feat: new version

* refactor: cleanup duplicated logic

* refactor: cleanup config.rs

* refactor: cleanup explain.rs

* refactor: cleanup instance.rs

* refactor: cleanup

* refactor: use UnixStream instead of status file

* Refactor vm lifetime (#8)

* fix: handle vm supervisor being killed

* fix: fixed the loop connection retry

* refactor: extracted vm_manager liveness check logic

* Script failure report (#9)

* feat: added script failure report

* feat: vm error report can also report ssh.sh

* refactor: liveness check when connecting to ssh

* fix: fixed the wrong InstanceError::VMError

* fix: fixed the is_lock_stable

Cargo.lock

@@ -82,6 +82,12 @@ dependencies = [
  "windows-sys 0.61.2",
 ]
 
+[[package]]
+name = "anyhow"
+version = "1.0.101"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5f0e0fee31ef5ed1ba1316088939cea399010ed7731dba877ed44aeb407a75ea"
+
 [[package]]
 name = "assert_cmd"
 version = "2.1.2"
@@ -144,6 +150,15 @@ version = "3.19.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "5dd9dc738b7a8311c7ade152424974d8115f2cdad61e8dab8dac9f2362298510"
 
+[[package]]
+name = "bytesize"
+version = "2.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6bd91ee7b2422bcb158d90ef4d14f75ef67f340943fc4149891dcce8f8b972a3"
+dependencies = [
+ "serde_core",
+]
+
 [[package]]
 name = "cassowary"
 version = "0.3.0"
@@ -1303,10 +1318,12 @@ checksum = "ba73ea9cf16a25df0c8caa16c51acb937d5712a8429db78a3ee29d5dcacd3a65"
 
 [[package]]
 name = "vibebox"
-version = "0.2.4"
+version = "0.3.2"
 dependencies = [
+ "anyhow",
  "assert_cmd",
  "block2",
+ "bytesize",
  "clap",
  "color-eyre",
  "crossterm",

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "vibebox"
-version = "0.2.4"
+version = "0.3.2"
 edition = "2024"
 authors = ["Finn Sheng"]
 description = "Ultrafast CLI on Apple Silicon macOS for fast, sandboxed development and LLM agents."
@@ -42,6 +42,8 @@ ratatui = { version = "0.29.0", features = ["unstable-rendered-line-info"] }
 tracing = "0.1"
 tracing-subscriber = { version = "0.3", features = ["env-filter"] }
 dialoguer = "0.12.0"
+bytesize = {version = "2.3.1",features = ["serde"]}
+anyhow = "1.0.101"
 
 [dev-dependencies]
 assert_cmd = "2"

docs/tasks.md

@@ -25,7 +25,7 @@
 5. [x] Implement rendering functions for header, terminal area, input area, completions, and status bar.
 6. [x] Implement async event loop (keyboard, resize, tick) with crossterm EventStream + tokio.
 7. [x] Add a standalone TUI CLI binary (no main.rs wiring) with placeholder VM info and TODOs for VM integration.
-8. [ ] Run tests and validate coverage for the new module.
+8. [x] Run tests and validate coverage for the new module.
 
 ## TUI
 
@@ -49,19 +49,23 @@
    4. [x] set up the cli.
 9. [x] fix ui overlap, and consistency issue.
 10. [x] `purge-cache` to clear the cache.
-11. [ ] intensive integration test.
+11. [x] intensive integration test.
 
 ## Publish
 
-1. [ ] write the docs.
+1. [x] write the docs.
-2. [ ] setup quick install link.
+2. [x] setup quick install link.
-3. [ ] setup website.
+3. [x] setup website.
 
 ## Stage 2
 
 1. [ ] retouch the cli ux.
-2. [ ] refactor the code.
+2. [x] refactor the code.
-3. [ ] Redirect vm output to log.
+3. [ ] refactor the mount system.
-4. [ ] Redirect vm output to vibebox starting it.
+4. [x] refactor the vm process lifetime. 
-5. [ ] use anyhow to sync api.
+5. [x] Redirect vm output to log.
-6. [ ] add support for ipv6.
+6. [x] Redirect vm output to vibebox starting it.
+7. [x] use anyhow to sync api.
+8. [ ] add support for ipv6.
+9. [x] use UnixStream instead of status file
+10. [x] liveness check should also happen when waiting for ssh port

src/bin/vibebox-supervisor.rs

@@ -21,15 +21,16 @@ fn main() -> Result<()> {
     color_eyre::install()?;
 
     tracing::info!("starting vm supervisor");
-    let cwd = env::current_dir().map_err(|err| color_eyre::eyre::eyre!(err.to_string()))?;
+    let project_dir = env::current_dir().map_err(|err| color_eyre::eyre::eyre!(err.to_string()))?;
-    let config = config::load_config(&cwd);
+    let config = config::load_config(&project_dir)
-    let instance_dir = instance::ensure_instance_dir(&cwd)
         .map_err(|err| color_eyre::eyre::eyre!(err.to_string()))?;
-    let _ = instance::touch_last_active(&instance_dir);
+    let _ = instance::ensure_instance_dir(&project_dir)
+        .map_err(|err| color_eyre::eyre::eyre!(err.to_string()))?;
+    let _ = instance::touch_last_active(&project_dir);
     let args = vm::VmArg {
         cpu_count: config.box_cfg.cpu_count,
-        ram_bytes: config.box_cfg.ram_mb.saturating_mul(1024 * 1024),
+        ram_bytes: config.box_cfg.ram_size.as_u64(),
-        disk_bytes: config.box_cfg.disk_gb.saturating_mul(1024 * 1024 * 1024),
+        disk_bytes: config.box_cfg.disk_size.as_u64(),
         no_default_mounts: false,
         mounts: config.box_cfg.mounts.clone(),
     };
@@ -37,7 +38,7 @@ fn main() -> Result<()> {
     tracing::info!(auto_shutdown_ms, "vm supervisor config");
 
     let result = vm_manager::run_manager(args, auto_shutdown_ms);
-    let _ = instance::touch_last_active(&instance_dir);
+    let _ = instance::touch_last_active(&project_dir);
     if let Err(err) = result {
         tracing::error!(error = %err, "vm supervisor exited");
         return Err(color_eyre::eyre::eyre!(err.to_string()));

src/bin/vibebox.rs

@@ -1,3 +1,7 @@
+use bytesize::ByteSize;
+use clap::Parser;
+use color_eyre::Result;
+use dialoguer::Confirm;
 use std::{
     env,
     ffi::OsString,
@@ -6,10 +10,6 @@ use std::{
     path::{Path, PathBuf},
     sync::{Arc, Mutex},
 };
-
-use clap::Parser;
-use color_eyre::Result;
-use dialoguer::Confirm;
 use time::OffsetDateTime;
 use time::format_description::well_known::Rfc3339;
 use tracing_subscriber::filter::LevelFilter;
@@ -17,8 +17,9 @@ use tracing_subscriber::registry::Registry;
 use tracing_subscriber::{EnvFilter, fmt, prelude::*, reload};
 
 use vibebox::tui::{AppState, VmInfo};
+use vibebox::utils::relative_to_home;
 use vibebox::{
-    SessionManager, commands, config, explain, instance, session_manager, tui, vm, vm_manager,
+    SessionManager, commands, config, explain, instance, session_manager, tui, vm_manager,
 };
 
 #[derive(Debug, Parser)]
@@ -56,53 +57,26 @@ fn main() -> Result<()> {
 
     let config_override = cli.config.clone();
     let raw_args: Vec<OsString> = env::args_os().collect();
-    let config = config::load_config_with_path(&cwd, config_override.as_deref());
+    let config = config::load_config_with_path(&cwd, config_override.as_deref())
+        .map_err(|err| color_eyre::eyre::eyre!(err.to_string()))?;
 
-    if env::var("VIBEBOX_VM_MANAGER").as_deref() == Ok("1") {
-        tracing::info!("starting vm manager mode");
-        let args = vm::VmArg {
-            cpu_count: config.box_cfg.cpu_count,
-            ram_bytes: config.box_cfg.ram_mb.saturating_mul(1024 * 1024),
-            disk_bytes: config.box_cfg.disk_gb.saturating_mul(1024 * 1024 * 1024),
-            no_default_mounts: false,
-            mounts: config.box_cfg.mounts.clone(),
-        };
-        let auto_shutdown_ms = config.supervisor.auto_shutdown_ms;
-        tracing::info!(auto_shutdown_ms, "vm manager config");
-        if let Err(err) = vm_manager::run_manager(args, auto_shutdown_ms) {
-            tracing::error!(error = %err, "vm manager exited");
-            return Err(color_eyre::eyre::eyre!(err.to_string()));
-        }
-        return Ok(());
-    }
-
-    vm::ensure_signed();
-
-    let vm_args = vm::VmArg {
-        cpu_count: config.box_cfg.cpu_count,
-        ram_bytes: config.box_cfg.ram_mb.saturating_mul(1024 * 1024),
-        disk_bytes: config.box_cfg.disk_gb.saturating_mul(1024 * 1024 * 1024),
-        no_default_mounts: false,
-        mounts: config.box_cfg.mounts.clone(),
-    };
-    let auto_shutdown_ms = config.supervisor.auto_shutdown_ms;
     let vm_info = VmInfo {
-        max_memory_mb: vm_args.ram_bytes / (1024 * 1024),
+        max_memory: config.box_cfg.ram_size,
-        cpu_cores: vm_args.cpu_count,
+        cpu_cores: config.box_cfg.cpu_count,
-        max_disk_gb: (vm_args.disk_bytes as f32) / 1024.0 / 1024.0 / 1024.0,
+        max_disk: config.box_cfg.disk_size,
         system_name: "Debian".to_string(), // TODO: read system name from the VM.
-        auto_shutdown_ms,
+        auto_shutdown_ms: config.supervisor.auto_shutdown_ms,
     };
     if let Ok(manager) = SessionManager::new() {
         if let Err(err) = manager.update_global_sessions(&cwd) {
             tracing::warn!(error = %err, "failed to update a global session list");
         }
     } else {
-        tracing::warn!("failed to initialize session manager");
+        tracing::error!("failed to initialize session manager");
+        std::process::exit(1);
     }
     let commands = commands::build_commands();
     let app = Arc::new(Mutex::new(AppState::new(cwd.clone(), vm_info, commands)));
-
     {
         let mut locked = app.lock().expect("app state poisoned");
         tui::render_tui_once(&mut locked)?;
@@ -112,17 +86,39 @@ fn main() -> Result<()> {
         writeln!(stdout)?;
         stdout.flush()?;
     }
-    warn_disk_size_mismatch(&cwd, vm_args.disk_bytes);
+    warn_disk_size_mismatch(&cwd, config.box_cfg.disk_size);
     if let Some(handle) = stderr_handle {
         let _ = handle.modify(|filter| *filter = LevelFilter::INFO);
     }
 
-    tracing::debug!(auto_shutdown_ms, "auto shutdown config");
+    tracing::debug!(config.supervisor.auto_shutdown_ms, "auto shutdown config");
-    let manager_conn =
+    let manager_conn = vm_manager::ensure_manager(
-        vm_manager::ensure_manager(&raw_args, auto_shutdown_ms, config_override.as_deref())
+        &raw_args,
-            .map_err(|err| color_eyre::eyre::eyre!(err.to_string()))?;
+        config.supervisor.auto_shutdown_ms,
+        config_override.as_deref(),
+    )
+    .map_err(|err| {
+        tracing::error!(error = %err, "failed to ensure vm manager");
+        color_eyre::eyre::eyre!(err.to_string())
+    })?;
 
-    instance::run_with_ssh(manager_conn).map_err(|err| color_eyre::eyre::eyre!(err.to_string()))?;
+    if let Err(err) = instance::run_with_ssh(manager_conn) {
+        if let Some(instance::InstanceError::UnexpectedDisconnection) =
+            err.downcast_ref::<instance::InstanceError>()
+        {
+            tracing::warn!("vm manager disconnected; exiting vibebox");
+        } else if let Some(instance::InstanceError::VMError(vm_error)) =
+            err.downcast_ref::<instance::InstanceError>()
+        {
+            tracing::error!("[vm]: {vm_error}");
+            tracing::info!("vibecoding paused: the VM says today is a rest day 😴");
+            std::process::exit(1);
+        } else {
+            let message = err.to_string();
+            tracing::error!(error = %message, "vibebox exited: uncaught error");
+            return Err(color_eyre::eyre::eyre!(message));
+        }
+    }
 
     tracing::info!("See you again — keep vibecoding (no SEVs, only vibes) 😈");
 
@@ -205,13 +201,14 @@ fn handle_command(command: Command, cwd: &Path, config_override: Option<&Path>)
                 "Purged {} file{} totaling {} from {}",
                 file_count,
                 if file_count == 1 { "" } else { "s" },
-                format_bytes(total_bytes),
+                ByteSize(total_bytes),
                 cache_dir.display()
             );
             Ok(())
         }
         Command::Explain => {
-            let config = config::load_config_with_path(cwd, config_override);
+            let config = config::load_config_with_path(cwd, config_override)
+                .map_err(|err| color_eyre::eyre::eyre!(err.to_string()))?;
             let mounts = explain::build_mount_rows(cwd, &config)
                 .map_err(|err| color_eyre::eyre::eyre!(err.to_string()))?;
             let networks = explain::build_network_rows(cwd)
@@ -234,20 +231,6 @@ fn project_name(directory: &Path) -> String {
         .to_string()
 }
 
-fn relative_to_home(directory: &Path) -> String {
-    let Ok(home) = env::var("HOME") else {
-        return directory.display().to_string();
-    };
-    let home_path = PathBuf::from(home);
-    if let Ok(stripped) = directory.strip_prefix(&home_path) {
-        if stripped.components().next().is_none() {
-            return "~".to_string();
-        }
-        return format!("~/{}", stripped.display());
-    }
-    directory.display().to_string()
-}
-
 fn cache_dir() -> Result<PathBuf> {
     let home = env::var("HOME").map(PathBuf::from)?;
     let cache_home = env::var("XDG_CACHE_HOME")
@@ -296,23 +279,6 @@ fn measure_dir(path: &Path) -> Result<(u64, u64)> {
     Ok((file_count, total_bytes))
 }
 
-fn format_bytes(bytes: u64) -> String {
-    const KB: f64 = 1024.0;
-    const MB: f64 = KB * 1024.0;
-    const GB: f64 = MB * 1024.0;
-
-    let b = bytes as f64;
-    if b >= GB {
-        format!("{:.2} GB", b / GB)
-    } else if b >= MB {
-        format!("{:.1} MB", b / MB)
-    } else if b >= KB {
-        format!("{:.1} KB", b / KB)
-    } else {
-        format!("{} B", bytes)
-    }
-}
-
 fn format_last_active(value: Option<&str>) -> String {
     let Some(raw) = value else {
         return "-".to_string();
@@ -354,24 +320,22 @@ fn format_last_active(value: Option<&str>) -> String {
     format!("{} day{} ago", days, if days == 1 { "" } else { "s" })
 }
 
-fn warn_disk_size_mismatch(cwd: &Path, configured_bytes: u64) {
+fn warn_disk_size_mismatch(cwd: &Path, configured_size: ByteSize) {
     let instance_raw = cwd
         .join(session_manager::INSTANCE_DIR_NAME)
         .join("instance.raw");
     let Ok(meta) = fs::metadata(&instance_raw) else {
         return;
     };
-    let current_bytes = meta.len();
+    let current_size = ByteSize::b(meta.len());
-    if current_bytes == configured_bytes {
+    if current_size == configured_size {
         return;
     }
-    let current_gb = current_bytes as f64 / (1024.0 * 1024.0 * 1024.0);
-    let target_gb = configured_bytes as f64 / (1024.0 * 1024.0 * 1024.0);
     tracing::warn!(
-        "instance disk size does not match config (current {:.2} GB, config {:.2} GB). \
+        "instance disk size does not match config (current {}, config {}). \
 disk_gb applies only on init. Run `vibebox reset` to recreate or set disk_gb to match; using the existing disk.",
-        current_gb,
+        current_size,
-        target_gb
+        configured_size,
     );
 }
 
@@ -380,13 +344,13 @@ type StderrHandle = reload::Handle<LevelFilter, Registry>;
 fn init_tracing(cwd: &Path) -> Option<StderrHandle> {
     let filter = EnvFilter::try_from_default_env().unwrap_or_else(|_| EnvFilter::new("debug"));
     let file_filter = filter.clone();
-    let stderr_is_tty = std::io::stderr().is_terminal();
+    let stderr_is_tty = io::stderr().is_terminal();
     let ansi = stderr_is_tty && env::var("VIBEBOX_LOG_NO_COLOR").is_err();
     let file = instance::ensure_instance_dir(cwd)
         .ok()
         .and_then(|instance_dir| {
             let log_path = instance_dir.join("cli.log");
-            std::fs::OpenOptions::new()
+            fs::OpenOptions::new()
                 .create(true)
                 .write(true)
                 .truncate(true)
@@ -400,7 +364,7 @@ fn init_tracing(cwd: &Path) -> Option<StderrHandle> {
             .with_target(false)
             .with_ansi(ansi)
             .without_time()
-            .with_writer(std::io::stderr)
+            .with_writer(io::stderr)
             .with_filter(stderr_filter);
         let subscriber = tracing_subscriber::registry().with(stderr_layer);
         if let Some(file) = file {
@@ -418,7 +382,7 @@ fn init_tracing(cwd: &Path) -> Option<StderrHandle> {
         let stderr_layer = fmt::layer()
             .with_target(false)
             .with_ansi(ansi)
-            .with_writer(std::io::stderr)
+            .with_writer(io::stderr)
             .with_filter(filter);
         let subscriber = tracing_subscriber::registry().with(stderr_layer);
         if let Some(file) = file {

src/config.rs

@@ -1,13 +1,14 @@
+use anyhow::{Context, Error, Result, bail};
+use bytesize::ByteSize;
+use serde::{Deserialize, Serialize};
 use std::{
     env, fs, io,
     path::{Path, PathBuf},
 };
 
-use serde::{Deserialize, Serialize};
-
 use crate::vm::DirectoryShare;
 
-pub const CONFIG_FILENAME: &str = "vibebox.toml";
+const CONFIG_FILENAME: &str = "vibebox.toml";
 pub const CONFIG_PATH_ENV: &str = "VIBEBOX_CONFIG_PATH";
 
 const DEFAULT_CPU_COUNT: usize = 2;
@@ -22,20 +23,83 @@ pub struct Config {
     pub supervisor: SupervisorConfig,
 }
 
