.go
index 8647c55..b07561b 100644
--- a/internal/interactive/repl_test.go
+++ b/internal/interactive/repl_test.go
@@ -45,7 +45,7 @@ func TestPrintHelp(_ *testing.T) {
 }
 
 func TestHandleResponseCommands(t *testing.T) {
-	err := handleResponse(&llm.Response{Type: "commands"}, testCfg(t), shell.Info{Shell: "cmd"})
+	err := handleResponse(&llm.Response{Type: "commands"}, testCfg(t), shell.Info{Shell: "cmd"}, false)
 	if err != nil {
 		t.Fatalf("handleResponse(commands) error = %v", err)
 	}
@@ -61,7 +61,7 @@ func TestHandleResponseConfig(t *testing.T) {
 	}
 
 	withStdin(t, "y\n", func() {
-		if err := handleResponse(resp, cfg, shell.Info{}); err != nil {
+		if err := handleResponse(resp, cfg, shell.Info{}, false); err != nil {
 			t.Fatalf("handleResponse(config) error = %v", err)
 		}
 	})
@@ -72,7 +72,7 @@ func TestHandleResponseConfig(t *testing.T) {
 }
 
 func TestHandleResponseUnknownType(t *testing.T) {
-	err := handleResponse(&llm.Response{Type: "mystery"}, testCfg(t), shell.Info{})
+	err := handleResponse(&llm.Response{Type: "mystery"}, testCfg(t), shell.Info{}, false)
 	if err == nil {
 		t.Fatal("handleResponse() error = nil, want error")
 	}
diff --git a/internal/shell/detect.go b/internal/shell/detect.go
index f77a1a4..919e92d 100644
--- a/internal/shell/detect.go
+++ b/internal/shell/detect.go
@@ -10,6 +10,7 @@ import (
 
 const (
 	shellPowershell = "powershell"
+	shellPwsh       = "pwsh"
 	shellWindows    = "windows"
 	shellUnknown    = "unknown"
 )
@@ -81,7 +82,7 @@ func detectShellVersion(shell string) string {
 	switch base {
 	case "bash", "zsh", "fish":
 		cmd = exec.CommandContext(context.Background(), shell, "--version")
-	case shellPowershell, "pwsh":
+	case shellPowershell, shellPwsh:
 		cmd = exec.CommandContext(context.Background(), shell, "-Command", "$PSVersionTable.PSVersion.ToString()")
 	default:
 		return shellUnknown
@@ -110,7 +111,7 @@ func shellBaseName(shell string) string {
 func Command(shell string) (bin string, args []string) {
 	base := shellBaseName(shell)
 	switch base {
-	case shellPowershell, "pwsh":
+	case shellPowershell, shellPwsh:
 		return shell, []string{"-Command"}
 	case "cmd":
 		return shell, []string{"/c"}
diff --git a/internal/shell/detect_other_test.go b/internal/shell/detect_other_test.go
new file mode 100644
index 0000000..2c1be74
--- /dev/null
+++ b/internal/shell/detect_other_test.go
@@ -0,0 +1,21 @@
+//go:build !windows
+
+package shell
+
+import "testing"
+
+func TestParentShellProcess_NonWindows(t *testing.T) {
+	// On non-Windows, parentShellProcess is a no-op stub that returns "".
+	got := parentShellProcess()
+	if got != "" {
+		t.Fatalf("parentShellProcess() = %q, want empty string on non-Windows", got)
+	}
+}
+
+func TestPreferredPowerShell_NonWindows(t *testing.T) {
+	// On non-Windows, preferredPowerShell returns "powershell".
+	got := preferredPowerShell()
+	if got != "powershell" {
+		t.Fatalf("preferredPowerShell() = %q, want powershell on non-Windows", got)
+	}
+}
diff --git a/internal/shell/detect_test.go b/internal/shell/detect_test.go
index 3bbff12..ef5aa0d 100644
--- a/internal/shell/detect_test.go
+++ b/internal/shell/detect_test.go
@@ -230,25 +230,3 @@ func TestDetectShellVersion_Branches(t *testing.T) {
 		}
 	}
 }
-
-func TestParentShellProcess_NonWindows(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("Skipping non-Windows test on Windows")
-	}
-	// On non-Windows, parentShellProcess is a no-op stub that returns "".
-	got := parentShellProcess()
-	if got != "" {
-		t.Fatalf("parentShellProcess() = %q, want empty string on non-Windows", got)
-	}
-}
-
-func TestPreferredPowerShell_NonWindows(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		t.Skip("Skipping non-Windows test on Windows")
-	}
-	// On non-Windows, preferredPowerShell returns "powershell".
-	got := preferredPowerShell()
-	if got != "powershell" {
-		t.Fatalf("preferredPowerShell() = %q, want powershell on non-Windows", got)
-	}
-}
diff --git a/internal/shell/detect_windows.go b/internal/shell/detect_windows.go
index 70566a5..cf1f811 100644
--- a/internal/shell/detect_windows.go
+++ b/internal/shell/detect_windows.go
@@ -79,7 +79,7 @@ func parentShellProcess() string {
 // preferredPowerShell returns the path to the best available PowerShell binary.
 // It prefers PowerShell 7+ (pwsh) over the legacy Windows PowerShell 5 (powershell).
 func preferredPowerShell() string {
-	if path, err := windowsLookPath("pwsh"); err == nil {
+	if path, err := windowsLookPath(shellPwsh); err == nil {
 		return path
 	}
 	for _, candidate := range []string{
@@ -90,12 +90,12 @@ func preferredPowerShell() string {
 			return candidate
 		}
 	}
-	return "powershell"
+	return shellPowershell
 }
 
