merge: integrate main into focus and blur PR

crates/bsk-cli/src/cli/atomic_output.rs

@@ -0,0 +1,75 @@
+//! Atomic visibility boundary for files received into a same-directory temp.
+
+use std::path::Path;
+
+pub fn commit(temp: &Path, out: &Path, overwrite: bool) -> std::io::Result<()> {
+    if !overwrite {
+        std::fs::hard_link(temp, out)?;
+        std::fs::remove_file(temp)?;
+        return Ok(());
+    }
+    replace(temp, out)
+}
+
+#[cfg(unix)]
+fn replace(temp: &Path, out: &Path) -> std::io::Result<()> {
+    // POSIX rename replaces an existing non-directory destination atomically.
+    std::fs::rename(temp, out)
+}
+
+#[cfg(windows)]
+fn replace(temp: &Path, out: &Path) -> std::io::Result<()> {
+    use std::os::windows::ffi::OsStrExt;
+    use windows_sys::Win32::Storage::FileSystem::{
+        MOVEFILE_REPLACE_EXISTING, MOVEFILE_WRITE_THROUGH, MoveFileExW,
+    };
+
+    let from: Vec<u16> = temp.as_os_str().encode_wide().chain(Some(0)).collect();
+    let to: Vec<u16> = out.as_os_str().encode_wide().chain(Some(0)).collect();
+    // SAFETY: both buffers are NUL-terminated and remain alive for the call.
+    let ok = unsafe {
+        MoveFileExW(
+            from.as_ptr(),
+            to.as_ptr(),
+            MOVEFILE_REPLACE_EXISTING | MOVEFILE_WRITE_THROUGH,
+        )
+    };
+    if ok == 0 {
+        Err(std::io::Error::last_os_error())
+    } else {
+        Ok(())
+    }
+}
+
+#[cfg(not(any(unix, windows)))]
+fn replace(temp: &Path, out: &Path) -> std::io::Result<()> {
+    std::fs::rename(temp, out)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn no_replace_never_overwrites_existing_output() {
+        let dir = tempfile::tempdir().unwrap();
+        let temp = dir.path().join("new.part");
+        let out = dir.path().join("out.bin");
+        std::fs::write(&temp, b"new").unwrap();
+        std::fs::write(&out, b"old").unwrap();
+        assert!(commit(&temp, &out, false).is_err());
+        assert_eq!(std::fs::read(&out).unwrap(), b"old");
+    }
+
+    #[test]
+    fn overwrite_replaces_existing_output_without_predelete() {
+        let dir = tempfile::tempdir().unwrap();
+        let temp = dir.path().join("new.part");
+        let out = dir.path().join("out.bin");
+        std::fs::write(&temp, b"new").unwrap();
+        std::fs::write(&out, b"old").unwrap();
+        commit(&temp, &out, true).unwrap();
+        assert_eq!(std::fs::read(&out).unwrap(), b"new");
+        assert!(!temp.exists());
+    }
+}

crates/bsk-cli/src/cli/business_rpc.rs

@@ -73,6 +73,14 @@ where
     P: Serialize + Send + 'static,
     R: DeserializeOwned + Send + 'static,
 {
+    let parent_cancel = stdin_cancellation();
+    if parent_cancel.as_ref().is_some_and(|rx| *rx.borrow()) {
+        return Err(CliError::from_rpc(RpcError {
+            code: ErrorCode::Cancelled,
+            message: "parent closed the cancellation pipe".into(),
+            data: Some(serde_json::json!({"reason": "cancelled_before_dispatch"})),
+        }));
+    }
     let rpc_id: RpcId = format!("{}-{}", rpc_id_prefix, random_short_id());
     let mut client = IpcClient::connect(sock.clone()).await?;
     let rpc_id_for_call = rpc_id.clone();
@@ -88,16 +96,16 @@ where
     let outcome = tokio::select! {
         biased;
         res = &mut main_fut => res?,
-        sig = wait_for_sigint() => {
-            sig.context("install SIGINT handler").map_err(CliError::Local)?;
-            debug!(rpc_id = %rpc_id_for_cancel, "SIGINT: forwarding cancel to daemon");
+        sig = wait_for_cancel(parent_cancel) => {
+            sig.context("wait for cancellation").map_err(CliError::Local)?;
+            debug!(rpc_id = %rpc_id_for_cancel, "forwarding cancel to daemon");
             let _ = send_cancel(&sock, &rpc_id_for_cancel).await;
             match tokio::time::timeout(CANCEL_RESPONSE_GRACE, &mut main_fut).await {
                 Ok(res) => res?,
                 Err(_) => {
                     debug!(
                         rpc_id = %rpc_id_for_cancel,
-                        "SIGINT: cancel grace elapsed; synthesising cancelled error"
+                        "cancel grace elapsed; synthesising cancelled error"
                     );
                     return Err(CliError::from_rpc(RpcError {
                         code: ErrorCode::Cancelled,
@@ -112,6 +120,46 @@ where
     outcome.map_err(CliError::from_rpc)
 }
 
+/// Release caller-owned staging even after cancellation. This deliberately
+/// exposes only transfer.release, not a general bypass for business RPCs.
+/// The entire batch, including connection setup, shares one bounded budget.
+pub fn release_transfers<'a>(
+    sock: &Path,
+    transfer_ids: impl IntoIterator<Item = &'a str>,
+) -> Result<(), CliError> {
+    use bsk_protocol::tools::{TransferIdParams, TransferReleaseResult};
+    const CLEANUP_TIMEOUT: Duration = Duration::from_secs(5);
+    let ids: Vec<_> = transfer_ids.into_iter().collect();
+    if ids.is_empty() {
+        return Ok(());
+    }
+    let rt = tokio::runtime::Builder::new_current_thread()
+        .enable_all()
+        .build()
+        .context("build runtime for transfer cleanup")?;
+    rt.block_on(async {
+        tokio::time::timeout(CLEANUP_TIMEOUT, async {
+            let mut client = IpcClient::connect(sock).await?;
+            for id in ids {
+                client
+                    .call::<_, TransferReleaseResult>(
+                        "transfer-release",
+                        Method::TransferRelease,
+                        Some(TransferIdParams {
+                            transfer_id: id.to_string(),
+                        }),
+                        CLEANUP_TIMEOUT,
+                    )
+                    .await?
+                    .map_err(CliError::from_rpc)?;
+            }
+            Ok::<_, CliError>(())
+        })
+        .await
+        .context("transfer cleanup exceeded its total time budget")?
+    })
+}
+
 /// Send a `cancel { rpc_id }` frame over a fresh connection so it
 /// lands on the daemon while the original call is still parked.
 ///
@@ -146,8 +194,57 @@ fn random_short_id() -> String {
     bytes.iter().map(|b| format!("{b:02x}")).collect()
 }
 
-/// Wait for either Ctrl-C or process SIGTERM. Returning here means
-/// the user has asked us to stop.
+/// Opt-in control pipe for parent processes that cannot send Ctrl-C (Node on
+/// Windows). One reader survives the short-lived runtimes used by multi-RPC
+/// commands such as upload. Ordinary CLI stdin and terminal Ctrl-C are unchanged.
+fn stdin_cancellation() -> Option<tokio::sync::watch::Receiver<bool>> {
+    if std::env::var("BSK_CANCEL_ON_STDIN_CLOSE").as_deref() != Ok("1") {
+        return None;
+    }
+    static CANCEL: std::sync::OnceLock<tokio::sync::watch::Receiver<bool>> =
+        std::sync::OnceLock::new();
+    Some(
+        CANCEL
+            .get_or_init(|| {
+                let (tx, rx) = tokio::sync::watch::channel(false);
+                std::thread::spawn(move || {
+                    use std::io::Read;
+                    let mut stdin = std::io::stdin().lock();
+                    let mut buf = [0; 64];
+                    loop {
+                        match stdin.read(&mut buf) {
+                            Ok(0) => break,
+                            Ok(_) => {}
+                            Err(err) if err.kind() == std::io::ErrorKind::Interrupted => continue,
+                            Err(_) => break,
+                        }
+                    }
+                    let _ = tx.send(true);
+                });
+                rx
+            })
+            .clone(),
+    )
+}
+
+async fn wait_for_cancel(
+    parent_cancel: Option<tokio::sync::watch::Receiver<bool>>,
+) -> anyhow::Result<()> {
+    let parent_closed = async move {
+        match parent_cancel {
+            Some(mut rx) => {
+                let _ = rx.wait_for(|closed| *closed).await;
+            }
+            None => std::future::pending::<()>().await,
+        }
+    };
+    tokio::select! {
+        result = wait_for_sigint() => result,
+        _ = parent_closed => Ok(()),
+    }
+}
+
+/// Wait for terminal Ctrl-C.
 async fn wait_for_sigint() -> anyhow::Result<()> {
     tokio::signal::ctrl_c().await.context("listen for SIGINT")?;
     Ok(())

crates/bsk-cli/src/cli/doctor.rs

@@ -351,10 +351,10 @@ fn check_version_compatible(status: Option<&StatusResult>) -> CheckResult {
     }
 }
 