+const MI_B: u64 = 1024 * 1024;
+const GI_B: u64 = 1024 * 1024 * 1024;
+
+mod serde_mb {
+    use super::{ByteSize, MI_B};
+    use serde::{Deserialize, Deserializer, Serializer};
+
+    pub fn serialize<S>(v: &ByteSize, s: S) -> Result<S::Ok, S::Error>
+    where
+        S: Serializer,
+    {
+        let bytes = v.0;
+        if !bytes.is_multiple_of(MI_B) {
+            return Err(serde::ser::Error::custom(
+                "ram_mb must be an integer number of MB",
+            ));
+        }
+        s.serialize_u64(bytes / MI_B)
+    }
+
+    pub fn deserialize<'de, D>(d: D) -> Result<ByteSize, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        let mb = u64::deserialize(d)?;
+        let bytes = mb
+            .checked_mul(MI_B)
+            .ok_or_else(|| serde::de::Error::custom("ram_mb overflow"))?;
+        Ok(ByteSize(bytes))
+    }
+}
+
+mod serde_gb {
+    use super::{ByteSize, GI_B};
+    use serde::{Deserialize, Deserializer, Serializer};
+
+    pub fn serialize<S>(v: &ByteSize, s: S) -> Result<S::Ok, S::Error>
+    where
+        S: Serializer,
+    {
+        let bytes = v.0;
+        if !bytes.is_multiple_of(GI_B) {
+            return Err(serde::ser::Error::custom(
+                "disk_gb must be an integer number of GB",
+            ));
+        }
+        s.serialize_u64(bytes / GI_B)
+    }
+
+    pub fn deserialize<'de, D>(d: D) -> Result<ByteSize, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        let gb = u64::deserialize(d)?;
+        let bytes = gb
+            .checked_mul(GI_B)
+            .ok_or_else(|| serde::de::Error::custom("disk_gb overflow"))?;
+        Ok(ByteSize(bytes))
+    }
+}
+
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct BoxConfig {
     pub cpu_count: usize,
-    pub ram_mb: u64,
+    #[serde(rename = "ram_mb", with = "serde_mb")]
-    pub disk_gb: u64,
+    pub ram_size: ByteSize,
+    #[serde(rename = "disk_gb", with = "serde_gb")]
+    pub disk_size: ByteSize,
     pub mounts: Vec<String>,
 }
 
 impl Default for BoxConfig {
     fn default() -> Self {
         Self {
-            cpu_count: default_cpu_count(),
+            cpu_count: DEFAULT_CPU_COUNT,
-            ram_mb: default_ram_mb(),
+            ram_size: ByteSize::mib(DEFAULT_RAM_MB),
-            disk_gb: default_disk_gb(),
+            disk_size: ByteSize::gib(DEFAULT_DISK_GB),
             mounts: default_mounts(),
         }
     }
@@ -49,23 +113,11 @@ pub struct SupervisorConfig {
 impl Default for SupervisorConfig {
     fn default() -> Self {
         Self {
-            auto_shutdown_ms: default_auto_shutdown_ms(),
+            auto_shutdown_ms: DEFAULT_AUTO_SHUTDOWN_MS,
         }
     }
 }
 
-fn default_cpu_count() -> usize {
-    DEFAULT_CPU_COUNT
-}
-
-fn default_ram_mb() -> u64 {
-    DEFAULT_RAM_MB
-}
-
-fn default_auto_shutdown_ms() -> u64 {
-    DEFAULT_AUTO_SHUTDOWN_MS
-}
-
 fn default_mounts() -> Vec<String> {
     vec![
         "~/.codex:~/.codex:read-write".into(),
@@ -73,19 +125,12 @@ fn default_mounts() -> Vec<String> {
     ]
 }
 
-fn default_disk_gb() -> u64 {
-    DEFAULT_DISK_GB
-}
-
 pub fn config_path(project_root: &Path) -> PathBuf {
     project_root.join(CONFIG_FILENAME)
 }
 
-pub fn ensure_config_file(
+pub fn ensure_config_file(project_root: &Path, override_path: Option<&Path>) -> Result<PathBuf> {
-    project_root: &Path,
+    let path = resolve_config_path(project_root, override_path)?;
-    override_path: Option<&Path>,
-) -> Result<PathBuf, io::Error> {
-    let path = resolve_config_path(project_root, override_path);
     if !path.exists() {
         let default_config = Config::default();
         let contents = toml::to_string_pretty(&default_config).unwrap_or_default();
@@ -95,32 +140,24 @@ pub fn ensure_config_file(
     Ok(path)
 }
 
-pub fn load_config(project_root: &Path) -> Config {
+pub fn load_config(project_root: &Path) -> Result<Config> {
     load_config_with_path(project_root, None)
 }
 
-pub fn load_config_with_path(project_root: &Path, override_path: Option<&Path>) -> Config {
+pub fn load_config_with_path(project_root: &Path, override_path: Option<&Path>) -> Result<Config> {
-    let path = match ensure_config_file(project_root, override_path) {
+    let path =
-        Ok(path) => path,
+        ensure_config_file(project_root, override_path).context("failed to create config")?;
-        Err(err) => die(&format!("failed to create config: {err}")),
+    let raw = fs::read_to_string(&path).context("failed to read config")?;
-    };
-    let raw = match fs::read_to_string(&path) {
-        Ok(raw) => raw,
-        Err(err) => die(&format!("failed to read config: {err}")),
-    };
     let trimmed = raw.trim();
     tracing::debug!(path = %path.display(), bytes = raw.len(), "loaded vibebox config");
     if trimmed.is_empty() {
-        die(&format!(
+        bail!(format!(
             "config file ({}) is empty. Required fields: [box].cpu_count (integer), [box].ram_mb (integer), [box].disk_gb (integer), [box].mounts (array of strings), [supervisor].auto_shutdown_ms (integer)",
             path.display()
         ));
     }
 
-    let value: toml::Value = match toml::from_str(trimmed) {
+    let value: toml::Value = toml::from_str(trimmed).context("invalid config")?;
-        Ok(value) => value,
-        Err(err) => die(&format!("invalid config: {err}")),
-    };
     let schema_errors = validate_schema(&value);
     if !schema_errors.is_empty() {
         let message = format!(
@@ -128,27 +165,28 @@ pub fn load_config_with_path(project_root: &Path, override_path: Option<&Path>)
             path.display(),
             schema_errors.join("\n- ")
         );
-        die(&message);
+        bail!(message);
     }
 
-    let config: Config = match toml::from_str(trimmed) {
+    let config: Config = toml::from_str(trimmed).context("invalid config")?;
-        Ok(config) => config,
+    validate_config(&config).map_err(Error::msg)?;
-        Err(err) => die(&format!("invalid config: {err}")),
+    Ok(config)
-    };
-    validate_or_exit(&config);
-    config
 }
 
-fn resolve_config_path(project_root: &Path, override_path: Option<&Path>) -> PathBuf {
+fn resolve_config_path(project_root: &Path, override_path: Option<&Path>) -> Result<PathBuf> {
-    let root = match fs::canonicalize(project_root) {
+    let env_override = env::var_os(CONFIG_PATH_ENV).map(PathBuf::from);
-        Ok(root) => root,
+    resolve_config_path_inner(project_root, override_path, env_override)
-        Err(err) => die(&format!("failed to resolve project root: {err}")),
+}
-    };
 
-    let override_path = override_path
+fn resolve_config_path_inner(
-        .map(PathBuf::from)
+    project_root: &Path,
-        .or_else(|| env::var_os(CONFIG_PATH_ENV).map(PathBuf::from));
+    override_path: Option<&Path>,
-    let raw_path = if let Some(path) = override_path {
+    env_override: Option<PathBuf>,
+) -> Result<PathBuf> {
+    let root = fs::canonicalize(project_root).context("failed to resolve project root")?;
+
+    let selected_path = override_path.map(PathBuf::from).or(env_override);
+    let raw_path = if let Some(path) = selected_path {
         if path.is_absolute() {
             path
         } else {
@@ -159,14 +197,50 @@ fn resolve_config_path(project_root: &Path, override_path: Option<&Path>) -> Pat
     };
 
     let normalized = normalize_path(&raw_path);
-    if !normalized.starts_with(&root) {
+    let resolved =
-        die(&format!(
+        resolve_path_for_boundary_check(&normalized).context("failed to resolve config path")?;
+    if !resolved.starts_with(&root) {
+        bail!(
             "config path must be within {}: {}",
             root.display(),
-            normalized.display()
+            resolved.display()
-        ));
+        );
     }
-    normalized
+    Ok(normalized)
+}
+
+fn resolve_path_for_boundary_check(path: &Path) -> Result<PathBuf, io::Error> {
+    if path.exists() {
+        return fs::canonicalize(path);
+    }
+    let (ancestor, missing) = nearest_existing_ancestor(path)?;
+    let mut resolved = fs::canonicalize(ancestor)?;
+    for part in missing {
+        resolved.push(part);
+    }
+    Ok(resolved)
+}
+
+fn nearest_existing_ancestor(path: &Path) -> Result<(&Path, Vec<std::ffi::OsString>), io::Error> {
+    let mut current = path;
+    let mut missing = Vec::new();
+    while !current.exists() {
+        let name = current.file_name().ok_or_else(|| {
+            io::Error::new(
+                io::ErrorKind::InvalidInput,
+                format!("path has no existing ancestor: {}", path.display()),
+            )
+        })?;
+        missing.push(name.to_os_string());
+        current = current.parent().ok_or_else(|| {
+            io::Error::new(
+                io::ErrorKind::InvalidInput,
+                format!("path has no parent: {}", path.display()),
+            )
+        })?;
+    }
+    missing.reverse();
+    Ok((current, missing))
 }
 
 fn normalize_path(path: &Path) -> PathBuf {
@@ -263,27 +337,343 @@ fn validate_string_array(
     }
 }
 
-fn validate_or_exit(config: &Config) {
+fn validate_config(config: &Config) -> Result<(), String> {
     if config.box_cfg.cpu_count == 0 {
-        die("box.cpu_count must be >= 1");
+        return Err("box.cpu_count must be >= 1".to_string());
     }
-    if config.box_cfg.ram_mb == 0 {
+    if config.box_cfg.ram_size.as_mib() == 0.0 {
-        die("box.ram_mb must be >= 1");
+        return Err("box.ram_mb must be >= 1".to_string());
     }
-    if config.box_cfg.disk_gb == 0 {
+    if config.box_cfg.disk_size.as_gib() == 0.0 {
-        die("box.disk_gb must be >= 1");
+        return Err("box.disk_gb must be >= 1".to_string());
     }
     if config.supervisor.auto_shutdown_ms == 0 {
-        die("supervisor.auto_shutdown_ms must be >= 1");
+        return Err("supervisor.auto_shutdown_ms must be >= 1".to_string());
     }
     for spec in &config.box_cfg.mounts {
         if let Err(err) = DirectoryShare::from_mount_spec(spec) {
-            die(&format!("invalid mount spec '{spec}': {err}"));
+            return Err(format!("invalid mount spec '{spec}': {err}"));
         }
     }
+    Ok(())
 }
 
-fn die(message: &str) -> ! {
+#[cfg(test)]
-    tracing::error!("{message}");
+mod tests {
-    std::process::exit(1);
+    use super::*;
+    use std::ffi::OsString;
+    use std::sync::Mutex;
+    use tempfile::TempDir;
+
+    static ENV_MUTEX: Mutex<()> = Mutex::new(());
+
+    struct EnvGuard {
+        key: &'static str,
+        previous: Option<OsString>,
+    }
+
+    impl EnvGuard {
+        fn set(key: &'static str, value: &Path) -> Self {
+            let previous = env::var_os(key);
+            unsafe {
+                env::set_var(key, value);
+            }
+            Self { key, previous }
+        }
+    }
+
+    impl Drop for EnvGuard {
+        fn drop(&mut self) {
+            match &self.previous {
+                Some(value) => unsafe {
+                    env::set_var(self.key, value);
+                },
+                None => unsafe {
+                    env::remove_var(self.key);
+                },
+            }
+        }
+    }
+
+    #[test]
+    fn default_config_serializes_with_legacy_keys() {
+        let cfg = Config::default();
+        let serialized = toml::to_string(&cfg).expect("default config should serialize");
+
+        assert!(serialized.contains("ram_mb = 2048"));
+        assert!(serialized.contains("disk_gb = 5"));
+        assert!(!serialized.contains("ram_size"));
+        assert!(!serialized.contains("disk_size"));
+    }
+
+    #[test]
+    fn config_deserializes_sizes_from_mb_and_gb() {
+        let raw = r#"
+[box]
+cpu_count = 4
+ram_mb = 3072
+disk_gb = 12
+mounts = ["~/src:~/src:read-write"]
+
+[supervisor]
+auto_shutdown_ms = 15000
+"#;
+        let cfg: Config = toml::from_str(raw).expect("config should deserialize");
+
+        assert_eq!(cfg.box_cfg.cpu_count, 4);
+        assert_eq!(cfg.box_cfg.ram_size.as_u64(), ByteSize::mib(3072).as_u64());
+        assert_eq!(cfg.box_cfg.disk_size.as_u64(), ByteSize::gib(12).as_u64());
+        assert_eq!(cfg.supervisor.auto_shutdown_ms, 15000);
+    }
+
+    #[test]
+    fn serialize_rejects_non_integral_mb_or_gb() {
+        let cfg = Config {
+            box_cfg: BoxConfig {
+                cpu_count: 2,
+                ram_size: ByteSize::b((2 * MI_B) + 1),
+                disk_size: ByteSize::gib(5),
+                mounts: default_mounts(),
+            },
+            supervisor: SupervisorConfig::default(),
+        };
+
+        let err = toml::to_string(&cfg).expect_err("serialization should reject invalid MB");
+        assert!(
+            err.to_string()
+                .contains("ram_mb must be an integer number of MB")
+        );
+
+        let cfg = Config {
+            box_cfg: BoxConfig {
+                cpu_count: 2,
+                ram_size: ByteSize::mib(2048),
+                disk_size: ByteSize::b((5 * GI_B) + 1),
+                mounts: default_mounts(),
+            },
+            supervisor: SupervisorConfig::default(),
+        };
+
+        let err = toml::to_string(&cfg).expect_err("serialization should reject invalid GB");
+        assert!(
+            err.to_string()
+                .contains("disk_gb must be an integer number of GB")
+        );
+    }
+
+    #[test]
+    fn normalize_path_removes_dot_and_parent_components() {
+        let normalized = normalize_path(Path::new("/tmp/project/./nested/../config.toml"));
+        assert_eq!(normalized, PathBuf::from("/tmp/project/config.toml"));
+    }
+
+    #[test]
+    fn validate_schema_returns_errors_for_missing_required_fields() {
+        let value: toml::Value = toml::from_str(
+            r#"
+[box]
+cpu_count = 2
+"#,
+        )
+        .expect("toml should parse");
+
+        let errors = validate_schema(&value);
+
+        assert!(errors.iter().any(|e| e == "missing [supervisor] table"));
+        assert!(errors.iter().any(|e| e == "missing [box].ram_mb (integer)"));
+        assert!(
+            errors
+                .iter()
+                .any(|e| e == "missing [box].disk_gb (integer)")
+        );
+        assert!(
+            errors
+                .iter()
+                .any(|e| e == "missing [box].mounts (array of strings)")
+        );
+    }
+
+    #[test]
+    fn validate_schema_errors_when_supervisor_is_not_table() {
+        let value: toml::Value = toml::from_str(
+            r#"
+supervisor = 123
+
+[box]
+cpu_count = 2
+ram_mb = 2048
+disk_gb = 5
+mounts = []
+"#,
+        )
+        .expect("toml should parse");
+
+        let errors = validate_schema(&value);
+        assert!(errors.iter().any(|e| e == "[supervisor] must be a table"));
+    }
+
+    #[test]
+    fn ensure_config_file_creates_default_config_if_absent() {
+        let temp = TempDir::new().expect("temp dir should be created");
+        let root = fs::canonicalize(temp.path()).expect("temp dir should canonicalize");
+
+        let path = ensure_config_file(&root, None).expect("config should be created");
+        let raw = fs::read_to_string(&path).expect("created config should be readable");
+        let parsed: Config = toml::from_str(&raw).expect("created config should be valid");
+
+        assert_eq!(path, root.join("vibebox.toml"));
+        assert_eq!(parsed.box_cfg.cpu_count, DEFAULT_CPU_COUNT);
+        assert_eq!(
+            parsed.box_cfg.ram_size.as_u64(),
+            ByteSize::mib(DEFAULT_RAM_MB).as_u64()
+        );
+        assert_eq!(
+            parsed.box_cfg.disk_size.as_u64(),
+            ByteSize::gib(DEFAULT_DISK_GB).as_u64()
+        );
+    }
+
+    #[test]
+    fn load_config_creates_and_loads_default_config() {
+        let _lock = ENV_MUTEX.lock().expect("env lock should be acquired");
+        let temp = TempDir::new().expect("temp dir should be created");
+        let root = fs::canonicalize(temp.path()).expect("temp dir should canonicalize");
+        let home = root.join("home");
+        fs::create_dir_all(home.join(".codex")).expect("home .codex should be created");
+        fs::create_dir_all(home.join(".claude")).expect("home .claude should be created");
+        let _home_guard = EnvGuard::set("HOME", &home);
+
+        let cfg = load_config(&root).expect("load_config should succeed");
+
+        assert_eq!(cfg.box_cfg.cpu_count, DEFAULT_CPU_COUNT);
+        assert_eq!(
+            cfg.box_cfg.ram_size.as_u64(),
+            ByteSize::mib(DEFAULT_RAM_MB).as_u64()
+        );
+        assert_eq!(
+            cfg.box_cfg.disk_size.as_u64(),
+            ByteSize::gib(DEFAULT_DISK_GB).as_u64()
+        );
+        assert!(root.join("vibebox.toml").exists());
+    }
+
+    #[test]
+    fn load_config_with_path_uses_override_path() {
+        let _lock = ENV_MUTEX.lock().expect("env lock should be acquired");
+        let temp = TempDir::new().expect("temp dir should be created");
+        let root = fs::canonicalize(temp.path()).expect("temp dir should canonicalize");
+        let home = root.join("home");
+        fs::create_dir_all(home.join(".codex")).expect("home .codex should be created");
+        fs::create_dir_all(home.join(".claude")).expect("home .claude should be created");
+        let _home_guard = EnvGuard::set("HOME", &home);
+        let override_path = root.join("custom.toml");
+
+        fs::write(
+            &override_path,
+            r#"
+[box]
+cpu_count = 6
+ram_mb = 4096
+disk_gb = 9
+mounts = ["~/.codex:~/.codex:read-write", "~/.claude:~/.claude:read-write"]
+
+[supervisor]
+auto_shutdown_ms = 12345
+"#,
+        )
+        .expect("override config should be written");
+
+        let cfg = load_config_with_path(&root, Some(Path::new("custom.toml")))
+            .expect("load_config_with_path should succeed");
+
+        assert_eq!(cfg.box_cfg.cpu_count, 6);
+        assert_eq!(cfg.box_cfg.ram_size.as_u64(), ByteSize::mib(4096).as_u64());
+        assert_eq!(cfg.box_cfg.disk_size.as_u64(), ByteSize::gib(9).as_u64());
+        assert_eq!(cfg.supervisor.auto_shutdown_ms, 12345);
+        assert!(!root.join("vibebox.toml").exists());
+    }
+
+    #[test]
+    fn resolve_config_path_uses_env_override_when_cli_override_missing() {
+        let temp = TempDir::new().expect("temp dir should be created");
+        let root = fs::canonicalize(temp.path()).expect("temp dir should canonicalize");
+
+        let resolved = resolve_config_path_inner(&root, None, Some(PathBuf::from("custom.toml")))
+            .expect("env override path should resolve");
+
+        assert_eq!(resolved, root.join("custom.toml"));
+    }
+
+    #[test]
+    fn resolve_config_path_rejects_env_override_outside_project() {
+        let temp = TempDir::new().expect("temp dir should be created");
+        let root = fs::canonicalize(temp.path()).expect("temp dir should canonicalize");
+
+        let err = resolve_config_path_inner(&root, None, Some(PathBuf::from("../escape.toml")))
+            .expect_err("outside-project path should be rejected");
+
+        assert!(
+            err.to_string().contains("config path must be within"),
+            "expected bounds-check error, got: {err}"
+        );
+    }
+
+    #[test]
+    fn validate_config_rejects_invalid_values() {
+        let cfg = Config {
+            box_cfg: BoxConfig {
+                cpu_count: 0,
+                ram_size: ByteSize::mib(2048),
+                disk_size: ByteSize::gib(5),
+                mounts: vec![],
+            },
+            supervisor: SupervisorConfig::default(),
+        };
+        let err = validate_config(&cfg).expect_err("cpu_count=0 should fail");
+        assert_eq!(err, "box.cpu_count must be >= 1");
+
+        let cfg = Config {
+            box_cfg: BoxConfig {
+                cpu_count: 2,
+                ram_size: ByteSize::mib(2048),
+                disk_size: ByteSize::gib(5),
+                mounts: vec!["/definitely/missing:/tmp/missing:read-write".to_string()],
+            },
+            supervisor: SupervisorConfig::default(),
+        };
+        let err = validate_config(&cfg).expect_err("invalid mount should fail");
+        assert!(err.starts_with("invalid mount spec"));
+    }
+
+    #[test]
+    fn resolve_config_path_accepts_symlinked_project_root() {
+        let temp = TempDir::new().expect("temp dir should be created");
+        let actual_root = temp.path().join("actual");
+        let link_root = temp.path().join("linked");
+        fs::create_dir_all(&actual_root).expect("actual root should exist");
+        std::os::unix::fs::symlink(&actual_root, &link_root).expect("symlink should be created");
+
+        let resolved = resolve_config_path_inner(&link_root, Some(Path::new("vibebox.toml")), None)
+            .expect("symlinked project root should resolve");
+
+        assert_eq!(resolved, link_root.join("vibebox.toml"));
+    }
+
+    #[test]
+    fn resolve_config_path_rejects_symlink_escape() {
+        let temp = TempDir::new().expect("temp dir should be created");
+        let project_root = temp.path().join("project");
+        let outside_root = temp.path().join("outside");
+        fs::create_dir_all(&project_root).expect("project root should exist");
+        fs::create_dir_all(&outside_root).expect("outside root should exist");
+        std::os::unix::fs::symlink(&outside_root, project_root.join("link"))
+            .expect("escape symlink should be created");
+
+        let err = resolve_config_path_inner(&project_root, Some(Path::new("link/cfg.toml")), None)
+            .expect_err("symlink escape should be rejected");
+        assert!(
+            err.to_string().contains("config path must be within"),
+            "expected bounds-check error, got: {err}"
+        );
+    }
 }