 // pwshExePath returns the full path to pwsh.exe, falling back to "pwsh".
 func pwshExePath() string {
-	if path, err := windowsLookPath("pwsh"); err == nil {
+	if path, err := windowsLookPath(shellPwsh); err == nil {
 		return path
 	}
 	for _, candidate := range []string{
@@ -106,5 +106,5 @@ func pwshExePath() string {
 			return candidate
 		}
 	}
-	return "pwsh"
+	return shellPwsh
 }
diff --git a/internal/tools/safety.go b/internal/tools/safety.go
index 11c54f9..73cd8a8 100644
--- a/internal/tools/safety.go
+++ b/internal/tools/safety.go
@@ -49,10 +49,10 @@ var sensitiveEnvKeys = []string{
 
 // ValidatePath resolves path to an absolute path and checks that it is
 // contained within cwd and does not match any blocked pattern.
-// Symlinks are resolved so that a link inside cwd pointing outside cannot
-// bypass the containment check.
+// It also resolves symlinks and re-validates the resolved path to prevent
+// symlink attacks that could point outside cwd or to blocked files.
 func ValidatePath(path, cwd string) (string, error) {
-	// Resolve relative to cwd lexically first
+	// Resolve relative to cwd
 	var abs string
 	if filepath.IsAbs(path) {
 		abs = filepath.Clean(path)
@@ -60,9 +60,11 @@ func ValidatePath(path, cwd string) (string, error) {
 		abs = filepath.Clean(filepath.Join(cwd, path))
 	}
 
+	// Always use forward slashes for cross-platform consistency in tests
+	// but convert back to OS specific separators for internal checks
 	cwdClean := filepath.Clean(cwd)
 
-	// Ensure the lexical path is under cwd
+	// Ensure the path is under cwd
 	if abs != cwdClean && !strings.HasPrefix(abs, cwdClean+string(os.PathSeparator)) {
 		return "", fmt.Errorf("path %q is outside the working directory", path)
 	}
diff --git a/internal/tools/safety_test.go b/internal/tools/safety_test.go
index 8b50765..a7bc26e 100644
--- a/internal/tools/safety_test.go
+++ b/internal/tools/safety_test.go
@@ -207,8 +207,8 @@ func TestValidatePath_SymlinkOutsideCWD(t *testing.T) {
 	if err == nil {
 		t.Error("ValidatePath should have rejected symlink pointing outside cwd")
 	}
-	if err != nil && !strings.Contains(err.Error(), "outside") {
-		t.Errorf("expected 'outside' in error, got: %v", err)
+	if err != nil && !strings.Contains(err.Error(), "outside") && !strings.Contains(err.Error(), "resolves outside") {
+		t.Errorf("expected 'outside' or 'resolves outside' in error, got: %v", err)
 	}
 }
 
diff --git a/internal/tools/tools.go b/internal/tools/tools.go
index 7d279c4..3978af5 100644
--- a/internal/tools/tools.go
+++ b/internal/tools/tools.go
@@ -9,6 +9,7 @@ import (
 	"os/exec"
 	"path/filepath"
 	"runtime"
+	"sort"
 	"strings"
 	"time"
 
@@ -334,6 +335,7 @@ func execDiskUsage() (string, error) {
 
 func execEnvironment() string {
 	vars := os.Environ()
+	sort.Strings(vars)
 	filtered := FilterEnvironment(vars)
 	return strings.Join(filtered, "\n")
 }
@@ -342,5 +344,11 @@ func truncateOutput(s string) string {
 	if len(s) <= maxOutputBytes {
 		return s
 	}
+
+	lastNewline := strings.LastIndex(s[:maxOutputBytes], "\n")
+	if lastNewline > maxOutputBytes/2 {
+		return s[:lastNewline] + "\n... [output truncated]"
+	}
+
 	return s[:maxOutputBytes] + "\n... [output truncated]"
 }