-/// `bsk doctor` check: every connected browser should use the same
-/// protocol version as the daemon. Minor protocol drift is accepted by
-/// the daemon but flagged here so the user can see who needs updating
-/// (M10.4).
+/// `bsk doctor` check: connected browsers should speak a protocol the
+/// daemon accepts. A different protocol string is still a live
+/// connection — report Ok so agents keep working. The detail is an
+/// upgrade reminder, not a blocker.
 ///
 /// Review M2 (round-1 minor): when no browsers are connected, the
 /// check has nothing to compare against, so it now reports
@@ -396,13 +396,12 @@ fn check_browsers_protocol_compatible(status: Option<&StatusResult>) -> CheckRes
         })
         .collect::<Vec<_>>()
         .join(", ");
-    CheckResult::fail(
+    CheckResult::ok(
         name,
         format!(
-            "{} browser(s) have protocol minor drift from the daemon: {stale}",
+            "{} browser(s) report a different protocol version (still usable — continue, and upgrade soon): {stale}",
             status.version_skew_browsers.len()
         ),
-        "upgrade the browser-skill extension or bsk CLI so both sides use the same protocol version",
     )
 }
 
@@ -581,7 +580,7 @@ mod m2_tests {
     }
 
     #[test]
-    fn browsers_check_reports_fail_when_skew_present() {
+    fn browsers_check_reports_ok_when_skew_present() {
         let status = fake_status(
             vec![BrowserStatusEntry {
                 instance_id: "alpha".into(),
@@ -605,9 +604,17 @@ mod m2_tests {
             }],
         );
         let check = check_browsers_protocol_compatible(Some(&status));
-        assert_eq!(check.status, CheckStatus::Fail);
+        assert_eq!(check.status, CheckStatus::Ok);
         assert!(check.detail.contains("alpha"));
-        assert!(check.hint.is_some());
+        assert!(
+            check
+                .detail
+                .contains("still usable — continue, and upgrade soon")
+        );
+        assert!(
+            check.hint.is_none(),
+            "a protocol-version note must not hint to stop"
+        );
     }
 
     #[test]
@@ -635,7 +642,7 @@ mod m2_tests {
             }],
         );
         let check = check_browsers_protocol_compatible(Some(&status));
-        assert_eq!(check.status, CheckStatus::Fail);
+        assert_eq!(check.status, CheckStatus::Ok);
         assert!(
             check
                 .detail

crates/bsk-cli/src/cli/download.rs

@@ -0,0 +1,140 @@
+//! `bsk download` — capture one browser download and atomically commit it.
+
+use std::fs::OpenOptions;
+use std::io::Write;
+use std::path::{Path, PathBuf};
+use std::time::Duration;
+
+use anyhow::Context;
+use base64::Engine;
+use bsk_protocol::Method;
+use bsk_protocol::tools::{
+    DownloadParams, DownloadResult, TransferChunkParams, TransferChunkResult,
+};
+use clap::Args;
+use uuid::Uuid;
+
+use crate::cli::atomic_output;
+use crate::cli::ensure_daemon::ensure_daemon;
+use crate::cli::error::{CliError, Format};
+use crate::cli::interaction::split_target;
+use crate::cli::navigate::parse_timeout_ms;
+
+#[derive(Debug, Clone, Args)]
+pub struct DownloadArgs {
+    /// Snapshot ref (`@e3`) or CSS selector for the download trigger.
+    pub target: Option<String>,
+    #[arg(long = "ref")]
+    pub ref_: Option<String>,
+    #[arg(long = "selector")]
+    pub selector: Option<String>,
+    #[arg(long)]
+    pub out: PathBuf,
+    #[arg(long)]
+    pub session: String,
+    #[arg(long = "tab-id")]
+    pub tab_id: Option<i64>,
+    #[arg(long, default_value = "2m", value_parser = parse_timeout_ms)]
+    pub timeout: u32,
+    #[arg(long)]
+    pub overwrite: bool,
+}
+
+pub fn dispatch(args: DownloadArgs, format: Format) -> Result<(), CliError> {
+    if args.out.exists() && !args.overwrite {
+        return Err(CliError::Local(anyhow::anyhow!(
+            "output already exists (pass --overwrite to replace it): {}",
+            args.out.display()
+        )));
+    }
+    let info = ensure_daemon().context("ensure daemon is running")?;
+    let (ref_, selector) = split_target(args.target, args.ref_, args.selector)?;
+    let params = DownloadParams {
+        session_id: args.session,
+        ref_,
+        selector,
+        tab_id: args.tab_id,
+        timeout_ms: Some(args.timeout),
+        browser_relative_dir: None,
+        max_byte_size: None,
+    };
+    let reply: DownloadResult = crate::cli::business_rpc::call(
+        info.sock_path.clone(),
+        "download",
+        Method::ToolDownload,
+        Some(params),
+        ipc_timeout(args.timeout),
+    )?;
+    let transfer_id = reply.transfer_id.clone().ok_or_else(|| {
+        CliError::Local(anyhow::anyhow!("daemon returned no download transfer id"))
+    })?;
+    let write_result = write_transfer(&info.sock_path, &transfer_id, &args.out, args.overwrite);
+    let _ = crate::cli::business_rpc::release_transfers(&info.sock_path, [transfer_id.as_str()]);
+    write_result?;
+    match format {
+        Format::Json => {
+            let mut value = serde_json::to_value(&reply).unwrap();
+            value["path"] = serde_json::json!(args.out.to_string_lossy());
+            if let Some(obj) = value.as_object_mut() {
+                obj.remove("transfer_id");
+                obj.remove("browser_path");
+            }
+            println!("{}", serde_json::to_string_pretty(&value).unwrap());
+        }
+        Format::Human => println!("{}", args.out.display()),
+    }
+    Ok(())
+}
+
+fn write_transfer(sock: &Path, id: &str, out: &Path, overwrite: bool) -> Result<(), CliError> {
+    let parent = out
+        .parent()
+        .filter(|p| !p.as_os_str().is_empty())
+        .unwrap_or(Path::new("."));
+    let temp = parent.join(format!(".bsk-download-{}.part", Uuid::new_v4().simple()));
+    let result = (|| {
+        let mut file = OpenOptions::new()
+            .write(true)
+            .create_new(true)
+            .open(&temp)
+            .with_context(|| format!("create temporary download {}", temp.display()))
+            .map_err(CliError::Local)?;
+        let mut offset = 0u64;
+        loop {
+            let chunk: TransferChunkResult = crate::cli::business_rpc::call(
+                sock.to_path_buf(),
+                "transfer-read",
+                Method::TransferRead,
+                Some(TransferChunkParams {
+                    transfer_id: id.to_string(),
+                    offset,
+                    data_base64: String::new(),
+                }),
+                Duration::from_secs(30),
+            )?;
+            let encoded = chunk.data_base64.as_deref().unwrap_or("");
+            let bytes = base64::engine::general_purpose::STANDARD
+                .decode(encoded)
+                .map_err(|e| CliError::Local(anyhow::anyhow!("decode download chunk: {e}")))?;
+            file.write_all(&bytes)
+                .map_err(|e| CliError::Local(e.into()))?;
+            offset = chunk.next_offset;
+            if chunk.eof {
+                break;
+            }
+        }
+        file.sync_all().map_err(|e| CliError::Local(e.into()))?;
+        drop(file);
+        atomic_output::commit(&temp, out, overwrite)
+            .with_context(|| format!("atomically commit download to {}", out.display()))
+            .map_err(CliError::Local)
+    })();
+    if result.is_err() {
+        let _ = std::fs::remove_file(&temp);
+    }
+    result
+}
+
+fn ipc_timeout(timeout_ms: u32) -> Duration {
+    Duration::from_millis(u64::from(timeout_ms) + 5_000)
+}

crates/bsk-cli/src/cli/interaction.rs

@@ -73,7 +73,7 @@ pub(crate) fn parse_modifiers(input: &str) -> Result<Vec<KeyModifier>, String> {
     Ok(out)
 }
 