src/error_report.sh

@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+set -Eeuo pipefail
+__vibebox_err_reported=0
+__vibebox_report_error() {
+  local rc="$1"
+  local line="$2"
+  local msg="${3:-}"
+  if [ "$__vibebox_err_reported" -eq 0 ]; then
+    msg="${msg//$'\n'/ }"
+    msg="${msg//$'\r'/ }"
+    if [ -n "$msg" ]; then
+      echo "VIBEBOX_SCRIPT_ERROR:__LABEL__:${line}:${rc} ${msg}"
+    else
+      echo "VIBEBOX_SCRIPT_ERROR:__LABEL__:${line}:${rc}"
+    fi
+    __vibebox_err_reported=1
+  fi
+}
+vibebox_fail() {
+  local msg="${1:-script failed}"
+  local rc="${2:-1}"
+  __vibebox_report_error "$rc" "${LINENO}" "$msg"
+  exit "$rc"
+}
+trap 'rc="$?"; __vibebox_report_error "$rc" "${LINENO}" "command failed: ${BASH_COMMAND:-unknown}"' ERR
+trap 'rc="$?"; if [ "$rc" -ne 0 ]; then __vibebox_report_error "$rc" "${LINENO}" "script exited with code ${rc}"; fi' EXIT
+
+__SCRIPT_BODY__

src/explain.rs

@@ -1,32 +1,32 @@
+use crate::instance::InstanceConfig;
+use crate::utils::relative_to_home;
+use crate::{config, instance, session_manager, tui};
+use anyhow::{Context, Result, bail};
 use std::{
     env,
-    error::Error,
     path::{Path, PathBuf},
 };
 
-use crate::{config, instance, session_manager, tui};
+pub fn build_mount_rows(project: &Path, config: &config::Config) -> Result<Vec<tui::MountListRow>> {
-
-pub fn build_mount_rows(
-    cwd: &Path,
-    config: &config::Config,
-) -> Result<Vec<tui::MountListRow>, Box<dyn Error + Send + Sync>> {
     let mut rows = Vec::new();
-    rows.extend(default_mounts(cwd)?);
+    rows.extend(default_mounts(project)?);
-    let guest_home = resolve_guest_home(cwd)?;
+    let guest_home = resolve_guest_home(project);
     for spec in &config.box_cfg.mounts {
-        rows.push(parse_mount_spec(cwd, spec, false, &guest_home)?);
+        rows.push(parse_mount_spec(project, spec, false, &guest_home)?);
     }
     Ok(rows)
 }
 
-pub fn build_network_rows(
+pub fn build_network_rows(project_dir: &Path) -> Result<Vec<tui::NetworkListRow>> {
-    cwd: &Path,
-) -> Result<Vec<tui::NetworkListRow>, Box<dyn Error + Send + Sync>> {
-    let instance_dir = cwd.join(session_manager::INSTANCE_DIR_NAME);
     let mut vm_ip = "-".to_string();
-    if let Ok(Some(ip)) = instance::read_instance_vm_ip(&instance_dir) {
+    if let Ok(config) = instance::read_instance_config(project_dir) {
+        match config.vm_ipv4 {
+            None => {}
+            Some(ip) => {
                 vm_ip = ip;
             }
+        }
+    }
     let host_to_vm = if vm_ip == "-" {
         "ssh: <pending>:22".to_string()
     } else {
@@ -41,13 +41,13 @@ pub fn build_network_rows(
     Ok(vec![row])
 }
 
-fn default_mounts(cwd: &Path) -> Result<Vec<tui::MountListRow>, Box<dyn Error + Send + Sync>> {
+fn default_mounts(cwd: &Path) -> Result<Vec<tui::MountListRow>> {
     let project_name = cwd
         .file_name()
         .and_then(|name| name.to_str())
-        .unwrap_or("project");
+        .with_context(|| "failed to get project name")?;
     let project_guest = format!("~/{project_name}");
-    let project_host = display_path(cwd);
+    let project_host = relative_to_home(cwd);
     let mut rows = vec![tui::MountListRow {
         host: project_host,
         guest: project_guest,
@@ -57,14 +57,14 @@ fn default_mounts(cwd: &Path) -> Result<Vec<tui::MountListRow>, Box<dyn Error +
 
     let home = env::var("HOME")
         .map(PathBuf::from)
-        .unwrap_or_else(|_| PathBuf::from("/"));
+        .with_context(|| "failed to get home directory")?;
     let cache_home = env::var("XDG_CACHE_HOME")
         .map(PathBuf::from)
         .unwrap_or_else(|_| home.join(".cache"));
     let cache_dir = cache_home.join(session_manager::GLOBAL_CACHE_DIR_NAME);
     let guest_mise_cache = cache_dir.join(".guest-mise-cache");
     rows.push(tui::MountListRow {
-        host: display_path(&guest_mise_cache),
+        host: relative_to_home(&guest_mise_cache),
         guest: "/root/.local/share/mise".to_string(),
         mode: "read-write".to_string(),
         default_mount: "yes".to_string(),
@@ -77,10 +77,10 @@ fn parse_mount_spec(
     spec: &str,
     default_mount: bool,
     guest_home: &str,
-) -> Result<tui::MountListRow, Box<dyn Error + Send + Sync>> {
+) -> Result<tui::MountListRow> {
     let parts: Vec<&str> = spec.split(':').collect();
     if parts.len() < 2 || parts.len() > 3 {
-        return Err(format!("invalid mount spec: {spec}").into());
+        bail!["invalid mount spec: {spec}"];
     }
     let host_part = parts[0];
     let guest_part = parts[1];
@@ -89,11 +89,10 @@ fn parse_mount_spec(
             "read-only" => "read-only",
             "read-write" => "read-write",
             other => {
-                return Err(format!(
+                bail![format!(
                     "invalid mount mode '{}'; expected read-only or read-write",
                     other
-                )
+                )];
-                .into());
             }
         }
     } else {
@@ -116,22 +115,22 @@ fn display_host_spec(cwd: &Path, host: &str) -> String {
     }
     let host_path = PathBuf::from(host);
     if host_path.is_absolute() {
-        return display_path(&host_path);
+        return relative_to_home(&host_path);
     }
     let candidate = cwd.join(&host_path);
     if candidate.is_absolute() {
-        display_path(&candidate)
+        relative_to_home(&candidate)
     } else {
         host.to_string()
     }
 }
 
-fn resolve_guest_home(cwd: &Path) -> Result<String, Box<dyn Error + Send + Sync>> {
+fn resolve_guest_home(project_dir: &Path) -> String {
-    let instance_dir = cwd.join(session_manager::INSTANCE_DIR_NAME);
+    let config = instance::read_instance_config(project_dir);
-    if let Ok(Some(user)) = instance::read_instance_ssh_user(&instance_dir) {
+    match config {
-        return Ok(format!("/home/{user}"));
+        Ok(config) => format!("/home/{}", config.ssh_user),
+        Err(_) => format!("/home/{}", InstanceConfig::default().ssh_user),
     }
-    Ok(format!("/home/{}", instance::DEFAULT_SSH_USER))
 }
 
 fn resolve_guest_display(guest: &str, guest_home: &str) -> String {
@@ -155,17 +154,3 @@ fn resolve_guest_display(guest: &str, guest_home: &str) -> String {
         format!("/root/{guest}")
     }
 }
-
-fn display_path(path: &Path) -> String {
-    let Ok(home) = env::var("HOME") else {
-        return path.display().to_string();
-    };
-    let home_path = PathBuf::from(home);
-    if let Ok(stripped) = path.strip_prefix(&home_path) {
-        if stripped.components().next().is_none() {
-            return "~".to_string();
-        }
-        return format!("~/{}", stripped.display());
-    }
-    path.display().to_string()
-}

src/instance.rs

@@ -1,82 +1,140 @@
+use anyhow::{Context, Result, bail};
+use serde::{Deserialize, Serialize};
+use std::os::unix::fs::FileTypeExt;
 use std::{
     env, fs,
-    io::{self},
+    io::{self, IsTerminal, Read},
     net::{SocketAddr, TcpStream},
     os::unix::{fs::PermissionsExt, net::UnixStream},
     path::{Path, PathBuf},
-    process::{Command, Stdio},
+    process::{Child, Command, Stdio},
-    sync::{Arc, Mutex},
+    sync::{
+        Arc, Mutex,
+        atomic::{AtomicBool, Ordering},
+        mpsc,
+    },
     thread,
     time::{Duration, Instant},
 };
-
-use serde::{Deserialize, Serialize};
 use time::{OffsetDateTime, format_description::well_known::Rfc3339};
 use uuid::Uuid;
 
 use crate::{
     commands,
-    session_manager::{INSTANCE_DIR_NAME, INSTANCE_FILENAME},
+    session_manager::INSTANCE_DIR_NAME,
+    session_manager::{VM_MANAGER_PID_NAME, VM_MANAGER_SOCKET_NAME},
     vm::{self, LoginAction},
 };
 
+pub const STATUS_VM_ERROR_PREFIX: &str = "error:";
+
 const SSH_KEY_NAME: &str = "ssh_key";
-#[cfg_attr(feature = "mock-vm", allow(dead_code))]
+const INSTANCE_FILENAME: &str = "instance.toml";
-pub(crate) const VM_ROOT_LOG_NAME: &str = "vm_root.log";
+const DEFAULT_SSH_USER: &str = "vibecoder";
-pub(crate) const STATUS_FILE_NAME: &str = "status.txt";
+const SSH_CONNECT_RETRIES: usize = 10;
-pub(crate) const DEFAULT_SSH_USER: &str = "vibecoder";
-const SSH_CONNECT_RETRIES: usize = 30;
 const SSH_CONNECT_DELAY_MS: u64 = 500;
 const SSH_SETUP_SCRIPT: &str = include_str!("ssh.sh");
+const STATUS_PREFIX: &str = "status:";
 
-#[derive(Debug, Clone, Serialize, Deserialize)]
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
-pub(crate) struct InstanceConfig {
+pub enum VmLiveness {
-    #[serde(default)]
+    RunningWithSocket { pid: u32 },
-    id: String,
+    RunningWithoutSocket { pid: u32 },
-    #[serde(default = "default_ssh_user")]
+    NotRunningOrMissing,
-    ssh_user: String,
-    #[serde(default)]
-    sudo_password: String,
-    #[serde(default)]
-    last_active: Option<String>,
-    #[serde(default)]
-    pub(crate) vm_ipv4: Option<String>,
 }
 
-impl InstanceConfig {
+pub fn vm_liveness(project_root: &Path) -> Result<VmLiveness> {
-    pub(crate) fn ssh_user_display(&self) -> String {
+    let instance_dir = ensure_instance_dir(project_root)?;
-        if self.ssh_user.trim().is_empty() {
+    let pid_path = instance_dir.join(VM_MANAGER_PID_NAME);
-            DEFAULT_SSH_USER.to_string()
+    let socket_path = instance_dir.join(VM_MANAGER_SOCKET_NAME);
+    fn pid_is_alive(pid: u32) -> bool {
+        let pid = pid as libc::pid_t;
+        let result = unsafe { libc::kill(pid, 0) };
+        if result == 0 {
+            return true;
+        }
+        match io::Error::last_os_error().raw_os_error() {
+            Some(code) if code == libc::EPERM => true,
+            Some(code) if code == libc::ESRCH => false,
+            _ => false,
+        }
+    }
+    let Ok(content) = fs::read_to_string(pid_path) else {
+        return Ok(VmLiveness::NotRunningOrMissing);
+    };
+    let Ok(pid) = content.trim().parse::<u32>() else {
+        return Ok(VmLiveness::NotRunningOrMissing);
+    };
+    if !pid_is_alive(pid) {
+        return Ok(VmLiveness::NotRunningOrMissing);
+    }
+    let has_socket = fs::metadata(socket_path)
+        .map(|meta| meta.file_type().is_socket())
+        .unwrap_or(false);
+    if has_socket {
+        Ok(VmLiveness::RunningWithSocket { pid })
     } else {
-            self.ssh_user.clone()
+        Ok(VmLiveness::RunningWithoutSocket { pid })
-        }
     }
 }
 
+#[derive(Debug, thiserror::Error)]
+pub enum InstanceError {
+    #[error("unexpected disconnection from vm manager")]
+    UnexpectedDisconnection,
+    #[error("{0}")]
+    VMError(String),
+}
+
 fn default_ssh_user() -> String {
     DEFAULT_SSH_USER.to_string()
 }
 
-pub fn run_with_ssh(manager_conn: UnixStream) -> Result<(), Box<dyn std::error::Error>> {
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct InstanceConfig {
+    #[serde(default)]
+    pub id: String,
+    #[serde(default = "default_ssh_user")]
+    pub ssh_user: String,
+    #[serde(default)]
+    sudo_password: String,
+    #[serde(default)]
+    pub last_active: Option<String>,
+    #[serde(default)]
+    pub vm_ipv4: Option<String>,
+}
+
+impl Default for InstanceConfig {
+    fn default() -> Self {
+        Self {
+            id: Uuid::now_v7().to_string(),
+            ssh_user: DEFAULT_SSH_USER.to_string(),
+            sudo_password: Uuid::now_v7().simple().to_string(),
+            last_active: None,
+            vm_ipv4: None,
+        }
+    }
+}
+
+pub fn run_with_ssh(manager_conn: UnixStream) -> Result<()> {
     let project_root = env::current_dir()?;
     tracing::info!(root = %project_root.display(), "starting ssh session");
     let instance_dir = ensure_instance_dir(&project_root)?;
     tracing::debug!(instance_dir = %instance_dir.display(), "instance dir ready");
     let (ssh_key, _ssh_pub) = ensure_ssh_keypair(&instance_dir)?;
 
-    let config = load_or_create_instance_config(&instance_dir)?;
+    let config = load_or_create_instance_config(&project_root)?;
     let ssh_user = config.ssh_user.clone();
     tracing::debug!(ssh_user = %ssh_user, "loaded instance config");
 
-    let _manager_conn = manager_conn;
+    wait_for_vm_ipv4(&project_root, Duration::from_secs(480), &manager_conn)?;
-    wait_for_vm_ipv4(&instance_dir, Duration::from_secs(480))?;
 
-    let ip = load_or_create_instance_config(&instance_dir)?
+    let ip = load_or_create_instance_config(&project_root)?
         .vm_ipv4
-        .ok_or("VM IPv4 not available")?;
+        .with_context(|| "failed to load instance IP address")?;
     tracing::info!(ip = %ip, "vm ipv4 ready");
 
-    run_ssh_session(ssh_key, ssh_user, ip)
+    run_ssh_session(ssh_key, ssh_user, ip, manager_conn, project_root)
 }
 
 pub fn ensure_instance_dir(project_root: &Path) -> Result<PathBuf, io::Error> {
@@ -85,9 +143,7 @@ pub fn ensure_instance_dir(project_root: &Path) -> Result<PathBuf, io::Error> {
     Ok(instance_dir)
 }
 
-pub(crate) fn ensure_ssh_keypair(
+pub fn ensure_ssh_keypair(instance_dir: &Path) -> Result<(PathBuf, PathBuf)> {
-    instance_dir: &Path,
-) -> Result<(PathBuf, PathBuf), Box<dyn std::error::Error>> {
     let private_key = instance_dir.join(SSH_KEY_NAME);
     let public_key = instance_dir.join(format!("{SSH_KEY_NAME}.pub"));
 
@@ -109,7 +165,9 @@ pub(crate) fn ensure_ssh_keypair(
             "-N",
             "",
             "-f",
-            private_key.to_str().ok_or("ssh key path not utf-8")?,
+            private_key
+                .to_str()
+                .with_context(|| "ssh key path not utf-8")?,
             "-C",
             "vibebox",
         ])
@@ -119,7 +177,7 @@ pub(crate) fn ensure_ssh_keypair(
         .status()?;
 
     if !status.success() {
-        return Err("ssh-keygen failed".into());
+        bail!("ssh-keygen failed");
     }
 
     fs::set_permissions(&private_key, fs::Permissions::from_mode(0o600))?;
@@ -128,98 +186,66 @@ pub(crate) fn ensure_ssh_keypair(
     Ok((private_key, public_key))
 }
 
-pub(crate) fn load_or_create_instance_config(
+pub fn load_or_create_instance_config(project_dir: &Path) -> Result<InstanceConfig> {
-    instance_dir: &Path,
+    let mut exist = true;
-) -> Result<InstanceConfig, Box<dyn std::error::Error>> {
+    let mut config = read_instance_config(project_dir).unwrap_or_else(|_| {
-    let config_path = instance_dir.join(INSTANCE_FILENAME);
+        exist = false;
-    let mut config = if config_path.exists() {
+        InstanceConfig::default()
-        let raw = fs::read_to_string(&config_path)?;
+    });
-        toml::from_str::<InstanceConfig>(&raw)?
-    } else {
-        InstanceConfig {
-            id: String::new(),
-            ssh_user: default_ssh_user(),
-            sudo_password: String::new(),
-            last_active: None,
-            vm_ipv4: None,
-        }
-    };
 
     let mut changed = false;
     if config.ssh_user.trim().is_empty() {
-        config.ssh_user = default_ssh_user();
+        config.ssh_user = InstanceConfig::default().ssh_user;
         changed = true;
     }
 
     if config.id.trim().is_empty() {
-        config.id = Uuid::now_v7().to_string();
+        config.id = InstanceConfig::default().id;
         changed = true;
     }
 
     if config.sudo_password.trim().is_empty() {
-        config.sudo_password = generate_password();
+        config.sudo_password = InstanceConfig::default().sudo_password;
         changed = true;
     }
 
-    if !config_path.exists() || changed {
+    if !exist || changed {
-        write_instance_config(&config_path, &config)?;
+        write_instance_config(project_dir, &config)?;
     }
 
     Ok(config)
 }
 
-fn read_instance_config(
+pub fn read_instance_config(project_dir: &Path) -> Result<InstanceConfig> {
-    instance_dir: &Path,
+    // todo maybe verify schema?
-) -> Result<Option<InstanceConfig>, Box<dyn std::error::Error>> {
+    let config_path = project_dir.join(INSTANCE_DIR_NAME).join(INSTANCE_FILENAME);
-    let config_path = instance_dir.join(INSTANCE_FILENAME);
     if !config_path.exists() {
-        return Ok(None);
+        bail!("instance config file does not exist");
     }
     let raw = fs::read_to_string(&config_path)?;
     let config = toml::from_str::<InstanceConfig>(&raw)?;
-    Ok(Some(config))
+    Ok(config)
 }
 
-pub fn read_instance_vm_ip(
+pub fn touch_last_active(project_dir: &Path) -> Result<()> {
-    instance_dir: &Path,
+    let mut config = load_or_create_instance_config(project_dir)?;
-) -> Result<Option<String>, Box<dyn std::error::Error>> {
-    let config = read_instance_config(instance_dir)?;
-    Ok(config.and_then(|cfg| cfg.vm_ipv4))
-}
-
-pub fn read_instance_ssh_user(
-    instance_dir: &Path,
-) -> Result<Option<String>, Box<dyn std::error::Error>> {
-    let config = read_instance_config(instance_dir)?;
-    Ok(config
-        .map(|cfg| cfg.ssh_user)
-        .filter(|user| !user.trim().is_empty()))
-}
-
-pub fn touch_last_active(instance_dir: &Path) -> Result<(), Box<dyn std::error::Error>> {
-    let mut config = load_or_create_instance_config(instance_dir)?;
     let now = OffsetDateTime::now_utc().format(&Rfc3339)?;
     config.last_active = Some(now);
-    write_instance_config(&instance_dir.join(INSTANCE_FILENAME), &config)?;
+    write_instance_config(project_dir, &config)?;
     Ok(())
 }
 
-pub(crate) fn write_instance_config(
+pub fn write_instance_config(project_dir: &Path, config: &InstanceConfig) -> Result<()> {
-    path: &Path,
+    let path = project_dir.join(INSTANCE_DIR_NAME).join(INSTANCE_FILENAME);
-    config: &InstanceConfig,
-) -> Result<(), Box<dyn std::error::Error>> {
     let data = toml::to_string_pretty(config)?;
-    fs::write(path, data)?;
+    fs::create_dir_all(project_dir.join(INSTANCE_DIR_NAME))?;
-    fs::set_permissions(path, fs::Permissions::from_mode(0o600))?;
+    fs::write(&path, data)?;
+    fs::set_permissions(&path, fs::Permissions::from_mode(0o600))?;
     Ok(())
 }
 
-fn generate_password() -> String {
-    Uuid::now_v7().simple().to_string()
-}
-
 #[cfg_attr(feature = "mock-vm", allow(dead_code))]
-pub(crate) fn extract_ipv4(line: &str) -> Option<String> {
+pub fn extract_ipv4(line: &str) -> Option<String> {
     let mut current = String::new();
     let mut best: Option<String> = None;
 
@@ -238,46 +264,68 @@ pub(crate) fn extract_ipv4(line: &str) -> Option<String> {
     best
 }
 
+fn handle_manager_line(line: &str, last_status: &mut Option<String>) -> Result<()> {
+    if let Some(status) = line.strip_prefix(STATUS_PREFIX) {
+        let status = status.trim();
+        if let Some(message) = status.strip_prefix(STATUS_VM_ERROR_PREFIX) {
+            let message = message.trim();
+            if message.is_empty() {
+                bail!("vm manager reported startup failure");
+            }
+            return Err(InstanceError::VMError(message.to_string()).into());
+        }
+        if !status.is_empty() && last_status.as_deref() != Some(status) {
+            tracing::info!("[background]: {}", status);
+            *last_status = Some(status.to_string());
+        }
+    }
+    Ok(())
+}
+
 fn wait_for_vm_ipv4(
-    instance_dir: &Path,
+    project_dir: &Path,
     timeout: Duration,
-) -> Result<(), Box<dyn std::error::Error>> {
+    manager_conn: &UnixStream,
+) -> Result<()> {
     let start = Instant::now();
     let mut next_log_at = start + Duration::from_secs(10);
-    let mut next_status_check = start;
+    let mut stream = manager_conn.try_clone()?;
+    let _ = stream.set_read_timeout(Some(Duration::from_millis(250)));
+    let mut read_buf = [0u8; 1024];
+    let mut pending = String::new();
     tracing::info!("waiting for vm ipv4");
-    let status_path = instance_dir.join(STATUS_FILE_NAME);
     let mut last_status: Option<String> = None;
-    let mut status_missing = true;
     let mut once_hint = false;
     loop {
-        let config = load_or_create_instance_config(instance_dir)?;
+        match stream.read(&mut read_buf) {
+            Ok(0) => {
+                bail!("vm manager disconnected before VM became ready");
+            }
+            Ok(n) => {
+                pending.push_str(&String::from_utf8_lossy(&read_buf[..n]));
+                while let Some(pos) = pending.find('\n') {
+                    let line = pending[..pos].trim().to_string();
+                    pending.drain(..=pos);
+                    handle_manager_line(&line, &mut last_status)?;
+                }
+            }
+            Err(err) if err.kind() == io::ErrorKind::WouldBlock => {}
+            Err(err) if err.kind() == io::ErrorKind::TimedOut => {}
+            Err(err) if err.kind() == io::ErrorKind::Interrupted => {}
+            Err(err) => {
+                tracing::warn!(error = %err, "failed to read vm manager status stream");
+            }
+        }
+
+        let config = load_or_create_instance_config(project_dir)?;
         if config.vm_ipv4.is_some() {
-            let _ = fs::remove_file(&status_path);
             return Ok(());
         }
         if start.elapsed() > timeout {
-            let _ = fs::remove_file(&status_path);
+            bail!("timed out waiting for VM IPv4");
-            return Err("Timed out waiting for VM IPv4".into());
         }
+
         let now = Instant::now();
-        if now >= next_status_check {
-            match fs::read_to_string(&status_path) {
-                Ok(status) => {
-                    status_missing = false;
-                    let status = status.trim().to_string();
-                    if !status.is_empty() && last_status.as_deref() != Some(status.as_str()) {
-                        tracing::info!("[background]: {}", status);
-                        last_status = Some(status);
-                        next_log_at = now + Duration::from_secs(20);
-                    }
-                }
-                Err(_) => {
-                    status_missing = true;
-                }
-            }
-            next_status_check = now + Duration::from_millis(500);
-        }
         if now >= next_log_at {
             let waited = start.elapsed();
             if waited.as_secs() > 15 && !once_hint {
@@ -286,12 +334,9 @@ fn wait_for_vm_ipv4(
                 );
                 once_hint = true;
             }
-            if status_missing {
             tracing::info!("still waiting for vm ipv4, {}s elapsed", waited.as_secs(),);
-            }
             next_log_at += Duration::from_secs(20);
         }
-        thread::sleep(Duration::from_millis(200));
     }
 }
 
@@ -299,9 +344,14 @@ fn run_ssh_session(
     ssh_key: PathBuf,
     ssh_user: String,
     ip: String,
-) -> Result<(), Box<dyn std::error::Error>> {
+    manager_conn: UnixStream,
+    project_root: PathBuf,
+) -> Result<()> {
     let mut attempts = 0usize;
     loop {
+        if matches!(vm_liveness(&project_root)?, VmLiveness::NotRunningOrMissing) {
+            return Err(InstanceError::UnexpectedDisconnection.into());
+        }
         attempts += 1;
         if !ssh_port_open(&ip) {
             tracing::debug!(attempts, "ssh port doesn't open yet");
@@ -313,9 +363,7 @@ fn run_ssh_session(
                 SSH_CONNECT_RETRIES
             );
             if attempts >= SSH_CONNECT_RETRIES {
-                return Err(
+                bail!("ssh port not ready after {SSH_CONNECT_RETRIES} attempts");
-                    format!("ssh port not ready after {SSH_CONNECT_RETRIES} attempts").into(),
-                );
             }
             thread::sleep(Duration::from_millis(SSH_CONNECT_DELAY_MS));
             continue;
@@ -329,10 +377,10 @@ fn run_ssh_session(
             attempts,
             SSH_CONNECT_RETRIES
         );
-        let status = Command::new("ssh")
+        let child = Command::new("ssh")
             .args([
                 "-i",
-                ssh_key.to_str().unwrap_or(".vibebox/ssh_key"),
+                ssh_key.to_str().with_context(|| "invalid path")?,
                 "-o",
                 "IdentitiesOnly=yes",
                 "-o",
@@ -357,27 +405,67 @@ fn run_ssh_session(
             .stdin(Stdio::inherit())
             .stdout(Stdio::inherit())
             .stderr(Stdio::inherit())
-            .status();
+            .spawn();
 
-        match status {
+        match child {
-            Ok(status) if status.success() => {
+            Ok(mut child) => {
+                let done = Arc::new(AtomicBool::new(false));
+                let done_for_monitor = done.clone();
+                let (disconnect_tx, disconnect_rx) = mpsc::channel::<()>();
+                let mut manager_stream = manager_conn.try_clone()?;
+                let _ = manager_stream.set_read_timeout(Some(Duration::from_millis(250)));
+                thread::spawn(move || {
+                    let mut buf = [0u8; 1];
+                    while !done_for_monitor.load(Ordering::Relaxed) {
+                        match manager_stream.read(&mut buf) {
+                            Ok(0) => {
+                                let _ = disconnect_tx.send(());
+                                return;
+                            }
+                            Ok(_) => {}
+                            Err(err) if err.kind() == io::ErrorKind::WouldBlock => {}
+                            Err(err) if err.kind() == io::ErrorKind::TimedOut => {}
+                            Err(err) if err.kind() == io::ErrorKind::Interrupted => {}
+                            Err(_) => {
+                                let _ = disconnect_tx.send(());
+                                return;
+                            }
+                        }
+                    }
+                });
+
+                let status = loop {
+                    if disconnect_rx.try_recv().is_ok() {
+                        done.store(true, Ordering::Relaxed);
+                        terminate_ssh_child(&mut child);
+                        restore_terminal_after_disconnect();
+                        return Err(InstanceError::UnexpectedDisconnection.into());
+                    }
+                    if let Some(status) = child.try_wait()? {
+                        done.store(true, Ordering::Relaxed);
+                        break status;
+                    }
+                    thread::sleep(Duration::from_millis(100));
+                };
+
+                if status.success() {
                     tracing::info!(status = %status, "ssh exited");
                     break;
                 }
-            Ok(status) if status.code() == Some(255) => {
+                if status.code() == Some(255) {
                     tracing::warn!(status = %status, "ssh connection failed");
                     if attempts >= SSH_CONNECT_RETRIES {
-                    return Err(format!("ssh failed after {SSH_CONNECT_RETRIES} attempts").into());
+                        bail!("ssh failed after {SSH_CONNECT_RETRIES} attempts");
                     }
                     thread::sleep(Duration::from_millis(500));
+                    continue;
                 }
-            Ok(status) => {
                 tracing::info!(status = %status, "ssh exited");
                 break;
             }
             Err(err) => {
                 tracing::error!(error = %err, "failed to start ssh");
-                return Err(format!("failed to start ssh: {err}").into());
+                bail!("failed to start ssh: {err}");
             }
         }
     }
@@ -385,6 +473,30 @@ fn run_ssh_session(
     Ok(())
 }
 
+fn terminate_ssh_child(child: &mut Child) {
+    let pid = child.id() as i32;
+    unsafe {
+        libc::kill(pid, libc::SIGTERM);
+    }
+    let deadline = Instant::now() + Duration::from_millis(700);
+    while Instant::now() < deadline {
+        match child.try_wait() {
+            Ok(Some(_)) => return,
+            Ok(None) => thread::sleep(Duration::from_millis(50)),
+            Err(_) => break,
+        }
+    }
+    let _ = child.kill();
+    let _ = child.wait();
+}
+
+fn restore_terminal_after_disconnect() {
+    if io::stdin().is_terminal() {
+        let _ = Command::new("stty").arg("sane").status();
+    }
+    eprintln!();
+}
+
 #[cfg_attr(feature = "mock-vm", allow(dead_code))]
 fn is_ipv4_candidate(candidate: &str) -> bool {
     let parts: Vec<&str> = candidate.split('.').collect();
@@ -407,10 +519,10 @@ fn ssh_port_open(ip: &str) -> bool {
         Ok(addr) => addr,
         Err(_) => return false,
     };
-    TcpStream::connect_timeout(&addr, std::time::Duration::from_millis(500)).is_ok()
+    TcpStream::connect_timeout(&addr, Duration::from_millis(500)).is_ok()
 }
 
-pub(crate) fn build_ssh_login_actions(
+pub fn build_ssh_login_actions(
     config: &Arc<Mutex<InstanceConfig>>,
     project_name: &str,
     project_guest_dir: &str,
@@ -425,7 +537,7 @@ pub(crate) fn build_ssh_login_actions(
 
     let key_path = format!("{guest_dir}/{key_name}.pub");
 
-    let setup_script = SSH_SETUP_SCRIPT
+    let ssh_script = SSH_SETUP_SCRIPT
         .replace("__SSH_USER__", &ssh_user)
         .replace("__SUDO_PASSWORD__", &sudo_password)
         .replace("__PROJECT_NAME__", project_name)
@@ -433,8 +545,36 @@ pub(crate) fn build_ssh_login_actions(
         .replace("__KEY_PATH__", &key_path)
         .replace("__VIBEBOX_SHELL_SCRIPT__", &commands::render_shell_script())
         .replace("__VIBEBOX_HOME_LINKS__", home_links_script);
-    let setup = vm::script_command_from_content("ssh_setup", &setup_script)
+    let setup = vm::script_command_from_content("ssh.sh", &ssh_script)
         .expect("ssh setup script contained invalid marker");
 
     vec![LoginAction::Send(setup)]
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn handle_manager_line_updates_status() {
+        let mut last_status = None;
+        handle_manager_line("status: preparing VM image...", &mut last_status)
+            .expect("status should be accepted");
+        assert_eq!(last_status.as_deref(), Some("preparing VM image..."));
+    }
+
+    #[test]
+    fn handle_manager_line_ignores_non_status_lines() {
+        let mut last_status = None;
+        handle_manager_line("pid=123", &mut last_status).expect("non-status lines are ignored");
+        assert!(last_status.is_none());
+    }
+
+    #[test]
+    fn handle_manager_line_surfaces_error_status() {
+        let mut last_status = None;
+        let err = handle_manager_line("status: error: vm failed to boot", &mut last_status)
+            .expect_err("error status should fail");
+        assert_eq!(err.to_string(), "vm failed to boot");
+    }
+}

src/lib.rs

@@ -8,3 +8,4 @@ pub mod vm_manager;
 
 pub use session_manager::{SessionError, SessionManager, SessionRecord};
 pub mod config;
+pub mod utils;

src/provision.sh

@@ -1,12 +1,12 @@
 #!/bin/bash
 set -eEux
 
-trap 'echo "[vibebox][error] provisioning failed"; echo "VIBEBOX_PROVISION_FAILED"; systemctl poweroff || true; exit 1' ERR
+trap 'rc=$?; echo "[vibebox][error] provisioning failed at: ${BASH_COMMAND} (exit ${rc})"; printf "%s%s\n" VIBEBOX_PROVISION_ FAILED; systemctl poweroff || true; exit 1' ERR
 
 # Wait for network + DNS before apt-get to avoid early boot flakiness.
 wait_for_network() {
   echo "[vibebox] waiting for network/DNS readiness"
-  local deadline=$((SECONDS + 60))
+  local deadline=$((SECONDS + 180))
   while [ "$SECONDS" -lt "$deadline" ]; do
     local has_route=0
     if ip -4 route show default >/dev/null 2>&1; then
@@ -21,7 +21,7 @@ wait_for_network() {
     fi
     sleep 1
   done
-  echo "[vibebox][warn] network/DNS still not ready after 60s; continuing" >&2
+  echo "[vibebox][warn] network/DNS still not ready after 180s; continuing" >&2
   echo "[vibebox][warn] /etc/resolv.conf:" >&2
   cat /etc/resolv.conf >&2 || true
   ip -br addr >&2 || true
@@ -44,9 +44,9 @@ apt_update_with_retries() {
 }
 