-fn split_target(
+pub(crate) fn split_target(
     positional: Option<String>,
     explicit_ref: Option<String>,
     explicit_selector: Option<String>,

crates/bsk-cli/src/cli/mod.rs

@@ -2,6 +2,7 @@
 
 use std::time::Duration;
 
+mod atomic_output;
 pub mod browser_wait;
 pub mod browsers;
 pub mod business_rpc;
@@ -9,6 +10,7 @@ pub mod console;
 pub mod daemon;
 pub mod dialogs;
 pub mod doctor;
+pub mod download;
 pub mod emulate;
 pub mod ensure_daemon;
 pub mod error;
@@ -22,6 +24,7 @@ pub mod navigate;
 pub mod network;
 pub mod observe;
 pub mod record;
+pub mod record_recovery;
 pub mod record_state;
 pub mod render_error;
 pub mod screenshot;
@@ -30,6 +33,7 @@ pub mod snapshot;
 pub mod status;
 pub mod tab;
 pub mod update;
+pub mod upload;
 pub mod waits;
 pub mod window;
 
@@ -37,6 +41,7 @@ use clap::{Args, Parser, Subcommand};
 
 use crate::cli::console::ConsoleArgs;
 use crate::cli::daemon::DaemonCmd;
+use crate::cli::download::DownloadArgs;
 use crate::cli::emulate::EmulateArgs;
 use crate::cli::evaluate::EvaluateArgs;
 use crate::cli::get_html::GetHtmlArgs;
@@ -54,6 +59,7 @@ use crate::cli::session::SessionCmd;
 use crate::cli::snapshot::SnapshotArgs;
 use crate::cli::tab::TabCmd;
 use crate::cli::update::UpdateArgs;
+use crate::cli::upload::UploadArgs;
 use crate::cli::waits::{WaitForNavigationArgs, WaitMsArgs};
 use crate::cli::window::WindowCmd;
 
@@ -184,6 +190,12 @@ pub enum Command {
     /// Set `<select>` option values by `value` attribute.
     Select(SelectArgs),
 
+    /// Upload files through a page file input or explicit drop target.
+    Upload(UploadArgs),
+
+    /// Capture one browser download and write it to a local path.
+    Download(DownloadArgs),
+
     /// Evaluate a JavaScript expression inside the Agent Window.
     Evaluate(EvaluateArgs),
 

crates/bsk-cli/src/cli/observe.rs

@@ -33,6 +33,12 @@ pub struct ObserveArgs {
     /// Include conditional surface probe diagnostics in JSON output.
     #[arg(long = "debug-surfaces")]
     pub debug_surfaces: bool,
+
+    /// Actively hover page controls to reveal hover-only menus and tooltips.
+    /// Costs seconds of wall clock and touches the live page, so it is off
+    /// unless a static observation looks like it is missing hover content.
+    #[arg(long = "probe-hover")]
+    pub probe_hover: bool,
 }
 
 pub fn dispatch(args: ObserveArgs, format: Format) -> Result<(), CliError> {
@@ -47,6 +53,7 @@ fn run(sock: PathBuf, args: ObserveArgs, format: Format) -> Result<(), CliError>
         max_depth: args.max_depth,
         max_tokens: args.max_tokens,
         debug_surfaces: args.debug_surfaces,
+        probe_hover: args.probe_hover,
     };
     let reply: ObserveResult = call(sock, params)?;
     match format {

crates/bsk-cli/src/cli/record.rs

@@ -1,23 +1,30 @@
 //! `bsk record start|stop` — capture user actions in the Agent Window.
 
-use std::fs;
-use std::path::PathBuf;
+use std::path::{Path, PathBuf};
 use std::time::Duration;
 
 use anyhow::Context;
-use bsk_protocol::Method;
 use bsk_protocol::tools::{
     RecordAwaitParams, RecordAwaitResult, RecordStartParams, RecordStartResult, RecordStopParams,
-    RecordStopResult, Trace,
+    RecordStopResult, RecordedTrace, TRACE_VERSION_V3,
 };
+use bsk_protocol::{ErrorCode, Method};
 use clap::{Args, Subcommand};
 
 use crate::cli::TOOL_IPC_TIMEOUT;
+
+mod export;
+
 use crate::cli::business_rpc;
 use crate::cli::ensure_daemon::ensure_daemon;
 use crate::cli::error::{CliError, Format};
+use crate::cli::record_recovery;
 use crate::cli::record_state;
 use crate::cli::session::{SessionStartOptions, start_session, stop_session};
+use export::{
+    ExportMeta, export_with_recovery, states_dir_for_output, trace_json_path,
+    validate_record_output,
+};
 
 /// Max wait for the user to click 结束 in the browser (24 hours).
 const RECORD_AWAIT_TIMEOUT_MS: u32 = 86_400_000;
@@ -50,7 +57,8 @@ pub struct RecordStartArgs {
     pub tab_id: Option<i64>,
 
     /// Navigate to this http(s) URL before recording. When omitted, defaults
-    /// to `https://example.com/`.
+    /// to `https://example.com/`. If that page does not load, pass `--url`
+    /// with a site you can open in this browser.
     #[arg(long)]
     pub url: Option<String>,
 
@@ -58,15 +66,25 @@ pub struct RecordStartArgs {
     #[arg(long)]
     pub purpose: Option<String>,
 
-    /// Output path for the trace JSON (default `./trace.json`).
-    #[arg(long, default_value = "trace.json")]
+    /// Max VOM tokens per page observation file (default 3000).
+    #[arg(long = "max-page-tokens")]
+    pub max_page_tokens: Option<u32>,
+
+    /// Redact all form values in page observations (`[filled]` only).
+    #[arg(long = "redact-values")]
+    pub redact_values: bool,
+
+    /// Output directory for the trace bundle (default `./trace`).
+    /// Writes `<dir>/trace.json` and `<dir>/states/*.txt`.
+    #[arg(long, default_value = "trace")]
     pub output: PathBuf,
 }
 
 #[derive(Debug, Clone, Args)]
 pub struct RecordStopArgs {
-    /// Output path for the trace JSON (default `./trace.json`).
-    #[arg(long, default_value = "trace.json")]
+    /// Output directory for the trace bundle (default `./trace`).
+    /// Writes `<dir>/trace.json` and `<dir>/states/*.txt`.
+    #[arg(long, default_value = "trace")]
     pub output: PathBuf,
 }
 
@@ -78,11 +96,7 @@ pub fn dispatch(cmd: RecordCmd, format: Format) -> Result<(), CliError> {
 }
 
 fn dispatch_start(args: RecordStartArgs, format: Format) -> Result<(), CliError> {
-    if record_state::read().is_ok() {
-        return Err(CliError::Local(anyhow::anyhow!(
-            "a recording is already in progress; run `bsk record stop` first"
-        )));
-    }
+    prepare_record_start(&args.output)?;
 
     let info = ensure_daemon().context("ensure daemon is running")?;
     let session = start_session(
@@ -96,8 +110,12 @@ fn dispatch_start(args: RecordStartArgs, format: Format) -> Result<(), CliError>
     let start_params = RecordStartParams {
         session_id: session.session_id.clone(),
         tab_id: args.tab_id,
-        url: args.url,
+        url: args.url.clone(),
         purpose: args.purpose.clone(),
+        max_page_tokens: args.max_page_tokens,
+        redact_values: Some(args.redact_values),
+        trace_version: Some(TRACE_VERSION_V3),
+        supports_tab_switch_steps: Some(true),
     };
     let start_result = business_rpc::call::<RecordStartParams, RecordStartResult>(
         info.sock_path.clone(),
@@ -111,7 +129,7 @@ fn dispatch_start(args: RecordStartArgs, format: Format) -> Result<(), CliError>
         Ok(result) => result,
         Err(err) => {
             let _ = stop_session(info.sock_path, &session.session_id);
-            return Err(err);
+            return Err(annotate_default_start_page_error(err, args.url.as_deref()));
         }
     };
 