 # Don't wait too long for slow mirrors.
-echo 'Acquire::http::Timeout "2";' | tee /etc/apt/apt.conf.d/99timeout
+echo 'Acquire::http::Timeout "10";' | tee /etc/apt/apt.conf.d/99timeout
-echo 'Acquire::https::Timeout "2";' | tee -a /etc/apt/apt.conf.d/99timeout
+echo 'Acquire::https::Timeout "10";' | tee -a /etc/apt/apt.conf.d/99timeout
-echo 'Acquire::Retries "2";' | tee -a /etc/apt/apt.conf.d/99timeout
+echo 'Acquire::Retries "5";' | tee -a /etc/apt/apt.conf.d/99timeout
 
 wait_for_network
 apt_update_with_retries
@@ -98,5 +98,5 @@ sleep 100 # sleep here so that we don't see the login screen flash up before the
 EOF
 
 # Done provisioning, power off the VM
-echo "VIBEBOX_PROVISION_OK"
+printf "%s%s\n" VIBEBOX_PROVISION_ OK
 systemctl poweroff

src/session_manager.rs

@@ -1,19 +1,18 @@
+use anyhow::Result;
 use std::{
     env, fs,
     io::{self, Write},
-    os::unix::fs::FileTypeExt,
     path::{Path, PathBuf},
 };
 
+use crate::config::config_path;
+use crate::instance::{VmLiveness, read_instance_config, vm_liveness};
 use serde::{Deserialize, Serialize};
 
-use crate::config::CONFIG_FILENAME;
-
 pub const INSTANCE_DIR_NAME: &str = ".vibebox";
 pub const GLOBAL_CACHE_DIR_NAME: &str = "vibebox";
 pub const GLOBAL_DIR_NAME: &str = ".vibebox";
-pub const INSTANCE_FILENAME: &str = "instance.toml";
+const SESSION_TOML_SUFFIX: &str = ".toml";
-pub const SESSION_TOML_SUFFIX: &str = ".toml";
 pub const VM_MANAGER_SOCKET_NAME: &str = "vm.sock";
 pub const VM_MANAGER_PID_NAME: &str = "vm.pid";
 const SESSIONS_DIR_NAME: &str = "sessions";
@@ -32,14 +31,6 @@ struct SessionEntry {
     pub id: String,
 }
 
-#[derive(Debug, Default, Deserialize)]
-struct InstanceMetadata {
-    #[serde(default)]
-    id: Option<String>,
-    #[serde(default)]
-    last_active: Option<String>,
-}
-
 #[derive(Debug)]
 pub struct SessionManager {
     sessions_dir: PathBuf,
@@ -61,7 +52,7 @@ pub enum SessionError {
     #[error("Session directory does not exist: {0}")]
     MissingDirectory(PathBuf),
     #[error(transparent)]
-    Io(#[from] std::io::Error),
+    Io(#[from] io::Error),
     #[error(transparent)]
     TomlDe(#[from] toml::de::Error),
     #[error(transparent)]
@@ -94,8 +85,8 @@ impl SessionManager {
         let mut added = false;
 
         if has_config {
-            let meta = read_instance_metadata(&directory)?;
+            let id = read_instance_config(&directory).map_or(None, |config| Some(config.id));
-            if let Some(id) = meta.id {
+            if let Some(id) = id {
                 let record = SessionEntry {
                     directory: directory.clone(),
                     id: id.clone(),
@@ -110,7 +101,6 @@ impl SessionManager {
             } else {
                 tracing::warn!(
                     directory = %directory.display(),
-                    file = INSTANCE_FILENAME,
                     "missing session id in instance file"
                 );
             }
@@ -148,12 +138,13 @@ impl SessionManager {
         }
         let mut records = Vec::with_capacity(sessions.len());
         for session in sessions {
-            let meta = read_instance_metadata(&session.directory)?;
+            let last_active =
+                read_instance_config(&session.directory).map_or(None, |option| option.last_active);
             let active = is_session_active(&session.directory);
             records.push(SessionRecord {
                 directory: session.directory,
                 id: session.id,
-                last_active: meta.last_active,
+                last_active,
                 active,
             });
         }
@@ -271,43 +262,22 @@ fn is_vibebox_dir(directory: &Path) -> bool {
     if !directory.is_absolute() {
         return false;
     }
-    directory.join(CONFIG_FILENAME).is_file()
+    config_path(directory).is_file()
 }
 
 fn is_session_active(directory: &Path) -> bool {
     let instance_dir = directory.join(INSTANCE_DIR_NAME);
     let pid_path = instance_dir.join(VM_MANAGER_PID_NAME);
-    let socket_path = instance_dir.join(VM_MANAGER_SOCKET_NAME);
+    match vm_liveness(directory) {
-
+        Ok(liveness) => match liveness {
-    let pid = read_pid(&pid_path);
+            VmLiveness::RunningWithSocket { .. } => true,
-    let is_alive = pid.map(pid_is_alive).unwrap_or(false);
+            VmLiveness::RunningWithoutSocket { .. } => true,
-    if !is_alive {
+            VmLiveness::NotRunningOrMissing => {
                 let _ = fs::remove_file(&pid_path);
-        return false;
+                false
             }
-
+        },
-    if let Ok(metadata) = fs::metadata(&socket_path) {
+        Err(_) => false,
-        return metadata.file_type().is_socket();
-    }
-
-    true
-}
-
-fn read_pid(path: &Path) -> Option<u32> {
-    let content = fs::read_to_string(path).ok()?;
-    content.trim().parse::<u32>().ok()
-}
-
-fn pid_is_alive(pid: u32) -> bool {
-    let pid = pid as libc::pid_t;
-    let result = unsafe { libc::kill(pid, 0) };
-    if result == 0 {
-        return true;
-    }
-    match std::io::Error::last_os_error().raw_os_error() {
-        Some(code) if code == libc::EPERM => true,
-        Some(code) if code == libc::ESRCH => false,
-        _ => false,
     }
 }
 
@@ -315,31 +285,11 @@ fn read_session_file(path: &Path) -> Result<SessionEntry, SessionError> {
     let raw = fs::read_to_string(path)?;
     let record: SessionEntry = toml::from_str(&raw)?;
     if record.id.trim().is_empty() {
-        return Err(std::io::Error::new(io::ErrorKind::InvalidData, "session id missing").into());
+        return Err(io::Error::new(io::ErrorKind::InvalidData, "session id missing").into());
     }
     Ok(record)
 }
 
-fn read_instance_metadata(directory: &Path) -> Result<InstanceMetadata, SessionError> {
-    let instance_path = directory.join(INSTANCE_DIR_NAME).join(INSTANCE_FILENAME);
-    if !instance_path.exists() {
-        return Ok(InstanceMetadata::default());
-    }
-    let raw = fs::read_to_string(&instance_path)?;
-    let mut meta: InstanceMetadata = toml::from_str(&raw)?;
-    if let Some(id) = &meta.id
-        && id.trim().is_empty()
-    {
-        meta.id = None;
-    }
-    if let Some(last_active) = &meta.last_active
-        && last_active.trim().is_empty()
-    {
-        meta.last_active = None;
-    }
-    Ok(meta)
-}
-
 fn atomic_write(path: &Path, content: &[u8]) -> io::Result<()> {
     let Some(parent) = path.parent() else {
         return Err(io::Error::new(
@@ -362,6 +312,7 @@ fn atomic_write(path: &Path, content: &[u8]) -> io::Result<()> {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::instance::{InstanceConfig, write_instance_config};
     use std::fs;
     use tempfile::TempDir;
 
@@ -375,11 +326,14 @@ mod tests {
         dir
     }
 
-    fn write_instance(project_dir: &Path, id: &str, last_active: &str) {
+    fn write_instance(project_dir: &Path, id: &str, last_active: &str) -> Result<()> {
-        let instance_dir = project_dir.join(INSTANCE_DIR_NAME);
+        fs::create_dir_all(project_dir)?;
-        fs::create_dir_all(&instance_dir).unwrap();
+
-        let content = format!("id = \"{id}\"\nlast_active = \"{last_active}\"\n");
+        let mut config = InstanceConfig::default();
-        fs::write(instance_dir.join(INSTANCE_FILENAME), content).unwrap();
+        config.id = id.to_string();
+        config.last_active = Some(last_active.to_string());
+
+        write_instance_config(project_dir, &config)
     }
 
     #[test]
@@ -387,11 +341,14 @@ mod tests {
         let temp = TempDir::new().unwrap();
         let mgr = manager(&temp);
         let project_dir = create_project_dir(&temp);
-        fs::write(project_dir.join(CONFIG_FILENAME), "").unwrap();
+        fs::write(config_path(project_dir.as_path()), "").unwrap();
+        assert!(
             write_instance(
                 &project_dir,
                 "019bf290-cccc-7c23-ba1d-dce7e6d40693",
                 "2026-02-07T05:00:00Z",
+            )
+            .is_ok()
         );
 
         let dirs = mgr.update_global_sessions(&project_dir).unwrap();
@@ -412,15 +369,18 @@ mod tests {
         let temp = TempDir::new().unwrap();
         let mgr = manager(&temp);
         let project_dir = create_project_dir(&temp);
-        fs::write(project_dir.join(CONFIG_FILENAME), "").unwrap();
+        fs::write(config_path(project_dir.as_path()), "").unwrap();
+        assert!(
             write_instance(
                 &project_dir,
                 "019bf290-cccc-7c23-ba1d-dce7e6d40693",
                 "2026-02-07T05:00:00Z",
+            )
+            .is_ok()
         );
         let _ = mgr.update_global_sessions(&project_dir).unwrap();
 
-        fs::remove_file(project_dir.join(CONFIG_FILENAME)).unwrap();
+        fs::remove_file(config_path(project_dir.as_path())).unwrap();
         let sessions = mgr.list_sessions().unwrap();
         assert!(sessions.is_empty());
 
@@ -451,11 +411,14 @@ mod tests {
         let temp = TempDir::new().unwrap();
         let mgr = manager(&temp);
         let project_dir = create_project_dir(&temp);
-        fs::write(project_dir.join(CONFIG_FILENAME), "").unwrap();
+        fs::write(config_path(project_dir.as_path()), "").unwrap();
+        assert!(
             write_instance(
                 &project_dir,
                 "019bf290-cccc-7c23-ba1d-dce7e6d40693",
                 "2026-02-07T05:00:00Z",
+            )
+            .is_ok()
         );
         let _ = mgr.update_global_sessions(&project_dir).unwrap();
 

src/ssh.sh

@@ -2,6 +2,7 @@
 set -eu
 
 SSH_USER="__SSH_USER__"
+SUDO_PASSWORD="__SUDO_PASSWORD__"
 PROJECT_NAME="__PROJECT_NAME__"
 PROJECT_GUEST_DIR="__PROJECT_GUEST_DIR__"
 KEY_PATH="__KEY_PATH__"
@@ -61,6 +62,10 @@ if ! id -u "$SSH_USER" >/dev/null 2>&1; then
   usermod -aG sudo "$SSH_USER" || true
 fi
 
+if [ -n "$SUDO_PASSWORD" ]; then
+  echo "${SSH_USER}:${SUDO_PASSWORD}" | chpasswd
+fi
+
 install -d -m 700 -o "$SSH_USER" -g "$SSH_USER" "/home/${SSH_USER}/.ssh"
 install -m 600 -o "$SSH_USER" -g "$SSH_USER" "$KEY_PATH" "/home/${SSH_USER}/.ssh/authorized_keys"
 
@@ -109,8 +114,8 @@ mise_ok() { command -v mise >/dev/null 2>&1 || [ -x "$MISE_BIN" ]; }
 mise_install() {
   if [ ! -x "$MISE_BIN" ] && ! command -v mise >/dev/null 2>&1; then
     if ! curl https://mise.run | HOME="$USER_HOME" sh; then
-      mise_warn "mise install script failed (continuing)"
+      mise_warn "mise install script failed"
-      return 0
+      return 1
     fi
   fi
   echo 'eval "$(~/.local/bin/mise activate bash)"' >> "${USER_HOME}/.bashrc"
@@ -136,18 +141,21 @@ MISE
   touch "${USER_HOME}/.config/mise/mise.lock"
   if [ -x "$MISE_BIN" ]; then
     if ! HOME="$USER_HOME" "$MISE_BIN" install; then
-      mise_warn "mise install failed (continuing)"
+      mise_warn "mise install failed"
-      return 0
+      return 1
     fi
   else
     if ! HOME="$USER_HOME" mise install; then
-      mise_warn "mise install failed (continuing)"
+      mise_warn "mise install failed"
-      return 0
+      return 1
     fi
   fi
 }
 
-mise_install || true
+if ! mise_install; then
+  diag "mise installation failed"
+  vibebox_fail "mise installation failed" 1
+fi
 
 # 3) start ssh (don't swallow failures)
 # If ssh is already active, don't force start/restart.
@@ -155,7 +163,7 @@ if ! systemctl is-active --quiet ssh; then
   if ! systemctl start ssh; then
     diag "systemctl start ssh failed"
     dump_diag
-    exit 1
+    vibebox_fail "failed to start ssh service" 1
   fi
 fi
 
@@ -193,7 +201,7 @@ done
 if [ -z "$dev" ] || [ -z "$ip" ]; then
   diag "no stable IPv4 on default route interface"
   dump_diag
-  exit 1
+  vibebox_fail "no stable ipv4 route on default interface" 1
 fi
 
 # 5) strong verify: ssh must listen externally (0.0.0.0:22 or $ip:22 or [::]:22)
@@ -212,7 +220,7 @@ done
 if ! listens_ok; then
   diag "sshd not listening on 0.0.0.0:22 / ${ip}:22"
   dump_diag
-  exit 1
+  vibebox_fail "sshd is not listening on the expected address" 1
 fi
 
 ip a

src/tui.rs

@@ -1,10 +1,4 @@
-use std::{
+use bytesize::ByteSize;
-    io::{self, Write},
-    os::unix::io::OwnedFd,
-    path::PathBuf,
-    sync::{Arc, Mutex},
-};
-
 use color_eyre::Result;
 use crossterm::{
     cursor::{MoveTo, Show},
@@ -22,6 +16,12 @@ use ratatui::{
     text::{Line, Span, Text},
     widgets::{Block, Borders, Cell, List, ListItem, Paragraph, Row, Table, Widget},
 };
+use std::{
+    io::{self, Write},
+    os::unix::io::OwnedFd,
+    path::PathBuf,
+    sync::{Arc, Mutex},
+};
 
 use crate::vm;
 
@@ -39,9 +39,9 @@ const INFO_LINE_COUNT: u16 = 5;
 
 #[derive(Debug, Clone)]
 pub struct VmInfo {
-    pub max_memory_mb: u64,
+    pub max_memory: ByteSize,
     pub cpu_cores: usize,
-    pub max_disk_gb: f32,
+    pub max_disk: ByteSize,
     pub system_name: String,
     pub auto_shutdown_ms: u64,
 }
@@ -474,27 +474,27 @@ fn write_buffer_with_style(buffer: &Buffer, out: &mut impl Write) -> io::Result<
     Ok(())
 }
 
-fn map_color(color: ratatui::style::Color) -> CrosstermColor {
+fn map_color(color: Color) -> CrosstermColor {
     match color {
-        ratatui::style::Color::Reset => CrosstermColor::Reset,
+        Color::Reset => CrosstermColor::Reset,
-        ratatui::style::Color::Black => CrosstermColor::Black,
+        Color::Black => CrosstermColor::Black,
-        ratatui::style::Color::Red => CrosstermColor::DarkRed,
+        Color::Red => CrosstermColor::DarkRed,
-        ratatui::style::Color::Green => CrosstermColor::DarkGreen,
+        Color::Green => CrosstermColor::DarkGreen,
-        ratatui::style::Color::Yellow => CrosstermColor::DarkYellow,
+        Color::Yellow => CrosstermColor::DarkYellow,
-        ratatui::style::Color::Blue => CrosstermColor::DarkBlue,
+        Color::Blue => CrosstermColor::DarkBlue,
-        ratatui::style::Color::Magenta => CrosstermColor::DarkMagenta,
+        Color::Magenta => CrosstermColor::DarkMagenta,
-        ratatui::style::Color::Cyan => CrosstermColor::DarkCyan,
+        Color::Cyan => CrosstermColor::DarkCyan,
-        ratatui::style::Color::Gray => CrosstermColor::Grey,
+        Color::Gray => CrosstermColor::Grey,
-        ratatui::style::Color::DarkGray => CrosstermColor::DarkGrey,
+        Color::DarkGray => CrosstermColor::DarkGrey,
-        ratatui::style::Color::LightRed => CrosstermColor::Red,
+        Color::LightRed => CrosstermColor::Red,
-        ratatui::style::Color::LightGreen => CrosstermColor::Green,
+        Color::LightGreen => CrosstermColor::Green,
-        ratatui::style::Color::LightYellow => CrosstermColor::Yellow,
+        Color::LightYellow => CrosstermColor::Yellow,
-        ratatui::style::Color::LightBlue => CrosstermColor::Blue,
+        Color::LightBlue => CrosstermColor::Blue,
-        ratatui::style::Color::LightMagenta => CrosstermColor::Magenta,
+        Color::LightMagenta => CrosstermColor::Magenta,
-        ratatui::style::Color::LightCyan => CrosstermColor::Cyan,
+        Color::LightCyan => CrosstermColor::Cyan,
-        ratatui::style::Color::White => CrosstermColor::White,
+        Color::White => CrosstermColor::White,
-        ratatui::style::Color::Rgb(r, g, b) => CrosstermColor::Rgb { r, g, b },
+        Color::Rgb(r, g, b) => CrosstermColor::Rgb { r, g, b },
-        ratatui::style::Color::Indexed(i) => CrosstermColor::AnsiValue(i),
+        Color::Indexed(i) => CrosstermColor::AnsiValue(i),
     }
 }
 
@@ -574,8 +574,8 @@ fn render_header(buffer: &mut Buffer, area: Rect, app: &AppState) {
             Span::raw("CPU / Memory / Disk: "),
             Span::styled(
                 format!(
-                    "{} cores / {} MB / {} GB",
+                    "{} cores / {} / {}",
-                    app.vm_info.cpu_cores, app.vm_info.max_memory_mb, app.vm_info.max_disk_gb
+                    app.vm_info.cpu_cores, app.vm_info.max_memory, app.vm_info.max_disk
                 ),
                 Style::default().fg(Color::Green),
             ),

src/utils.rs

@@ -0,0 +1,18 @@
+use std::{
+    env,
+    path::{Path, PathBuf},
+};
+
+pub fn relative_to_home(directory: &Path) -> String {
+    let Ok(home) = env::var("HOME") else {
+        return directory.display().to_string();
+    };
+    let home_path = PathBuf::from(home);
+    if let Ok(stripped) = directory.strip_prefix(&home_path) {
+        if stripped.components().next().is_none() {
+            return "~".to_string();
+        }
+        return format!("~/{}", stripped.display());
+    }
+    directory.display().to_string()
+}

src/vm.rs

@@ -1,5 +1,5 @@
-use crate::instance::STATUS_FILE_NAME;
 use crate::session_manager::{GLOBAL_CACHE_DIR_NAME, INSTANCE_DIR_NAME};
+use anyhow::{Context, Error, Result, bail};
 use std::{
     env, fs,
     io::{self, Write},
@@ -23,6 +23,7 @@ use std::{
 };
 
 use block2::RcBlock;
+use bytesize::ByteSize;
 use dispatch2::DispatchQueue;
 use objc2::{AnyThread, rc::Retained, runtime::ProtocolObject};
 use objc2_foundation::*;
@@ -34,49 +35,19 @@ const DEBIAN_COMPRESSED_SIZE_BYTES: u64 = 280901576;
 const SHARED_DIRECTORIES_TAG: &str = "shared";
 pub const PROJECT_GUEST_BASE: &str = "/usr/local/vibebox-mounts";
 
-const BYTES_PER_MB: u64 = 1024 * 1024;
-const DEFAULT_CPU_COUNT: usize = 2;
-const DEFAULT_RAM_MB: u64 = 2048;
-const DEFAULT_RAM_BYTES: u64 = DEFAULT_RAM_MB * BYTES_PER_MB;
 const START_TIMEOUT: Duration = Duration::from_secs(60);
 const LOGIN_EXPECT_TIMEOUT: Duration = Duration::from_secs(120);
 const PROVISION_EXPECT_TIMEOUT: Duration = Duration::from_secs(900);
 
-struct StatusFile {
+const ERROR_REPORT_SCRIPT: &str = include_str!("error_report.sh");
-    path: PathBuf,
-    cleared: AtomicBool,
-}
-
-impl StatusFile {
-    fn new(path: PathBuf) -> Self {
-        Self {
-            path,
-            cleared: AtomicBool::new(false),
-        }
-    }
-
-    fn update(&self, message: &str) {
-        let _ = fs::write(&self.path, message);
-    }
-}
-
-impl Drop for StatusFile {
-    fn drop(&mut self) {
-        if !self.cleared.load(Ordering::SeqCst) {
-            let _ = fs::remove_file(&self.path);
-            self.cleared.store(true, Ordering::SeqCst);
-        }
-    }
-}
 const PROVISION_SCRIPT: &str = include_str!("provision.sh");
-const PROVISION_SCRIPT_NAME: &str = "provision.sh";
 const RESIZE_DISK_SCRIPT: &str = include_str!("resize_disk.sh");
 const DEFAULT_RAW_NAME: &str = "default.raw";
 const INSTANCE_RAW_NAME: &str = "instance.raw";
 const BASE_DISK_RAW_NAME: &str = "disk.raw";
 
 #[derive(Clone)]
-pub(crate) enum LoginAction {
+pub enum LoginAction {
     Expect {
         text: String,
         timeout: Duration,
@@ -88,23 +59,20 @@ pub(crate) enum LoginAction {
     },
     Send(String),
 }
+use crate::config::BoxConfig;
 use LoginAction::*;
 
 #[derive(Clone)]
-pub(crate) struct DirectoryShare {
+pub struct DirectoryShare {
     host: PathBuf,
     guest: PathBuf,
     read_only: bool,
 }
 
 impl DirectoryShare {
-    pub(crate) fn new(
+    pub fn new(host: PathBuf, mut guest: PathBuf, read_only: bool) -> Result<Self> {
-        host: PathBuf,
-        mut guest: PathBuf,
-        read_only: bool,
-    ) -> Result<Self, Box<dyn std::error::Error>> {
         if !host.exists() {
-            return Err(format!("Host path does not exist: {}", host.display()).into());
+            bail!(format!("host path does not exist: {}", host.display()));
         }
         if !guest.is_absolute() {
             guest = PathBuf::from("/root").join(guest);
@@ -116,10 +84,10 @@ impl DirectoryShare {
         })
     }
 
-    pub(crate) fn from_mount_spec(spec: &str) -> Result<Self, Box<dyn std::error::Error>> {
+    pub fn from_mount_spec(spec: &str) -> Result<Self> {
         let parts: Vec<&str> = spec.split(':').collect();
         if parts.len() < 2 || parts.len() > 3 {
-            return Err(format!("Invalid mount spec: {spec}").into());
+            bail!(format!("invalid mount spec: {spec}"));
         }
         let host = expand_tilde_path(parts[0]);
         let guest = PathBuf::from(parts[1]);
@@ -128,11 +96,10 @@ impl DirectoryShare {
                 "read-only" => true,
                 "read-write" => false,
                 _ => {
-                    return Err(format!(
+                    bail!(format!(
                         "Invalid mount mode '{}'; expected read-only or read-write",
                         parts[2]
-                    )
+                    ));
-                    .into());
                 }
             }
         } else {
@@ -176,19 +143,28 @@ pub struct VmArg {
     pub mounts: Vec<String>,
 }
 
-pub fn run_with_args<F>(args: VmArg, io_handler: F) -> Result<(), Box<dyn std::error::Error>>
+type StatusEmitter<'a> = dyn Fn(&str) + std::marker::Send + Sync + 'a;
+
+fn emit_status(status: Option<&StatusEmitter<'_>>, message: &str) {
+    if let Some(status) = status {
+        status(message);
+    }
+}
+
+pub fn run_with_args<F>(args: VmArg, io_handler: F) -> Result<()>
 where
     F: FnOnce(Arc<OutputMonitor>, OwnedFd, OwnedFd) -> IoContext,
 {
-    run_with_args_and_extras(args, io_handler, Vec::new(), Vec::new())
+    run_with_args_and_extras(args, io_handler, Vec::new(), Vec::new(), None)
 }
 
-pub(crate) fn run_with_args_and_extras<F>(
+pub fn run_with_args_and_extras<F>(
     args: VmArg,
     io_handler: F,
     extra_login_actions: Vec<LoginAction>,
     extra_directory_shares: Vec<DirectoryShare>,
-) -> Result<(), Box<dyn std::error::Error>>
+    status: Option<&StatusEmitter<'_>>,
+) -> Result<()>
 where
     F: FnOnce(Arc<OutputMonitor>, OwnedFd, OwnedFd) -> IoContext,
 {
@@ -197,7 +173,7 @@ where
     let project_root = env::current_dir()?;
     let project_name = project_root
         .file_name()
-        .ok_or("Project directory has no name")?
+        .with_context(|| "Project directory has no name")?
         .to_string_lossy()
         .into_owned();
 
@@ -209,8 +185,10 @@ where
     let guest_mise_cache = cache_dir.join(".guest-mise-cache");
 
     let instance_dir = project_root.join(INSTANCE_DIR_NAME);
-    let status_file = StatusFile::new(instance_dir.join(STATUS_FILE_NAME));
+    fs::create_dir_all(&instance_dir)?;
-    status_file.update("preparing VM image...");
+    emit_status(status, "preparing VM image...");
+    tracing::info!("preparing VM image...");
+    let provision_log = instance_dir.join("provision.log");
 
     let basename_compressed = DEBIAN_COMPRESSED_DISK_URL.rsplit('/').next().unwrap();
     let base_compressed = cache_dir.join(basename_compressed);
@@ -234,13 +212,14 @@ where
         &base_compressed,
         &default_raw,
         std::slice::from_ref(&mise_directory_share),
-        Some(&status_file),
+        Some(&provision_log),
+        status,
     )?;
     let _ = ensure_instance_disk(
         &instance_raw,
         &default_raw,
-        args.disk_bytes,
+        ByteSize(args.disk_bytes),
-        Some(&status_file),
+        status,
     )?;
     let base_size = fs::metadata(&default_raw)?.len();
     let instance_size = fs::metadata(&instance_raw)?.len();
@@ -270,7 +249,7 @@ where
     }
 
     if needs_resize {
-        let resize_cmd = script_command_from_content("resize_disk", RESIZE_DISK_SCRIPT)?;
+        let resize_cmd = script_command_from_content("resize_disk.sh", RESIZE_DISK_SCRIPT)?;
         login_actions.push(Send(resize_cmd));
     }
 
@@ -286,25 +265,29 @@ where
         &directory_shares[..],
         args.cpu_count,
         args.ram_bytes,
-        Some(&status_file),
+        status,
         io_handler,
     )
 }
 
-pub(crate) fn script_command_from_content(
+pub fn script_command_from_content(label: &str, script: &str) -> Result<String> {
-    label: &str,
-    script: &str,
-) -> Result<String, Box<dyn std::error::Error>> {
     let marker = "VIBE_SCRIPT_EOF";
     let guest_dir = "/tmp/vibe-scripts";
     let guest_path = format!("{guest_dir}/{label}.sh");
+    let script_body = match script.split_once('\n') {
+        Some((first, rest)) if first.starts_with("#!") => rest,
+        _ => script,
+    };
+    let wrapped_script = ERROR_REPORT_SCRIPT
+        .replace("__LABEL__", label)
+        .replace("__SCRIPT_BODY__", script_body);
     let command = format!(
-        "mkdir -p {guest_dir}\ncat >{guest_path} <<'{marker}'\n{script}\n{marker}\nchmod +x {guest_path}\n{guest_path}"
+        "mkdir -p {guest_dir}\ncat >{guest_path} <<'{marker}'\n{wrapped_script}\n{marker}\nchmod +x {guest_path}\n{guest_path}"
     );
     if script.contains(marker) {
-        return Err(
+        bail!(format!(
-            format!("Script '{label}' contains marker '{marker}', cannot safely upload").into(),
+            "Script '{label}' contains marker '{marker}', cannot safely upload"
-        );
+        ));
     }
     Ok(command)
 }
@@ -501,18 +484,16 @@ impl IoControl {
 fn ensure_base_image(
     base_raw: &Path,
     base_compressed: &Path,
-    status: Option<&StatusFile>,
+    status: Option<&StatusEmitter<'_>>,
-) -> Result<(), Box<dyn std::error::Error>> {
+) -> Result<()> {
     if base_raw.exists() {
         return Ok(());
     }
 
     if !base_compressed.exists()
-        || std::fs::metadata(base_compressed).map(|m| m.len())? < DEBIAN_COMPRESSED_SIZE_BYTES
+        || fs::metadata(base_compressed).map(|m| m.len())? < DEBIAN_COMPRESSED_SIZE_BYTES
     {
-        if let Some(status) = status {
+        emit_status(status, "downloading base image...");
-            status.update("downloading base image...");
-        }
         tracing::info!("downloading base image");
         let status = Command::new("curl")
             .args([
@@ -527,15 +508,14 @@ fn ensure_base_image(
             ])
             .status()?;
         if !status.success() {
-            return Err("Failed to download base image".into());
+            bail!("failed to download base image");
         }
     }
 
     // Check SHA
     {
-        if let Some(status) = status {
+        emit_status(status, "verifying base image...");
-            status.update("verifying base image...");
+        tracing::info!("verifying base image...");
-        }
         let input = format!("{}  {}\n", DEBIAN_COMPRESSED_SHA, base_compressed.display());
 
         let mut child = Command::new("/usr/bin/shasum")
@@ -553,25 +533,25 @@ fn ensure_base_image(
 
         let status = child.wait().expect("failed to wait on child");
         if !status.success() {
-            return Err(format!("SHA validation failed for {DEBIAN_COMPRESSED_DISK_URL}").into());
+            bail!(format!(
+                "SHA validation failed for {DEBIAN_COMPRESSED_DISK_URL}"
+            ));
         }
     }
 
-    if let Some(status) = status {
+    emit_status(status, "decompressing base image...");
-        status.update("decompressing base image...");
+    tracing::info!("decompressing base image...");
-    }
-    tracing::info!("decompressing base image");
     let status = Command::new("tar")
         .args([
             "-xOf",
             &base_compressed.to_string_lossy(),
             BASE_DISK_RAW_NAME,
         ])
-        .stdout(std::fs::File::create(base_raw)?)
+        .stdout(fs::File::create(base_raw)?)
         .status()?;
 
     if !status.success() {
-        return Err("Failed to decompress base image".into());
+        bail!("Failed to decompress base image");
     }
 
     Ok(())
@@ -582,21 +562,20 @@ fn ensure_default_image(
     base_compressed: &Path,
     default_raw: &Path,
     directory_shares: &[DirectoryShare],
-    status: Option<&StatusFile>,
+    provision_log: Option<&Path>,
-) -> Result<(), Box<dyn std::error::Error>> {
+    status: Option<&StatusEmitter<'_>>,
+) -> Result<()> {
     if default_raw.exists() {
         return Ok(());
     }
 
     ensure_base_image(base_raw, base_compressed, status)?;
 
-    if let Some(status) = status {
+    emit_status(status, "configuring base image...");
-        status.update("configuring base image...");
+    tracing::info!("configuring base image...");
-    }
-    tracing::info!("configuring base image");
     fs::copy(base_raw, default_raw)?;
 
-    let provision_command = script_command_from_content(PROVISION_SCRIPT_NAME, PROVISION_SCRIPT)?;
+    let provision_command = script_command_from_content("provision.sh", PROVISION_SCRIPT)?;
     let provision_actions = [
         Send(provision_command),
         ExpectEither {
@@ -605,14 +584,31 @@ fn ensure_default_image(
             timeout: PROVISION_EXPECT_TIMEOUT,
         },
     ];
-    if let Err(err) = run_vm(
+    let provision_result = if let Some(log_path) = provision_log {
+        let log_path = log_path.to_path_buf();
+        run_vm_with_io(
             default_raw,
             &provision_actions,
             directory_shares,
-        DEFAULT_CPU_COUNT,
+            BoxConfig::default().cpu_count,
-        DEFAULT_RAM_BYTES,
+            BoxConfig::default().ram_size.as_u64(),
-        None,
+            status,
-    ) {
+            move |output_monitor, vm_output_fd, vm_input_fd| {
+                spawn_vm_io_with_log(output_monitor, vm_output_fd, vm_input_fd, log_path)
+            },
+        )
+    } else {
+        run_vm(
+            default_raw,
+            &provision_actions,
+            directory_shares,
+            BoxConfig::default().cpu_count,
+            BoxConfig::default().ram_size.as_u64(),
+            status,
+        )
+    };
+
+    if let Err(err) = provision_result {
         let _ = fs::remove_file(default_raw);
         return Err(err);
     }
@@ -623,18 +619,16 @@ fn ensure_default_image(
 fn ensure_instance_disk(
     instance_raw: &Path,
     template_raw: &Path,
-    target_bytes: u64,
+    target_bytes: ByteSize,
-    status: Option<&StatusFile>,
+    status: Option<&StatusEmitter<'_>>,
-) -> Result<bool, Box<dyn std::error::Error>> {
+) -> Result<bool> {
     if instance_raw.exists() {
-        let current_size = fs::metadata(instance_raw)?.len();
+        let current_size = ByteSize(fs::metadata(instance_raw)?.len());
         if current_size != target_bytes {
-            let current_gb = current_size as f64 / (1024.0 * 1024.0 * 1024.0);
+            let current_gb = current_size;
-            let target_gb = target_bytes as f64 / (1024.0 * 1024.0 * 1024.0);
+            let target_gb = target_bytes;
             tracing::warn!(
-                current_bytes = current_size,
+                "instance disk size does not match config (current {}, config {}); disk_gb applies only on init. Run `vibebox reset` to recreate or set disk_gb to match; using existing disk.",
-                target_bytes,
-                "instance disk size does not match config (current {:.2} GB, config {:.2} GB); disk_gb applies only on init. Run `vibebox reset` to recreate or set disk_gb to match; using existing disk.",
                 current_gb,
                 target_gb
             );
@@ -642,30 +636,28 @@ fn ensure_instance_disk(
         return Ok(false);
     }
 
-    let template_size = fs::metadata(template_raw)?.len();
+    let template_size = ByteSize(fs::metadata(template_raw)?.len());
     if target_bytes < template_size {
-        return Err(format!(
+        bail!(format!(
-            "Requested disk size {} bytes is smaller than base image size {} bytes",
+            "Requested disk size {} is smaller than base image size {}",
             target_bytes, template_size
-        )
+        ));
-        .into());
     }
     let target_size = target_bytes;
     let needs_resize = target_size > template_size;
 
-    if let Some(status) = status {
+    emit_status(status, "creating instance disk...");
-        status.update("creating instance disk...");
+    tracing::info!("creating instance disk...");
-    }
     tracing::info!(path = %template_raw.display(), "creating instance disk");
-    std::fs::create_dir_all(instance_raw.parent().unwrap())?;
+    fs::create_dir_all(instance_raw.parent().unwrap())?;
     if target_size == template_size {
         fs::copy(template_raw, instance_raw)?;
         return Ok(needs_resize);
     }
 
-    let mut dst = std::fs::File::create(instance_raw)?;
+    let mut dst = fs::File::create(instance_raw)?;
-    dst.set_len(target_size)?;
+    dst.set_len(target_size.as_u64())?;
-    let mut src = std::fs::File::open(template_raw)?;
+    let mut src = fs::File::open(template_raw)?;
     std::io::copy(&mut src, &mut dst)?;
     Ok(needs_resize)
 }
@@ -683,18 +675,17 @@ pub fn create_pipe() -> (OwnedFd, OwnedFd) {
     (read_stream.into(), write_stream.into())
 }
 
-pub fn spawn_vm_io_with_hooks<F, G>(
+pub fn spawn_vm_io_with_hooks<
+    F: FnMut(&str) -> bool + std::marker::Send + 'static,
+    G: FnMut(&[u8]) + std::marker::Send + 'static,
+>(
     output_monitor: Arc<OutputMonitor>,
     vm_output_fd: OwnedFd,
     vm_input_fd: OwnedFd,
     io_control: Arc<IoControl>,
     mut on_line: F,
     mut on_output: G,
-) -> IoContext
+) -> IoContext {
-where
-    F: FnMut(&str) -> bool + ::std::marker::Send + 'static,
-    G: FnMut(&[u8]) + ::std::marker::Send + 'static,
-{
     let (input_tx, input_rx): (Sender<VmInput>, Receiver<VmInput>) = mpsc::channel();
 