@@ -142,8 +160,8 @@ fn dispatch_start(args: RecordStartArgs, format: Format) -> Result<(), CliError>
     // Keep write/render inside a Result so `?` cannot skip session teardown.
     let run_result: Result<(), CliError> = match await_result {
         Ok(result) => (|| {
-            write_trace_file(&args.output, &result.trace)?;
-            render_finish(&result.trace, &args.output, format)
+            let exported = export_with_recovery(&args.output, &result.trace)?;
+            render_finish(&result.trace, &args.output, &exported, format)
         })(),
         Err(err) => Err(err),
     };
@@ -157,11 +175,12 @@ fn dispatch_start(args: RecordStartArgs, format: Format) -> Result<(), CliError>
 }
 
 fn dispatch_stop(args: RecordStopArgs, format: Format) -> Result<(), CliError> {
-    let state = record_state::read().map_err(CliError::Local)?;
-    let info = ensure_daemon().context("ensure daemon is running")?;
-    let session_id = state.session_id.clone();
+    validate_record_output(&args.output)?;
+
+    if let Ok(state) = record_state::read() {
+        let info = ensure_daemon().context("ensure daemon is running")?;
+        let session_id = state.session_id.clone();
 
-    let run_result: Result<(), CliError> = (|| {
         let params = RecordStopParams {
             session_id: session_id.clone(),
         };
@@ -172,16 +191,85 @@ fn dispatch_stop(args: RecordStopArgs, format: Format) -> Result<(), CliError> {
             Some(params),
             TOOL_IPC_TIMEOUT,
         )?;
-        write_trace_file(&args.output, &result.trace)?;
-        render_stop(&result, &args.output, format)
-    })();
 
-    let session_stop_result = stop_session(info.sock_path, &session_id);
-    record_state::clear();
+        let run_result: Result<(), CliError> = (|| {
+            let exported = export_with_recovery(&args.output, &result.trace)?;
+            render_stop(&result, &args.output, &exported, format)
+        })();
 
-    run_result?;
-    session_stop_result?;
-    Ok(())
+        let session_stop_result = stop_session(info.sock_path, &session_id);
+        record_state::clear();
+
+        run_result?;
+        session_stop_result?;
+        return Ok(());
+    }
+
+    let Some(trace) = record_recovery::load().map_err(CliError::Local)? else {
+        return Err(CliError::Local(anyhow::anyhow!(
+            "no recording in progress; run `bsk record start` first"
+        )));
+    };
+
+    let exported = export_with_recovery(&args.output, &trace)?;
+    render_finish(&trace, &args.output, &exported, format)
+}
+
+/// When `record start` omitted `--url` and the default example.com page
+/// never loaded, rewrite the RPC error so the CLI hint points at `--url`.
+fn annotate_default_start_page_error(err: CliError, user_url: Option<&str>) -> CliError {
+    if user_url.is_some() {
+        return err;
+    }
+    let CliError::Rpc {
+        code,
+        message,
+        data,
+        source,
+    } = err
+    else {
+        return err;
+    };
+    if !is_default_start_page_failure(code, &message) {
+        return CliError::Rpc {
+            code,
+            message,
+            data,
+            source,
+        };
+    }
+    CliError::Rpc {
+        code,
+        message,
+        data: Some(serde_json::json!({
+            "reason": crate::cli::render_error::reason::RECORD_START_PAGE_UNREACHABLE
+        })),
+        source,
+    }
+}
+
+fn is_default_start_page_failure(code: ErrorCode, message: &str) -> bool {
+    match code {
+        ErrorCode::CdpFailed | ErrorCode::Timeout => true,
+        ErrorCode::InvalidParams => {
+            message.contains("restricted URL") || message.contains("about:blank")
+        }
+        _ => false,
+    }
+}
+
+fn prepare_record_start(output: &Path) -> Result<(), CliError> {
+    if record_state::read().is_ok() {
+        return Err(CliError::Local(anyhow::anyhow!(
+            "a recording is already in progress; run `bsk record stop` first"
+        )));
+    }
+    if record_recovery::exists() {
+        return Err(CliError::Local(anyhow::anyhow!(
+            "a previous recording was not exported; run `bsk record stop --output <dir>` to recover it first"
+        )));
+    }
+    validate_record_output(output)
 }
 
 fn record_await_ipc_timeout(timeout_ms: u32) -> Duration {
@@ -190,39 +278,66 @@ fn record_await_ipc_timeout(timeout_ms: u32) -> Duration {
         .unwrap_or(Duration::from_secs(u64::from(timeout_ms / 1_000) + 15))
 }
 
-fn write_trace_file(output: &PathBuf, trace: &Trace) -> Result<(), CliError> {
-    let json = serde_json::to_string_pretty(trace)
-        .context("serialize trace JSON")
-        .map_err(CliError::Local)?;
-    if let Some(parent) = output.parent() {
-        if !parent.as_os_str().is_empty() {
-            fs::create_dir_all(parent)
-                .with_context(|| format!("create output directory {}", parent.display()))
-                .map_err(CliError::Local)?;
-        }
-    }
-    fs::write(output, format!("{json}\n"))
-        .with_context(|| format!("write trace to {}", output.display()))
-        .map_err(CliError::Local)?;
-    Ok(())
-}
-
-fn render_finish(trace: &Trace, output: &PathBuf, format: Format) -> Result<(), CliError> {
+fn render_finish(
+    trace: &RecordedTrace,
+    output_dir: &PathBuf,
+    exported: &ExportMeta,
+    format: Format,
+) -> Result<(), CliError> {
+    let trace_path = trace_json_path(output_dir);
     match format {
         Format::Json => {
+            let payload = match trace {
+                RecordedTrace::V3(t) => serde_json::json!({
+                    "output": output_dir,
+                    "trace_json": trace_path,
+                    "trace_version": exported.trace_version,
+                    "states_dir": exported.states_dir,
+                    "trace": t,
+                    "window_closed": true,
+                }),
+                RecordedTrace::V2(t) => serde_json::json!({
+                    "output": output_dir,
+                    "trace_json": trace_path,
+                    "trace_version": exported.trace_version,
+                    "states_dir": null,
+                    "trace": t,
+                    "window_closed": true,
+                }),
+            };
             println!(
                 "{}",
-                serde_json::to_string(&serde_json::json!({
-                    "output": output,
-                    "trace": trace,
-                    "window_closed": true,
-                }))
-                .map_err(|e| CliError::Local(anyhow::anyhow!(e)))?
+                serde_json::to_string(&payload).map_err(|e| CliError::Local(anyhow::anyhow!(e)))?
             );
         }
-        Format::Human => {
-            println!("saved {} steps to {}", trace.steps.len(), output.display());
-        }
+        Format::Human => match trace {
+            RecordedTrace::V3(t) => {
+                let states_dir = exported
+                    .states_dir
+                    .as_ref()
+                    .map(|p| p.display().to_string())
+                    .unwrap_or_else(|| states_dir_for_output(output_dir).display().to_string());
+                println!(
+                    "saved {} steps to {} and {} states to {}",
+                    t.steps.len(),
+                    trace_path.display(),
+                    t.states.len(),
+                    states_dir
+                );
+            }
+            RecordedTrace::V2(t) => {
+                if exported.v2_fallback {
+                    eprintln!(
+                        "note: extension returned trace v2 (no page observations); update the BrowserSkill extension for v3 bundles"
+                    );
+                }
+                println!(
+                    "saved {} steps to {} (trace v2)",
+                    t.steps.len(),
+                    trace_path.display()
+                );
+            }
+        },
     }
     Ok(())
 }
@@ -230,33 +345,89 @@ fn render_finish(trace: &Trace, output: &PathBuf, format: Format) -> Result<(),
 fn render_stop(
     result: &RecordStopResult,
     output: &PathBuf,
+    exported: &ExportMeta,
     format: Format,
 ) -> Result<(), CliError> {
-    render_finish(&result.trace, output, format)
+    render_finish(&result.trace, output, exported, format)
 }
 