     // raw_guard is set when we've put the user's terminal into raw mode because we've attached stdin/stdout to the VM.
@@ -842,14 +833,14 @@ where
                     PollResult::Spurious => continue,
                     PollResult::Ready(bytes) => {
                         if io_control.forward_output() {
-                            // enable raw mode, if we haven't already
+                            // enable raw mode if we haven't already
                             if raw_guard.lock().unwrap().is_none()
                                 && let Ok(guard) = enable_raw_mode(libc::STDIN_FILENO)
                             {
                                 *raw_guard.lock().unwrap() = Some(guard);
                             }
 
-                            let mut stdout = std::io::stdout().lock();
+                            let mut stdout = io::stdout().lock();
                             if stdout.write_all(bytes).is_err() {
                                 break;
                             }
@@ -865,7 +856,7 @@ where
 
     // Copies data from mpsc channel into VM, so vibe can "type" stuff and run scripts.
     let mux_thread = thread::spawn(move || {
-        let mut vm_writer = std::fs::File::from(vm_input_fd);
+        let mut vm_writer = fs::File::from(vm_input_fd);
         loop {
             match input_rx.recv() {
                 Ok(VmInput::Bytes(data)) => {
@@ -895,7 +886,7 @@ pub fn spawn_vm_io_with_line_handler<F>(
     on_line: F,
 ) -> IoContext
 where
-    F: FnMut(&str) -> bool + ::std::marker::Send + 'static,
+    F: FnMut(&str) -> bool + std::marker::Send + 'static,
 {
     spawn_vm_io_with_hooks(
         output_monitor,
@@ -915,6 +906,36 @@ pub fn spawn_vm_io(
     spawn_vm_io_with_line_handler(output_monitor, vm_output_fd, vm_input_fd, |_| false)
 }
 
+fn spawn_vm_io_with_log(
+    output_monitor: Arc<OutputMonitor>,
+    vm_output_fd: OwnedFd,
+    vm_input_fd: OwnedFd,
+    log_path: PathBuf,
+) -> IoContext {
+    let log_file = fs::OpenOptions::new()
+        .create(true)
+        .write(true)
+        .truncate(true)
+        .open(&log_path)
+        .ok()
+        .map(|file| Arc::new(Mutex::new(file)));
+
+    spawn_vm_io_with_hooks(
+        output_monitor,
+        vm_output_fd,
+        vm_input_fd,
+        IoControl::new(),
+        |_| false,
+        move |bytes| {
+            if let Some(log) = &log_file
+                && let Ok(mut file) = log.lock()
+            {
+                let _ = file.write_all(bytes);
+            }
+        },
+    )
+}
+
 impl IoContext {
     pub fn shutdown(self) {
         let _ = self.input_tx.send(VmInput::Shutdown);
@@ -932,7 +953,7 @@ fn create_vm_configuration(
     vm_writes_to_fd: OwnedFd,
     cpu_count: usize,
     ram_bytes: u64,
-) -> Result<Retained<VZVirtualMachineConfiguration>, Box<dyn std::error::Error>> {
+) -> Result<Retained<VZVirtualMachineConfiguration>> {
     unsafe {
         let platform =
             VZGenericPlatformConfiguration::init(VZGenericPlatformConfiguration::alloc());
@@ -966,7 +987,7 @@ fn create_vm_configuration(
             false,
             VZDiskImageCachingMode::Automatic,
             VZDiskImageSynchronizationMode::Full,
-        ).unwrap();
+        )?;
 
             let disk_device = VZVirtioBlockDeviceConfiguration::initWithAttachment(
                 VZVirtioBlockDeviceConfiguration::alloc(),
@@ -1053,7 +1074,7 @@ fn create_vm_configuration(
         // Validate
         config.validateWithError().map_err(|e| {
             io::Error::other(format!(
-                "Invalid VM configuration: {:?}",
+                "invalid VM configuration: {:?}",
                 e.localizedDescription()
             ))
         })?;
@@ -1062,9 +1083,9 @@ fn create_vm_configuration(
     }
 }
 
-fn load_efi_variable_store() -> Result<Retained<VZEFIVariableStore>, Box<dyn std::error::Error>> {
+fn load_efi_variable_store() -> Result<Retained<VZEFIVariableStore>> {
     unsafe {
-        let temp_dir = std::env::temp_dir();
+        let temp_dir = env::temp_dir();
         let temp_path = temp_dir.join(format!("efi_variable_store_{}.efivars", std::process::id()));
         let url = nsurl_from_path(&temp_path)?;
         let options = VZEFIVariableStoreInitializationOptions::AllowOverwrite;
@@ -1080,8 +1101,8 @@ fn load_efi_variable_store() -> Result<Retained<VZEFIVariableStore>, Box<dyn std
 fn spawn_login_actions_thread(
     login_actions: Vec<LoginAction>,
     output_monitor: Arc<OutputMonitor>,
-    input_tx: mpsc::Sender<VmInput>,
+    input_tx: Sender<VmInput>,
-    vm_output_tx: mpsc::Sender<VmOutput>,
+    vm_output_tx: Sender<VmOutput>,
 ) -> thread::JoinHandle<()> {
     thread::spawn(move || {
         for a in login_actions {
@@ -1131,9 +1152,9 @@ fn run_vm_with_io<F>(
     directory_shares: &[DirectoryShare],
     cpu_count: usize,
     ram_bytes: u64,
-    status: Option<&StatusFile>,
+    status: Option<&StatusEmitter<'_>>,
     io_handler: F,
-) -> Result<(), Box<dyn std::error::Error>>
+) -> Result<()>
 where
     F: FnOnce(Arc<OutputMonitor>, OwnedFd, OwnedFd) -> IoContext,
 {
@@ -1180,23 +1201,22 @@ where
 
         match rx.try_recv() {
             Ok(result) => {
-                result.map_err(|e| format!("Failed to start VM: {}", e))?;
+                result.map_err(|e| Error::msg(format!("Failed to start VM: {}", e)))?;
                 break;
             }
             Err(mpsc::TryRecvError::Empty) => continue,
             Err(mpsc::TryRecvError::Disconnected) => {
-                return Err("VM start channel disconnected".into());
+                bail!("VM start channel disconnected");
             }
         }
     }
 
     if Instant::now() >= start_deadline {
-        return Err("Timed out waiting for VM to start".into());
+        bail!("Timed out waiting for VM to start");
     }
 
-    if let Some(status) = status {
+    emit_status(status, "vm booting... go vibecoder!");
-        status.update("vm booting... go vibecoder!");
+    tracing::info!("vm booting... go vibecoder!");
-    }
     tracing::info!("vm booting");
 
     let output_monitor = Arc::new(OutputMonitor::default());
@@ -1242,7 +1262,7 @@ where
     );
 
     let mut last_state = None;
-    let mut exit_result = Ok(());
+    let mut exit_result: Result<(), String> = Ok(());
     loop {
         unsafe {
             NSRunLoop::mainRunLoop().runMode_beforeDate(
@@ -1261,8 +1281,7 @@ where
                 exit_result = Err(format!(
                     "Login action ({}) timed out after {:?}; shutting down.",
                     action, timeout
-                )
+                ));
-                .into());
                 unsafe {
                     if vm.canRequestStop() {
                         if let Err(err) = vm.requestStopWithError() {
@@ -1279,8 +1298,7 @@ where
                 exit_result = Err(format!(
                     "Login action ({}) failed: {}; shutting down.",
                     action, reason
-                )
+                ));
-                .into());
                 unsafe {
                     if vm.canRequestStop() {
                         if let Err(err) = vm.requestStopWithError() {
@@ -1296,7 +1314,7 @@ where
             Err(mpsc::TryRecvError::Empty) => {}
             Err(mpsc::TryRecvError::Disconnected) => {}
         }
-        if state != objc2_virtualization::VZVirtualMachineState::Running {
+        if state != VZVirtualMachineState::Running {
             //eprintln!("VM stopped with state: {:?}", state);
             break;
         }
@@ -1306,7 +1324,7 @@ where
 
     io_ctx.shutdown();
 
-    exit_result
+    exit_result.map_err(Error::msg)
 }
 
 fn run_vm(
@@ -1315,8 +1333,8 @@ fn run_vm(
     directory_shares: &[DirectoryShare],
     cpu_count: usize,
     ram_bytes: u64,
-    status: Option<&StatusFile>,
+    status: Option<&StatusEmitter<'_>>,
-) -> Result<(), Box<dyn std::error::Error>> {
+) -> Result<()> {
     run_vm_with_io(
         disk_path,
         login_actions,
@@ -1328,7 +1346,7 @@ fn run_vm(
     )
 }
 
-fn nsurl_from_path(path: &Path) -> Result<Retained<NSURL>, Box<dyn std::error::Error>> {
+fn nsurl_from_path(path: &Path) -> Result<Retained<NSURL>> {
     let abs_path = if path.is_absolute() {
         path.to_path_buf()
     } else {
@@ -1337,7 +1355,7 @@ fn nsurl_from_path(path: &Path) -> Result<Retained<NSURL>, Box<dyn std::error::E
     let ns_path = NSString::from_str(
         abs_path
             .to_str()
-            .ok_or("Non-UTF8 path encountered while building NSURL")?,
+            .with_context(|| "non-UTF8 path encountered while building NSURL")?,
     );
     Ok(NSURL::fileURLWithPath(&ns_path))
 }
@@ -1352,7 +1370,7 @@ fn enable_raw_mode(fd: i32) -> io::Result<RawModeGuard> {
     let original = attributes;
 
     // Disable translation of carriage return to newline on input
-    attributes.c_iflag &= !(libc::ICRNL);
+    attributes.c_iflag &= !libc::ICRNL;
     // Disable canonical mode (line buffering), echo, and signal generation
     attributes.c_lflag &= !(libc::ICANON | libc::ECHO | libc::ISIG);
     attributes.c_cc[libc::VMIN] = 0;
@@ -1380,10 +1398,10 @@ impl Drop for RawModeGuard {
 
 // Ensure the running binary has com.apple.security.virtualization entitlements by checking and, if not, signing and relaunching.
 pub fn ensure_signed() {
-    if std::env::var("VIBEBOX_SKIP_CODESIGN").as_deref() == Ok("1") {
+    if env::var("VIBEBOX_SKIP_CODESIGN").as_deref() == Ok("1") {
         return;
     }
-    let exe = std::env::current_exe().expect("failed to get current exe path");
+    let exe = env::current_exe().expect("failed to get current exe path");
     let exe_str = exe.to_str().expect("exe path not valid utf-8");
 
     let has_required_entitlements = {
@@ -1405,8 +1423,8 @@ pub fn ensure_signed() {
     }
 
     const ENTITLEMENTS: &str = include_str!("entitlements.plist");
-    let entitlements_path = std::env::temp_dir().join("entitlements.plist");
+    let entitlements_path = env::temp_dir().join("entitlements.plist");
-    std::fs::write(&entitlements_path, ENTITLEMENTS).expect("failed to write entitlements");
+    fs::write(&entitlements_path, ENTITLEMENTS).expect("failed to write entitlements");
 
     let output = Command::new("codesign")
         .args([
@@ -1419,7 +1437,7 @@ pub fn ensure_signed() {
         ])
         .output();
 
-    let _ = std::fs::remove_file(&entitlements_path);
+    let _ = fs::remove_file(&entitlements_path);
 
     match output {
         Ok(o) if o.status.success() => {
@@ -1427,7 +1445,7 @@ pub fn ensure_signed() {
             if !stderr.trim().is_empty() {
                 tracing::debug!(codesign_stderr = %stderr.trim(), "codesign output");
             }
-            let err = Command::new(&exe).args(std::env::args_os().skip(1)).exec();
+            let err = Command::new(&exe).args(env::args_os().skip(1)).exec();
             tracing::error!(error = %err, "failed to re-exec after signing");
             std::process::exit(1);
         }

src/vm_manager.rs

@@ -1,12 +1,22 @@
+use crate::instance::{VmLiveness, vm_liveness};
+use crate::session_manager::INSTANCE_DIR_NAME;
+use crate::{
+    config::CONFIG_PATH_ENV,
+    instance::{
+        InstanceConfig, STATUS_VM_ERROR_PREFIX, build_ssh_login_actions, ensure_instance_dir,
+        ensure_ssh_keypair, extract_ipv4, load_or_create_instance_config, write_instance_config,
+    },
+    session_manager::{GLOBAL_DIR_NAME, VM_MANAGER_PID_NAME, VM_MANAGER_SOCKET_NAME},
+    vm::{self, DirectoryShare, LoginAction, PROJECT_GUEST_BASE, VmInput},
+};
+use anyhow::{Error, Result, anyhow, bail};
 use std::{
     env, fs,
     io::{Read, Write},
     os::unix::{
-        fs::FileTypeExt,
         fs::PermissionsExt,
         io::AsRawFd,
         net::{UnixListener, UnixStream},
-        process::CommandExt,
     },
     path::{Path, PathBuf},
     process::{Command, Stdio},
@@ -15,19 +25,8 @@ use std::{
     time::{Duration, Instant},
 };
 
-use crate::{
+#[cfg_attr(feature = "mock-vm", allow(dead_code))]
-    config::CONFIG_PATH_ENV,
+const VM_ROOT_LOG_NAME: &str = "vm_root.log";
-    instance::VM_ROOT_LOG_NAME,
-    instance::{
-        DEFAULT_SSH_USER, InstanceConfig, build_ssh_login_actions, ensure_instance_dir,
-        ensure_ssh_keypair, extract_ipv4, load_or_create_instance_config, write_instance_config,
-    },
-    session_manager::{
-        GLOBAL_DIR_NAME, INSTANCE_FILENAME, VM_MANAGER_PID_NAME, VM_MANAGER_SOCKET_NAME,
-    },
-    vm::{self, DirectoryShare, LoginAction, PROJECT_GUEST_BASE, VmInput},
-};
-
 const VM_MANAGER_LOCK_NAME: &str = "vm.lock";
 const VM_MANAGER_LOG_NAME: &str = "vm_manager.log";
 const SHUTDOWN_RETRY_MS: u64 = 500;
@@ -35,6 +34,10 @@ const SHUTDOWN_RETRY_MS: u64 = 500;
 const HARD_SHUTDOWN_TIMEOUT_MS: u64 = 1_000;
 #[cfg(not(test))]
 const HARD_SHUTDOWN_TIMEOUT_MS: u64 = 12_000;
+const STATUS_PREFIX: &str = "status:";
+
+type ClientStreams = Arc<Mutex<Vec<UnixStream>>>;
+type SharedStatus = Arc<Mutex<String>>;
 
 #[cfg(not(test))]
 fn force_exit(_reason: &str) -> ! {
@@ -50,16 +53,16 @@ pub fn ensure_manager(
     raw_args: &[std::ffi::OsString],
     auto_shutdown_ms: u64,
     config_path: Option<&Path>,
-) -> Result<UnixStream, Box<dyn std::error::Error>> {
+) -> Result<UnixStream> {
     let project_root = env::current_dir()?;
     tracing::debug!(root = %project_root.display(), "ensure vm manager");
     let instance_dir = ensure_instance_dir(&project_root)?;
-    cleanup_stale_manager(&instance_dir);
+    cleanup_stale_manager(&project_root)?;
     let socket_path = instance_dir.join(VM_MANAGER_SOCKET_NAME);
 
     if let Ok(stream) = UnixStream::connect(&socket_path) {
         send_client_pid(&stream);
-        tracing::info!(path = %socket_path.display(), "connected to existing vm manager");
+        tracing::info!(path = %socket_path.display(), "connected to an existing vm manager");
         return Ok(stream);
     }
 
@@ -97,12 +100,11 @@ pub fn ensure_manager(
                         drop(lock_file.take());
                         let _ = fs::remove_file(&lock_path);
                     }
-                    return Err(format!(
+                    bail!(format!(
-                        "Timed out waiting for vm manager socket: {} ({})",
+                        "timed out waiting for vm manager socket: {} ({})",
                         socket_path.display(),
                         err
-                    )
+                    ));
-                    .into());
                 }
                 thread::sleep(Duration::from_millis(100));
             }
@@ -110,12 +112,19 @@ pub fn ensure_manager(
     }
 }
 
-pub fn run_manager(
+pub fn run_manager(args: vm::VmArg, auto_shutdown_ms: u64) -> Result<()> {
-    args: vm::VmArg,
-    auto_shutdown_ms: u64,
-) -> Result<(), Box<dyn std::error::Error>> {
     let project_root = env::current_dir()?;
     tracing::info!(root = %project_root.display(), "vm manager starting");
+    #[cfg(not(feature = "mock-vm"))]
+    {
+        unsafe {
+            env::remove_var("VIBEBOX_SKIP_CODESIGN");
+        }
+        vm::ensure_signed();
+        unsafe {
+            env::set_var("VIBEBOX_SKIP_CODESIGN", "1");
+        }
+    }
     let _pid_guard = ensure_pid_file(&project_root)?;
     #[cfg(feature = "mock-vm")]
     tracing::info!("vm manager using mock executor");
@@ -133,7 +142,6 @@ pub fn run_manager(
         #[cfg(feature = "mock-vm")]
         {
             ManagerOptions {
-                ensure_signed: false,
                 detach: true,
                 prepare_vm: false,
             }
@@ -141,7 +149,6 @@ pub fn run_manager(
         #[cfg(not(feature = "mock-vm"))]
         {
             ManagerOptions {
-                ensure_signed: true,
                 detach: true,
                 prepare_vm: true,
             }
@@ -155,25 +162,22 @@ fn spawn_manager_process(
     auto_shutdown_ms: u64,
     instance_dir: &Path,
     config_path: Option<&Path>,
-) -> Result<(), Box<dyn std::error::Error>> {
+) -> Result<()> {
     let exe = env::current_exe()?;
     let mut supervisor_exe = exe.clone();
     supervisor_exe.set_file_name("vibebox-supervisor");
-    let use_supervisor = supervisor_exe.exists();
+    // intentional
-    let mut cmd = if use_supervisor {
+    if !supervisor_exe.exists() {
-        Command::new(supervisor_exe)
+        bail!(format!(
-    } else {
+            "vibebox-supervisor not found at {}",
-        let mut cmd = Command::new(exe);
+            supervisor_exe.display()
-        cmd.arg0("vibebox-supervisor");
+        ));
-        cmd
+    }
-    };
+    let mut cmd = Command::new(supervisor_exe);
     if raw_args.len() > 1 {
         cmd.args(&raw_args[1..]);
     }
     cmd.env("VIBEBOX_INTERNAL", "1");
-    if !use_supervisor {
-        cmd.env("VIBEBOX_VM_MANAGER", "1");
-    }
     cmd.env("VIBEBOX_LOG_NO_COLOR", "1");
     cmd.env("VIBEBOX_AUTO_SHUTDOWN_MS", auto_shutdown_ms.to_string());
     if let Some(path) = config_path {
@@ -201,38 +205,41 @@ fn spawn_manager_process(
     Ok(())
 }
 