 #[cfg(test)]
 mod tests {
     use super::*;
 
-    #[test]
-    fn default_output_is_trace_json() {
-        let args = RecordStopArgs {
-            output: PathBuf::from("trace.json"),
-        };
-        assert_eq!(args.output, PathBuf::from("trace.json"));
+    fn with_temp_home<F: FnOnce()>(f: F) {
+        let _lock = record_recovery::test_env_lock();
+        let tmp = tempfile::tempdir().unwrap();
+        unsafe {
+            std::env::set_var(crate::daemon::paths::BSK_HOME_ENV, tmp.path());
+        }
+        f();
+        record_recovery::clear();
+        record_state::clear();
+        unsafe {
+            std::env::remove_var(crate::daemon::paths::BSK_HOME_ENV);
+        }
     }
 
     #[test]
-    fn start_args_default_output_is_trace_json() {
+    fn prepare_record_start_rejects_existing_trace_json_file() {
+        with_temp_home(|| {
+            let dir = tempfile::tempdir().unwrap();
+            let output = dir.path().join("trace.json");
+            std::fs::write(&output, "{}\n").unwrap();
+
+            let err = prepare_record_start(&output).unwrap_err();
+            let msg = err.to_string();
+            assert!(
+                msg.contains("JSON file") || msg.contains("json file"),
+                "{msg}"
+            );
+            assert!(
+                msg.contains("states/") || msg.contains("--output trace"),
+                "{msg}"
+            );
+        });
+    }
+
+    #[test]
+    fn prepare_record_start_rejects_unexported_recovery() {
+        with_temp_home(|| {
+            crate::daemon::paths::ensure_bsk_home().unwrap();
+            std::fs::write(
+                crate::daemon::paths::record_recovery_path().unwrap(),
+                "{}\n",
+            )
+            .unwrap();
+
+            let err = prepare_record_start(Path::new("trace")).unwrap_err();
+            let msg = err.to_string();
+            assert!(
+                msg.contains("not exported") || msg.contains("recover"),
+                "{msg}"
+            );
+        });
+    }
+
+    #[test]
+    fn default_output_is_trace_dir() {
+        let args = RecordStopArgs {
+            output: PathBuf::from("trace"),
+        };
+        assert_eq!(args.output, PathBuf::from("trace"));
+    }
+
+    #[test]
+    fn start_args_default_output_is_trace_dir() {
         let args = RecordStartArgs {
             browser: None,
             tab_id: None,
             url: None,
             purpose: None,
-            output: PathBuf::from("trace.json"),
+            max_page_tokens: None,
+            redact_values: false,
+            output: PathBuf::from("trace"),
         };
-        assert_eq!(args.output, PathBuf::from("trace.json"));
+        assert_eq!(args.output, PathBuf::from("trace"));
     }
 
     #[test]
@@ -264,4 +435,35 @@ mod tests {
         let got = record_await_ipc_timeout(RECORD_AWAIT_TIMEOUT_MS);
         assert!(got >= Duration::from_secs(86_400));
     }
+
+    #[test]
+    fn annotate_default_start_page_error_adds_url_reason() {
+        let err = CliError::from_rpc(bsk_protocol::RpcError {
+            code: ErrorCode::CdpFailed,
+            message: "Page.navigate rejected: net::ERR_ABORTED".into(),
+            data: None,
+        });
+        let annotated = annotate_default_start_page_error(err, None);
+        assert_eq!(
+            crate::cli::render_error::reason_for_data(annotated.data()),
+            Some(crate::cli::render_error::reason::RECORD_START_PAGE_UNREACHABLE)
+        );
+        let info =
+            crate::cli::render_error::info_for_error(annotated.code().unwrap(), annotated.data());
+        assert!(info.hint.unwrap().contains("--url"));
+    }
+
+    #[test]
+    fn annotate_default_start_page_error_skips_when_user_passed_url() {
+        let err = CliError::from_rpc(bsk_protocol::RpcError {
+            code: ErrorCode::CdpFailed,
+            message: "Page.navigate rejected: net::ERR_ABORTED".into(),
+            data: None,
+        });
+        let annotated = annotate_default_start_page_error(err, Some("https://www.example.org/"));
+        assert_eq!(
+            crate::cli::render_error::reason_for_data(annotated.data()),
+            None
+        );
+    }
 }

crates/bsk-cli/src/cli/record/export.rs

@@ -0,0 +1,976 @@
+use std::collections::HashSet;
+use std::fs::{self, OpenOptions};
+use std::io;
+use std::path::{Path, PathBuf};
+
+use anyhow::Context;
+use bsk_protocol::tools::{RecordedTrace, TRACE_VERSION_V2, TRACE_VERSION_V3, TraceV2, TraceV3};
+use serde::Serialize;
+
+use crate::cli::error::CliError;
+
+#[derive(Debug)]
+pub(super) struct ExportMeta {
+    pub(super) states_dir: Option<PathBuf>,
+    pub(super) trace_version: u32,
+    /// True when the CLI requested v3 but the extension returned legacy v2.
+    pub(super) v2_fallback: bool,
+}
+
+pub(super) fn export_recorded_trace(
+    output_dir: &Path,
+    trace: &RecordedTrace,
+) -> Result<ExportMeta, CliError> {
+    match trace {
+        RecordedTrace::V3(trace) => {
+            let states_dir = write_trace_bundle(output_dir, trace)?;
+            Ok(ExportMeta {
+                states_dir: Some(states_dir),
+                trace_version: TRACE_VERSION_V3,
+                v2_fallback: false,
+            })
+        }
+        RecordedTrace::V2(trace) => {
+            write_trace_v2(output_dir, trace)?;
+            Ok(ExportMeta {
+                states_dir: None,
+                trace_version: TRACE_VERSION_V2,
+                v2_fallback: true,
+            })
+        }
+    }
+}
+
+/// Save the completed Trace first, then write the bundle. Recovery is
+/// removed only after the export commits so a bad `--output` cannot drop
+/// a recording the extension already returned.
+pub(super) fn export_with_recovery(
+    output_dir: &Path,
+    trace: &RecordedTrace,
+) -> Result<ExportMeta, CliError> {
+    let save_result = crate::cli::record_recovery::save(trace);
+    match export_recorded_trace(output_dir, trace) {
+        Ok(meta) => {
+            crate::cli::record_recovery::clear();
+            Ok(meta)
+        }
+        Err(export_err) => {
+            let export_err = annotate_recovery_hint(export_err, save_result.is_ok());
+            if let Err(save_err) = save_result {
+                return Err(match export_err {
+                    CliError::Local(inner) => CliError::Local(
+                        inner.context(format!("also failed to save recovery data: {save_err}")),
+                    ),
+                    other => other,
+                });
+            }
+            Err(export_err)
+        }
+    }
+}
+
+fn annotate_recovery_hint(err: CliError, recovery_saved: bool) -> CliError {
+    if !recovery_saved {
+        return err;
+    }
+    match err {
+        CliError::Local(inner) => CliError::Local(inner.context(
+            "export failed; the recorded trace was saved and can be recovered with `bsk record stop --output <dir>`",
+        )),
+        other => other,
+    }
+}
+
+fn looks_like_json_output(path: &Path) -> bool {
+    path.extension()
+        .and_then(|ext| ext.to_str())
+        .is_some_and(|ext| ext.eq_ignore_ascii_case("json"))
+}
+
+fn legacy_json_output_error(path: &Path) -> CliError {
+    CliError::Local(anyhow::anyhow!(
+        "--output {} is a JSON file path; Trace v3 writes a bundle directory \
+         (`<dir>/trace.json` and `<dir>/states/`), not a single file. \
+         Use `--output trace` or another directory.",
+        path.display()
+    ))
+}
+
+/// Reject legacy `--output trace.json` (and any existing non-directory)
+/// before a recording starts, so the user is not blocked only after Finish.
+pub(super) fn validate_record_output(path: &Path) -> Result<(), CliError> {
+    match fs::symlink_metadata(path) {
+        Ok(metadata) if metadata.file_type().is_symlink() => Err(CliError::Local(anyhow::anyhow!(
+            "--output {} must not be a symlink",
+            path.display()
+        ))),
+        Ok(metadata) if metadata.is_dir() => Ok(()),
+        Ok(_) if looks_like_json_output(path) => Err(legacy_json_output_error(path)),
+        Ok(_) => Err(CliError::Local(anyhow::anyhow!(
+            "--output {} is not a directory; Trace v3 writes a bundle directory \
+             (`<dir>/trace.json` and `<dir>/states/`). Use `--output trace`.",
+            path.display()
+        ))),
+        Err(err) if err.kind() == io::ErrorKind::NotFound => {
+            if looks_like_json_output(path) {
+                Err(legacy_json_output_error(path))
+            } else {
+                Ok(())
+            }
+        }
+        Err(err) => Err(CliError::Local(
+            anyhow::Error::new(err).context(format!("inspect --output {}", path.display())),
+        )),
+    }
+}
+
+fn acquire_export_lock(output_dir: &Path) -> Result<std::fs::File, CliError> {
+    validate_bundle_directory(output_dir, "output bundle directory")?;
+    fs::create_dir_all(output_dir)
+        .with_context(|| format!("create output dir {}", output_dir.display()))
+        .map_err(CliError::Local)?;
+    let lock_path = output_dir.join(".bsk-record-export.lock");
+    let lock_file = OpenOptions::new()
+        .read(true)
+        .write(true)
+        .create(true)
+        .truncate(false)
+        .open(&lock_path)
+        .with_context(|| format!("open export lock {}", lock_path.display()))
+        .map_err(CliError::Local)?;
+    fs2::FileExt::lock_exclusive(&lock_file)
+        .with_context(|| format!("lock export bundle {}", output_dir.display()))
+        .map_err(CliError::Local)?;
+    Ok(lock_file)
+}
+
+pub(super) fn write_trace_v2(output_dir: &Path, trace: &TraceV2) -> Result<(), CliError> {
+    let trace_path = trace_json_path(output_dir);
+    let states_dir = states_dir_for_output(output_dir);
+    let json = serde_json::to_string_pretty(trace)
+        .context("serialize trace JSON")
+        .map_err(CliError::Local)?;
+
+    let _lock = acquire_export_lock(output_dir)?;
+    validate_replaceable_file(&trace_path, "trace JSON")?;
+
+    let transaction_id = uuid::Uuid::new_v4();
+    let staging_dir = output_dir.join(format!(".bsk-record-stage-{transaction_id}"));
+    fs::create_dir_all(&staging_dir)
+        .with_context(|| format!("create staging dir {}", staging_dir.display()))
+        .map_err(CliError::Local)?;
+
+    let staged_trace = staging_dir.join("trace.json");
+    let stage_result: Result<(), CliError> = (|| {
+        fs::write(&staged_trace, format!("{json}\n"))
+            .with_context(|| format!("write staged trace to {}", staged_trace.display()))
+            .map_err(CliError::Local)?;
+        Ok(())
+    })();
+    if let Err(err) = stage_result {
+        let _ = fs::remove_dir_all(&staging_dir);
+        return Err(err);
+    }
+
+    commit_export_transaction(
+        &staging_dir,
+        &states_dir,
+        &HashSet::new(),
+        transaction_id,
+        vec![(
+            staged_trace,
+            trace_path,
+            output_dir.join(format!(".trace.json.{transaction_id}.bak")),
+        )],
+    )?;
+
+    Ok(())
+}
+
+pub(super) fn states_dir_for_output(output_dir: &Path) -> PathBuf {
+    output_dir.join("states")
+}
+
+pub(super) fn trace_json_path(output_dir: &Path) -> PathBuf {
+    output_dir.join("trace.json")
+}
+
+fn is_canonical_state_id(id: &str) -> bool {
+    let Some(number) = id.strip_prefix('s') else {
+        return false;
+    };
+    let mut digits = number.bytes();
+    matches!(digits.next(), Some(b'1'..=b'9')) && digits.all(|byte| byte.is_ascii_digit())
+}
+
+fn validate_bundle_directory(path: &Path, description: &str) -> Result<(), CliError> {
+    match fs::symlink_metadata(path) {
+        Ok(metadata) if metadata.file_type().is_symlink() => Err(CliError::Local(anyhow::anyhow!(
+            "{description} {} must not be a symlink",
+            path.display()
+        ))),
+        Ok(metadata) if metadata.is_dir() => Ok(()),
+        Ok(_) if looks_like_json_output(path) => Err(legacy_json_output_error(path)),
+        Ok(_) => Err(CliError::Local(anyhow::anyhow!(
+            "{description} {} is not a directory; Trace v3 writes `<dir>/trace.json` and `<dir>/states/`. Use `--output trace`.",
+            path.display()
+        ))),
+        Err(err) if err.kind() == io::ErrorKind::NotFound => Ok(()),
+        Err(err) => Err(CliError::Local(
+            anyhow::Error::new(err).context(format!("inspect {description} {}", path.display())),
+        )),
+    }
+}
+
+fn validate_replaceable_file(path: &Path, description: &str) -> Result<(), CliError> {
+    match fs::symlink_metadata(path) {
+        Ok(metadata) if metadata.file_type().is_file() => Ok(()),
+        Ok(_) => Err(CliError::Local(anyhow::anyhow!(
+            "{description} {} is not a regular file",
+            path.display()
+        ))),
+        Err(err) if err.kind() == io::ErrorKind::NotFound => Ok(()),
+        Err(err) => Err(CliError::Local(
+            anyhow::Error::new(err).context(format!("inspect {description} {}", path.display())),
+        )),
+    }
+}
+
+fn commit_staged_file(staged: &Path, target: &Path, backup: &Path) -> io::Result<Option<PathBuf>> {
+    let old_file = match fs::symlink_metadata(target) {
+        Ok(_) => {
+            fs::rename(target, backup)?;
+            Some(backup.to_path_buf())
+        }
+        Err(err) if err.kind() == io::ErrorKind::NotFound => None,
+        Err(err) => return Err(err),
+    };
+
+    if let Err(err) = fs::rename(staged, target) {