-fn ensure_pid_file(project_root: &Path) -> Result<PidFileGuard, Box<dyn std::error::Error>> {
+fn ensure_pid_file(project_root: &Path) -> Result<PidFileGuard> {
     let instance_dir = ensure_instance_dir(project_root)?;
     let pid_path = instance_dir.join(VM_MANAGER_PID_NAME);
     let socket_path = instance_dir.join(VM_MANAGER_SOCKET_NAME);
-    if let Ok(content) = fs::read_to_string(&pid_path)
+    match vm_liveness(project_root)? {
-        && let Ok(pid) = content.trim().parse::<u32>()
+        VmLiveness::RunningWithSocket { pid } => {
-        && pid_is_alive(pid)
+            bail!("vm manager already running (pid {pid})");
-    {
-        if is_socket_path(&socket_path) {
-            return Err(format!("vm manager already running (pid {pid})").into());
         }
+        VmLiveness::RunningWithoutSocket { pid } => {
             tracing::warn!(
                 pid,
                 path = %socket_path.display(),
                 "stale pid file detected with missing socket"
             );
         }
+        VmLiveness::NotRunningOrMissing => {}
+    }
     let _ = fs::remove_file(&pid_path);
     fs::write(&pid_path, format!("{}\n", std::process::id()))?;
     let _ = fs::set_permissions(&pid_path, fs::Permissions::from_mode(0o600));
     Ok(PidFileGuard { path: pid_path })
 }
 
-fn cleanup_stale_manager(instance_dir: &Path) {
+fn cleanup_stale_manager(project_root: &Path) -> Result<()> {
-    let pid_path = instance_dir.join(VM_MANAGER_PID_NAME);
+    let pid_path = project_root
-    if let Ok(content) = fs::read_to_string(&pid_path)
+        .join(INSTANCE_DIR_NAME)
-        && let Ok(pid) = content.trim().parse::<u32>()
+        .join(VM_MANAGER_PID_NAME);
-        && pid_is_alive(pid)
+    if matches!(
-    {
+        vm_liveness(project_root)?,
-        return;
+        VmLiveness::RunningWithSocket { .. } | VmLiveness::RunningWithoutSocket { .. }
+    ) {
+        return Ok(());
     }
     let _ = fs::remove_file(&pid_path);
+    Ok(())
 }
 
 fn inject_project_mount(
@@ -259,12 +266,6 @@ fn inject_project_mount(
     mounts.insert(0, format!("{host}:{guest_tilde}:read-write"));
 }
 
-fn is_socket_path(path: &Path) -> bool {
-    fs::metadata(path)
-        .map(|meta| meta.file_type().is_socket())
-        .unwrap_or(false)
-}
-
 fn prepare_mounts_and_links(mut args: vm::VmArg, ssh_user: &str) -> (vm::VmArg, String) {
     let mut links = Vec::new();
     let mut mounts = Vec::with_capacity(args.mounts.len());
@@ -408,6 +409,30 @@ fn wait_for_disconnect(mut stream: UnixStream) {
     }
 }
 
+fn remove_client(streams: &ClientStreams, fd: std::os::fd::RawFd) {
+    if let Ok(mut clients) = streams.lock() {
+        clients.retain(|stream| stream.as_raw_fd() != fd);
+    }
+}
+
+fn send_status_line(stream: &mut UnixStream, status: &str) -> bool {
+    let mut payload = String::with_capacity(STATUS_PREFIX.len() + status.len() + 1);
+    payload.push_str(STATUS_PREFIX);
+    payload.push_str(status);
+    payload.push('\n');
+    stream.write_all(payload.as_bytes()).is_ok()
+}
+
+#[cfg_attr(feature = "mock-vm", allow(dead_code))]
+fn broadcast_status(streams: &ClientStreams, latest_status: &SharedStatus, status: &str) {
+    if let Ok(mut current) = latest_status.lock() {
+        *current = status.to_string();
+    }
+    if let Ok(mut clients) = streams.lock() {
+        clients.retain_mut(|stream| send_status_line(stream, status));
+    }
+}
+
 fn send_client_pid(stream: &UnixStream) {
     let pid = std::process::id();
     let payload = format!("pid={pid}\n");
@@ -417,7 +442,7 @@ fn send_client_pid(stream: &UnixStream) {
     }
 }
 
-fn acquire_spawn_lock(lock_path: &Path) -> Result<Option<fs::File>, Box<dyn std::error::Error>> {
+fn acquire_spawn_lock(lock_path: &Path) -> Result<Option<fs::File>> {
     match fs::OpenOptions::new()
         .write(true)
         .create_new(true)
@@ -451,6 +476,12 @@ fn is_lock_stale(lock_path: &Path) -> bool {
     }
 }
 
+fn read_lock_pid(lock_path: &Path) -> Option<u32> {
+    let content = fs::read_to_string(lock_path).ok()?;
+    let line = content.lines().next()?;
+    line.strip_prefix("pid=")?.trim().parse::<u32>().ok()
+}
+
 fn pid_is_alive(pid: u32) -> bool {
     let pid = pid as libc::pid_t;
     let result = unsafe { libc::kill(pid, 0) };
@@ -464,12 +495,6 @@ fn pid_is_alive(pid: u32) -> bool {
     }
 }
 
-fn read_lock_pid(lock_path: &Path) -> Option<u32> {
-    let content = fs::read_to_string(lock_path).ok()?;
-    let line = content.lines().next()?;
-    line.strip_prefix("pid=")?.trim().parse::<u32>().ok()
-}
-
 fn read_client_pid(stream: &UnixStream) -> Option<u32> {
     let mut stream = stream.try_clone().ok()?;
     let _ = stream.set_read_timeout(Some(Duration::from_millis(200)));
@@ -505,12 +530,14 @@ fn read_client_pid(stream: &UnixStream) -> Option<u32> {
 #[cfg_attr(feature = "mock-vm", allow(dead_code))]
 fn spawn_manager_io(
     config: Arc<Mutex<InstanceConfig>>,
-    instance_dir: PathBuf,
+    project_dir: PathBuf,
+    clients: ClientStreams,
+    latest_status: SharedStatus,
     output_monitor: Arc<vm::OutputMonitor>,
     vm_output_fd: std::os::unix::io::OwnedFd,
     vm_input_fd: std::os::unix::io::OwnedFd,
 ) -> vm::IoContext {
-    let log_path = instance_dir.join(VM_ROOT_LOG_NAME);
+    let log_path = project_dir.join(INSTANCE_DIR_NAME).join(VM_ROOT_LOG_NAME);
     let log_file = fs::OpenOptions::new()
         .create(true)
         .write(true)
@@ -519,7 +546,6 @@ fn spawn_manager_io(
         .ok()
         .map(|file| Arc::new(Mutex::new(file)));
 
-    let instance_path = instance_dir.join(INSTANCE_FILENAME);
     let config_for_output = config.clone();
     let log_for_output = log_file.clone();
     let mut line_buf = String::new();
@@ -542,6 +568,17 @@ fn spawn_manager_io(
             line_buf.drain(..=pos);
 
             let cleaned = line.trim_start_matches(['\r', ' ']);
+            if let Some(script_failure) = cleaned.strip_prefix("VIBEBOX_SCRIPT_ERROR:") {
+                let failure = script_failure.trim();
+                if !failure.is_empty() {
+                    tracing::error!(script_failure = %failure, "[vm] script reported failure");
+                    broadcast_status(
+                        &clients,
+                        &latest_status,
+                        &format!("{STATUS_VM_ERROR_PREFIX} {failure}"),
+                    );
+                }
+            }
             if let Some(pos) = cleaned.find("VIBEBOX_IPV4=") {
                 let ip_raw = &cleaned[(pos + "VIBEBOX_IPV4=".len())..];
                 let ip = extract_ipv4(ip_raw).unwrap_or_default();
@@ -550,7 +587,7 @@ fn spawn_manager_io(
                     && cfg.vm_ipv4.as_deref() != Some(ip.as_str())
                 {
                     cfg.vm_ipv4 = Some(ip.clone());
-                    let _ = write_instance_config(&instance_path, &cfg);
+                    let _ = write_instance_config(&project_dir, &cfg);
                 }
             }
         }
@@ -573,12 +610,12 @@ enum ManagerEvent {
 }
 
 struct ManagerOptions {
-    ensure_signed: bool,
     detach: bool,
     prepare_vm: bool,
 }
 
 trait VmExecutor {
+    #[allow(clippy::too_many_arguments)]
     fn run_vm(
         &self,
         args: vm::VmArg,
@@ -586,8 +623,10 @@ trait VmExecutor {
         extra_shares: Vec<DirectoryShare>,
         config: Arc<Mutex<InstanceConfig>>,
         instance_dir: PathBuf,
+        clients: ClientStreams,
+        latest_status: SharedStatus,
         vm_input_tx: Arc<Mutex<Option<mpsc::Sender<VmInput>>>>,
-    ) -> Result<(), Box<dyn std::error::Error>>;
+    ) -> Result<()>;
 }
 
 #[cfg_attr(feature = "mock-vm", allow(dead_code))]
@@ -600,15 +639,22 @@ impl VmExecutor for RealVmExecutor {
         extra_login_actions: Vec<LoginAction>,
         extra_shares: Vec<DirectoryShare>,
         config: Arc<Mutex<InstanceConfig>>,
-        instance_dir: PathBuf,
+        project_dir: PathBuf,
+        clients: ClientStreams,
+        latest_status: SharedStatus,
         vm_input_tx: Arc<Mutex<Option<mpsc::Sender<VmInput>>>>,
-    ) -> Result<(), Box<dyn std::error::Error>> {
+    ) -> Result<()> {
+        let status_callback = |status: &str| {
+            broadcast_status(&clients, &latest_status, status);
+        };
         vm::run_with_args_and_extras(
             args,
             |output_monitor, vm_output_fd, vm_input_fd| {
                 let io_ctx = spawn_manager_io(
                     config.clone(),
-                    instance_dir.clone(),
+                    project_dir.clone(),
+                    clients.clone(),
+                    latest_status.clone(),
                     output_monitor,
                     vm_output_fd,
                     vm_input_fd,
@@ -618,6 +664,7 @@ impl VmExecutor for RealVmExecutor {
             },
             extra_login_actions,
             extra_shares,
+            Some(&status_callback),
         )
     }
 }
@@ -634,8 +681,10 @@ impl VmExecutor for MockVmExecutor {
         _extra_shares: Vec<DirectoryShare>,
         _config: Arc<Mutex<InstanceConfig>>,
         _instance_dir: PathBuf,
+        _clients: ClientStreams,
+        _latest_status: SharedStatus,
         vm_input_tx: Arc<Mutex<Option<mpsc::Sender<VmInput>>>>,
-    ) -> Result<(), Box<dyn std::error::Error>> {
+    ) -> Result<()> {
         let (tx, rx) = mpsc::channel::<VmInput>();
         *vm_input_tx.lock().unwrap() = Some(tx);
         tracing::info!("mock vm executor running");
@@ -661,24 +710,14 @@ fn run_manager_with(
     auto_shutdown_ms: u64,
     executor: &dyn VmExecutor,
     options: ManagerOptions,
-) -> Result<(), Box<dyn std::error::Error>> {
+) -> Result<()> {
-    if options.ensure_signed {
-        let _had_skip = env::var("VIBEBOX_SKIP_CODESIGN").ok();
-        unsafe {
-            env::remove_var("VIBEBOX_SKIP_CODESIGN");
-        }
-        vm::ensure_signed();
-        unsafe {
-            env::set_var("VIBEBOX_SKIP_CODESIGN", "1");
-        }
-    }
     if options.detach {
         detach_from_terminal();
     }
 
     let project_name = project_root
         .file_name()
-        .ok_or("Project directory has no name")?
+        .ok_or_else(|| anyhow!("Project directory has no name"))?
         .to_string_lossy()
         .into_owned();
     let instance_dir = ensure_instance_dir(project_root)?;
@@ -686,16 +725,16 @@ fn run_manager_with(
         let _ = ensure_ssh_keypair(&instance_dir)?;
     }
 
-    let mut config = load_or_create_instance_config(&instance_dir)?;
+    let mut config = load_or_create_instance_config(project_root)?;
     if config.vm_ipv4.is_some() {
         config.vm_ipv4 = None;
-        write_instance_config(&instance_dir.join(INSTANCE_FILENAME), &config)?;
+        write_instance_config(project_root, &config)?;
     }
     let config = Arc::new(Mutex::new(config));
     let ssh_user = config
         .lock()
-        .map(|cfg| cfg.ssh_user_display())
+        .map(|cfg| cfg.ssh_user.clone())
-        .unwrap_or_else(|_| DEFAULT_SSH_USER.to_string());
+        .map_err(|_| anyhow!("failed to acquire ssh user display"))?;
     if !args.no_default_mounts {
         inject_project_mount(&mut args.mounts, project_root, &ssh_user, &project_name);
     }
@@ -703,11 +742,10 @@ fn run_manager_with(
 
     let project_guest_dir = format!("{PROJECT_GUEST_BASE}/{project_name}");
     let ssh_guest_dir = format!("/root/{}", GLOBAL_DIR_NAME);
-    let extra_shares = vec![DirectoryShare::new(
+    let extra_shares = vec![
-        instance_dir.clone(),
+        DirectoryShare::new(instance_dir.clone(), ssh_guest_dir.clone().into(), true)
-        ssh_guest_dir.clone().into(),
+            .map_err(|err| anyhow!(err.to_string()))?,
-        true,
+    ];
-    )?];
     let extra_login_actions = build_ssh_login_actions(
         &config,
         &project_name,
@@ -730,17 +768,43 @@ fn run_manager_with(
     let _ = fs::set_permissions(&socket_path, fs::Permissions::from_mode(0o600));
     tracing::info!(path = %socket_path.display(), "vm manager socket bound");
 
+    let clients: ClientStreams = Arc::new(Mutex::new(Vec::new()));
+    let latest_status: SharedStatus = Arc::new(Mutex::new(String::new()));
     let (event_tx, event_rx) = mpsc::channel::<ManagerEvent>();
     let event_tx_accept = event_tx.clone();
+    let clients_accept = clients.clone();
+    let latest_status_accept = latest_status.clone();
     thread::spawn(move || {
         for stream in listener.incoming() {
             match stream {
                 Ok(stream) => {
+                    let mut client_fd: Option<std::os::fd::RawFd> = None;
+                    let latest_status_snapshot = latest_status_accept
+                        .lock()
+                        .ok()
+                        .map(|status| status.clone());
+                    if let Ok(writer) = stream.try_clone() {
+                        let writer_fd = writer.as_raw_fd();
+                        if let Ok(mut connected) = clients_accept.lock() {
+                            connected.push(writer);
+                            client_fd = Some(writer_fd);
+                            if let Some(last) = connected.last_mut()
+                                && let Some(status) = latest_status_snapshot.as_deref()
+                                && !status.is_empty()
+                            {
+                                let _ = send_status_line(last, status);
+                            }
+                        }
+                    }
                     let event_tx_conn = event_tx_accept.clone();
+                    let clients_conn = clients_accept.clone();
                     thread::spawn(move || {
                         let pid = read_client_pid(&stream);
                         let _ = event_tx_conn.send(ManagerEvent::Inc(pid));
                         wait_for_disconnect(stream);
+                        if let Some(fd) = client_fd {
+                            remove_client(&clients_conn, fd);
+                        }
                         let _ = event_tx_conn.send(ManagerEvent::Dec(pid));
                     });
                 }
@@ -760,34 +824,43 @@ fn run_manager_with(
         extra_login_actions,
         extra_shares,
         config.clone(),
-        instance_dir.clone(),
+        project_root.to_path_buf(),
+        clients.clone(),
+        latest_status.clone(),
         vm_input_tx.clone(),
     );
     tracing::info!("vm manager vm run completed");
     let vm_err = vm_result.err().map(|e| e.to_string());
+    if let Some(err) = &vm_err {
+        broadcast_status(
+            &clients,
+            &latest_status,
+            &format!("{STATUS_VM_ERROR_PREFIX} {err}"),
+        );
+    }
     let _ = event_tx.send(ManagerEvent::VmExited(vm_err.clone()));
-    let event_loop_result: Result<(), String> = event_loop_handle
+    let event_loop_result = event_loop_handle
         .join()
-        .unwrap_or_else(|_| Err("vm manager event loop panicked".into()))
+        .unwrap_or_else(|_| Err(Error::msg("vm manager event loop panicked")))
         .map_err(|err| err.to_string());
     let _ = fs::remove_file(&socket_path);
     if let Err(err) = &event_loop_result {
         tracing::error!(error = %err, "vm manager exiting due to event loop error");
-        return Err(err.to_string().into());
+        bail!(err.to_string());
     }
     if let Some(err) = vm_err {
         tracing::error!(error = %err, "vm manager exiting due to vm error");
-        return Err(err.into());
+        bail!(err);
     }
     tracing::info!("vm manager exiting");
-    Ok(event_loop_result?)
+    event_loop_result.map_err(Error::msg)
 }
 
 fn manager_event_loop(
     event_rx: mpsc::Receiver<ManagerEvent>,
     vm_input_tx: Arc<Mutex<Option<mpsc::Sender<VmInput>>>>,
     auto_shutdown_ms: u64,
-) -> Result<(), String> {
+) -> Result<()> {
     let mut ref_count: usize = 0;
     let mut shutdown_deadline: Option<Instant> = None;
     let mut shutdown_sent = false;
@@ -897,7 +970,7 @@ fn manager_event_loop(
 #[cfg(test)]
 mod tests {
     use super::*;
-    use std::{sync::mpsc, thread, time::Duration};
+    use std::{fs, sync::mpsc, thread, time::Duration};
 
     #[test]
     fn manager_powers_off_after_grace_when_no_refs() {
@@ -979,4 +1052,26 @@ mod tests {
         let _ = event_tx.send(ManagerEvent::VmExited(None));
         let _ = manager_thread.join();
     }
+
+    #[test]
+    fn lock_is_not_stale_when_owner_pid_is_alive() {
+        let temp = tempfile::Builder::new()
+            .prefix("vb")
+            .tempdir_in("/tmp")
+            .expect("tempdir");
+        let lock_path = temp.path().join("vm.lock");
+        fs::write(&lock_path, format!("pid={}\n", std::process::id())).expect("write lock");
+        assert!(!is_lock_stale(&lock_path));
+    }
+
+    #[test]
+    fn lock_is_stale_when_owner_pid_is_missing() {
+        let temp = tempfile::Builder::new()
+            .prefix("vb")
+            .tempdir_in("/tmp")
+            .expect("tempdir");
+        let lock_path = temp.path().join("vm.lock");
+        fs::write(&lock_path, "pid=999999\n").expect("write lock");
+        assert!(is_lock_stale(&lock_path));
+    }
 }