+        if let Some(old_file) = old_file.as_deref() {
+            if let Err(restore_err) = fs::rename(old_file, target) {
+                return Err(io::Error::new(
+                    restore_err.kind(),
+                    format!(
+                        "{err}; additionally failed to restore {}: {restore_err}",
+                        target.display()
+                    ),
+                ));
+            }
+        }
+        return Err(err);
+    }
+
+    Ok(old_file)
+}
+
+fn rollback_installed_files(installed: &[(PathBuf, Option<PathBuf>)]) -> io::Result<()> {
+    let mut rollback_error = None;
+    for (target, backup) in installed.iter().rev() {
+        match fs::remove_file(target) {
+            Ok(()) => {}
+            Err(err) if err.kind() == io::ErrorKind::NotFound => {}
+            Err(err) => {
+                rollback_error.get_or_insert(err);
+                continue;
+            }
+        }
+        if let Some(backup) = backup {
+            if let Err(err) = fs::rename(backup, target) {
+                rollback_error.get_or_insert(err);
+            }
+        }
+    }
+    match rollback_error {
+        Some(err) => Err(err),
+        None => Ok(()),
+    }
+}
+
+fn is_generated_state_name(name: &str) -> bool {
+    name.strip_suffix(".txt").is_some_and(is_canonical_state_id)
+}
+
+fn restore_staged_stale_states(staged: &[(PathBuf, PathBuf)]) -> io::Result<()> {
+    let mut restore_error = None;
+    for (original, backup) in staged.iter().rev() {
+        if let Err(err) = fs::rename(backup, original) {
+            restore_error.get_or_insert(err);
+        }
+    }
+    match restore_error {
+        Some(err) => Err(err),
+        None => Ok(()),
+    }
+}
+
+/// Move stale generated states aside so replacement and cleanup form one
+/// rollback boundary. User-owned files, including arbitrary non-`sN.txt`, stay.
+fn stage_stale_states(
+    states_dir: &Path,
+    keep: &HashSet<String>,
+    transaction_id: uuid::Uuid,
+) -> Result<Vec<(PathBuf, PathBuf)>, CliError> {
+    let entries = match fs::read_dir(states_dir) {
+        Ok(entries) => entries,
+        Err(err) if err.kind() == io::ErrorKind::NotFound => return Ok(Vec::new()),
+        Err(err) => {
+            return Err(CliError::Local(
+                anyhow::Error::new(err)
+                    .context(format!("read states dir {}", states_dir.display())),
+            ));
+        }
+    };
+
+    let mut staged = Vec::new();
+    for entry in entries {
+        let entry = match entry {
+            Ok(entry) => entry,
+            Err(err) => {
+                let restore_result = restore_staged_stale_states(&staged);
+                let message = match restore_result {
+                    Ok(()) => format!("read stale state entry: {err}"),
+                    Err(restore_err) => {
+                        format!("read stale state entry: {err}; restore also failed: {restore_err}")
+                    }
+                };
+                return Err(CliError::Local(anyhow::anyhow!(message)));
+            }
+        };
+        let name = entry.file_name();
+        let Some(name) = name.to_str() else { continue };
+        if !is_generated_state_name(name) || keep.contains(name) {
+            continue;
+        }
+
+        let original = entry.path();
+        let file_type = match entry.file_type() {
+            Ok(file_type) => file_type,
+            Err(err) => {
+                let restore_result = restore_staged_stale_states(&staged);
+                let message = match restore_result {
+                    Ok(()) => format!("inspect stale state {}: {err}", original.display()),
+                    Err(restore_err) => format!(
+                        "inspect stale state {}: {err}; restore also failed: {restore_err}",
+                        original.display()
+                    ),
+                };
+                return Err(CliError::Local(anyhow::anyhow!(message)));
+            }
+        };
+        if file_type.is_dir() {
+            let restore_result = restore_staged_stale_states(&staged);
+            let message = match restore_result {
+                Ok(()) => format!("stale state {} is a directory", original.display()),
+                Err(restore_err) => format!(
+                    "stale state {} is a directory; restore also failed: {restore_err}",
+                    original.display()
+                ),
+            };
+            return Err(CliError::Local(anyhow::anyhow!(message)));
+        }
+
+        let backup = states_dir.join(format!(".{name}.{transaction_id}.stale"));
+        if let Err(err) = fs::rename(&original, &backup) {
+            let restore_result = restore_staged_stale_states(&staged);
+            let message = match restore_result {
+                Ok(()) => format!("stage stale state {}: {err}", original.display()),
+                Err(restore_err) => format!(
+                    "stage stale state {}: {err}; restore also failed: {restore_err}",
+                    original.display()
+                ),
+            };
+            return Err(CliError::Local(anyhow::anyhow!(message)));
+        }
+        staged.push((original, backup));
+    }
+    Ok(staged)
+}
+
+fn commit_export_transaction(
+    staging_dir: &Path,
+    states_dir: &Path,
+    keep_states: &HashSet<String>,
+    transaction_id: uuid::Uuid,
+    replacements: Vec<(PathBuf, PathBuf, PathBuf)>,
+) -> Result<(), CliError> {
+    let mut installed = Vec::with_capacity(replacements.len());
+    for (staged, target, backup) in replacements {
+        match commit_staged_file(&staged, &target, &backup) {
+            Ok(old_file) => installed.push((target, old_file)),
+            Err(err) => {
+                let rollback_result = rollback_installed_files(&installed);
+                let _ = fs::remove_dir_all(staging_dir);
+                let message = match rollback_result {
+                    Ok(()) => format!("commit replacement {}: {err}", target.display()),
+                    Err(rollback_err) => format!(
+                        "commit replacement {}: {err}; rollback also failed: {rollback_err}",
+                        target.display()
+                    ),
+                };
+                return Err(CliError::Local(anyhow::anyhow!(message)));
+            }
+        }
+    }
+
+    let stale_states = match stage_stale_states(states_dir, keep_states, transaction_id) {
+        Ok(stale) => stale,
+        Err(err) => {
+            let rollback_result = rollback_installed_files(&installed);
+            let _ = fs::remove_dir_all(staging_dir);
+            let message = match rollback_result {
+                Ok(()) => err.to_string(),
+                Err(rollback_err) => {
+                    format!("{err}; replacement rollback also failed: {rollback_err}")
+                }
+            };
+            return Err(CliError::Local(anyhow::anyhow!(message)));
+        }
+    };
+
+    for (_, backup) in &installed {
+        if let Some(backup) = backup {
+            let _ = fs::remove_file(backup);
+        }
+    }
+    for (_, backup) in stale_states {
+        let _ = fs::remove_file(backup);
+    }
+    let _ = fs::remove_dir_all(staging_dir);
+    Ok(())
+}
+
+#[derive(Serialize)]
+struct BundleState<'a> {
+    id: &'a str,
+    url: &'a str,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    title: Option<&'a str>,
+    page: String,
+    #[serde(skip_serializing_if = "std::ops::Not::not")]
+    truncated: bool,
+}
+
+#[derive(Serialize)]
+struct BundleTrace<'a> {
+    version: u32,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    purpose: Option<&'a str>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    started_at: Option<&'a str>,
+    recorded_at: &'a str,
+    stopped_by: &'a bsk_protocol::tools::StopReason,
+    entry: &'a bsk_protocol::tools::TraceEntry,
+    recorder: &'a bsk_protocol::tools::RecorderInfo,
+    states: Vec<BundleState<'a>>,
+    steps: &'a [bsk_protocol::tools::StepV3],
+}
+
+pub(super) fn write_trace_bundle(output_dir: &Path, trace: &TraceV3) -> Result<PathBuf, CliError> {
+    let states_dir = states_dir_for_output(output_dir);
+    let trace_path = trace_json_path(output_dir);
+
+    if trace.version != TRACE_VERSION_V3 {
+        return Err(CliError::Local(anyhow::anyhow!(
+            "trace version {} is not supported (expected {})",
+            trace.version,
+            TRACE_VERSION_V3
+        )));
+    }
+
+    let mut state_ids = HashSet::with_capacity(trace.states.len());
+    for state in &trace.states {
+        if !is_canonical_state_id(&state.id) {
+            return Err(CliError::Local(anyhow::anyhow!(
+                "state id {:?} is not canonical (expected s1, s2, ...)",
+                state.id
+            )));
+        }
+        if !state_ids.insert(state.id.as_str()) {
+            return Err(CliError::Local(anyhow::anyhow!(
+                "duplicate state id {:?}",
+                state.id
+            )));
+        }
+    }
+
+    let mut disk_states = Vec::with_capacity(trace.states.len());
+    let mut state_writes = Vec::new();
+    let mut keep = HashSet::new();
+    for state in &trace.states {
+        let filename = format!("{}.txt", state.id);
+        let state_path = states_dir.join(&filename);
+        if state_path.parent() != Some(states_dir.as_path()) {
+            return Err(CliError::Local(anyhow::anyhow!(
+                "state path {} escapes states directory {}",
+                state_path.display(),
+                states_dir.display()
+            )));
+        }
+        keep.insert(filename.clone());
+        state_writes.push((filename.clone(), state.body.as_str()));
+        disk_states.push(BundleState {
+            id: &state.id,
+            url: &state.url,
+            title: state.title.as_deref(),
+            page: filename,
+            truncated: state.truncated,
+        });
+    }
+
+    let disk_trace = BundleTrace {
+        version: trace.version,
+        purpose: trace.purpose.as_deref(),
+        started_at: trace.started_at.as_deref(),
+        recorded_at: &trace.recorded_at,
+        stopped_by: &trace.stopped_by,
+        entry: &trace.entry,
+        recorder: &trace.recorder,
+        states: disk_states,
+        steps: &trace.steps,
+    };
+    let json = serde_json::to_string_pretty(&disk_trace)
+        .context("serialize trace JSON")
+        .map_err(CliError::Local)?;
+
+    validate_bundle_directory(output_dir, "output bundle directory")?;
+    fs::create_dir_all(output_dir)
+        .with_context(|| format!("create output dir {}", output_dir.display()))
+        .map_err(CliError::Local)?;
+
+    validate_bundle_directory(&states_dir, "states directory")?;
+
+    let _lock = acquire_export_lock(output_dir)?;
+
+    fs::create_dir_all(&states_dir)
+        .with_context(|| format!("create states dir {}", states_dir.display()))
+        .map_err(CliError::Local)?;
+
+    for (filename, _) in &state_writes {
+        validate_replaceable_file(&states_dir.join(filename), "state observation")?;
+    }
+    validate_replaceable_file(&trace_path, "trace JSON")?;
+
+    let transaction_id = uuid::Uuid::new_v4();
+    let staging_dir = output_dir.join(format!(".bsk-record-stage-{transaction_id}"));
+    let staging_states = staging_dir.join("states");
+    fs::create_dir_all(&staging_states)
+        .with_context(|| format!("create staging dir {}", staging_states.display()))
+        .map_err(CliError::Local)?;
+
+    let stage_result: Result<(), CliError> = (|| {
+        for (filename, body) in &state_writes {
+            let staged_path = staging_states.join(filename);
+            fs::write(&staged_path, body)
+                .with_context(|| {
+                    format!("write staged state observation {}", staged_path.display())
+                })
+                .map_err(CliError::Local)?;
+        }
+        let staged_trace = staging_dir.join("trace.json");
+        fs::write(&staged_trace, format!("{json}\n"))
+            .with_context(|| format!("write staged trace to {}", staged_trace.display()))
+            .map_err(CliError::Local)?;
+        Ok(())
+    })();
+    if let Err(err) = stage_result {
+        let _ = fs::remove_dir_all(&staging_dir);
+        return Err(err);
+    }
+
+    let mut replacements: Vec<(PathBuf, PathBuf, PathBuf)> =
+        Vec::with_capacity(state_writes.len() + 1);
+    for (filename, _) in &state_writes {
+        replacements.push((
+            staging_states.join(filename),
+            states_dir.join(filename),
+            states_dir.join(format!(".{filename}.{transaction_id}.bak")),
+        ));
+    }
+    replacements.push((
+        staging_dir.join("trace.json"),
+        trace_path.clone(),
+        output_dir.join(format!(".trace.json.{transaction_id}.bak")),
+    ));
+
+    commit_export_transaction(
+        &staging_dir,
+        &states_dir,
+        &keep,
+        transaction_id,
+        replacements,
+    )?;
+
+    Ok(states_dir)
+}
+
+#[cfg(test)]
+mod tests {
+    use std::fs;
+
+    use super::*;
+    use bsk_protocol::tools::{
+        NavigationCause, PageRefV2 as PageRef, RecorderInfo, StepCommonV3 as StepCommon,
+        StepResultV3 as StepResult, StepV3 as Step, StopReason, TRACE_VERSION_V2, TRACE_VERSION_V3,
+        TraceEntry, TraceStateV3, TraceV2, TraceV3, VOM_FORMAT_VERSION,
+    };
+
+    fn sample_trace(state_id: &str, body: &str) -> TraceV3 {
+        TraceV3 {
+            version: TRACE_VERSION_V3,
+            recorded_at: "2026-08-10T02:12:55.360Z".into(),
+            stopped_by: StopReason::UserFinish,
+            entry: TraceEntry {
+                start_url: "https://example.com/".into(),
+            },
+            recorder: RecorderInfo {
+                bsk: "0.1.10".into(),
+                vom: VOM_FORMAT_VERSION,
+            },
+            states: vec![TraceStateV3 {
+                id: state_id.into(),
+                url: "https://example.com/".into(),
+                title: Some("Example".into()),
+                body: body.into(),
+                truncated: false,
+            }],
+            steps: vec![Step::Navigate {
+                common: StepCommon {
+                    id: 1,
+                    state: state_id.into(),
+                    result: StepResult {
+                        state: state_id.into(),
+                    },
+                },
+                to: "https://example.com/".into(),
+                cause: NavigationCause::UserTyped,
+            }],
+            purpose: None,
+            started_at: None,
+        }
+    }
+
+    #[test]
+    fn write_trace_bundle_writes_dir_layout() {
+        let dir = tempfile::tempdir().unwrap();
+        let output = dir.path().join("flow");
+        let trace = sample_trace("s1", "@vom 1\nL1 page");
+        let states_dir = write_trace_bundle(&output, &trace).unwrap();
+        let trace_path = output.join("trace.json");
+        assert!(trace_path.exists());
+        assert_eq!(states_dir, output.join("states"));
+        assert!(states_dir.join("s1.txt").exists());
+        let disk: serde_json::Value =
+            serde_json::from_str(&fs::read_to_string(&trace_path).unwrap()).unwrap();
+        assert_eq!(disk["states"][0]["page"], "s1.txt");
+        assert!(disk["states"][0].get("body").is_none());
+        assert_eq!(
+            fs::read_to_string(states_dir.join("s1.txt")).unwrap(),
+            "@vom 1\nL1 page"
+        );
+
+        // Re-exporting drops states from the previous run but keeps the
+        // directory in place for the concurrent writer.
+        fs::write(states_dir.join("s9.txt"), "stale").unwrap();
+        fs::write(states_dir.join("notes.md"), "keep me").unwrap();
+        fs::write(states_dir.join("reference.txt"), "user file").unwrap();
+        write_trace_bundle(&output, &trace).unwrap();
+        assert!(!states_dir.join("s9.txt").exists());
+        assert!(states_dir.join("s1.txt").exists());
+        assert!(states_dir.join("notes.md").exists());
+        assert_eq!(
+            fs::read_to_string(states_dir.join("reference.txt")).unwrap(),
+            "user file"
+        );
+    }
+
+    #[test]
+    fn write_trace_bundle_rejects_traversal_state_id() {
+        let dir = tempfile::tempdir().unwrap();
+        let output = dir.path().join("flow");
+        let trace = sample_trace("../escaped", "@vom 1\nL1 page");
+
+        let err = write_trace_bundle(&output, &trace).unwrap_err();
+
+        assert!(err.to_string().contains("state id"));
+        assert!(!output.join("escaped.txt").exists());
+        assert!(!output.join("trace.json").exists());
+    }
+
+    #[test]
+    fn write_trace_bundle_rejects_duplicate_state_id() {
+        let dir = tempfile::tempdir().unwrap();
+        let output = dir.path().join("flow");
+        let mut trace = sample_trace("s1", "@vom 1\nL1 first");
+        trace.states.push(TraceStateV3 {
+            body: "@vom 1\nL1 duplicate".into(),
+            ..trace.states[0].clone()
+        });
+
+        let err = write_trace_bundle(&output, &trace).unwrap_err();
+
+        assert!(err.to_string().contains("duplicate state id"));
+        assert!(!output.exists());
+    }
+
+    #[test]
+    fn unsupported_version_preserves_existing_bundle() {
+        let dir = tempfile::tempdir().unwrap();
+        let output = dir.path().join("flow");
+        write_trace_bundle(&output, &sample_trace("s1", "original")).unwrap();
+        let original_json = fs::read(output.join("trace.json")).unwrap();
+
+        let mut replacement = sample_trace("s2", "replacement");
+        replacement.version += 1;
+        let err = write_trace_bundle(&output, &replacement).unwrap_err();
+
+        assert!(err.to_string().contains("not supported"));
+        assert_eq!(fs::read(output.join("trace.json")).unwrap(), original_json);
+        assert_eq!(
+            fs::read_to_string(output.join("states/s1.txt")).unwrap(),
+            "original"
+        );
+        assert!(!output.join("states/s2.txt").exists());
+    }
+
+    #[test]
+    fn state_replacement_failure_preserves_existing_bundle() {
+        let dir = tempfile::tempdir().unwrap();
+        let output = dir.path().join("flow");
+        write_trace_bundle(&output, &sample_trace("s1", "original")).unwrap();
+        let original_json = fs::read(output.join("trace.json")).unwrap();
+        fs::write(output.join("states/notes.md"), "keep me").unwrap();
+        fs::create_dir(output.join("states/s2.txt")).unwrap();
+
+        let err = write_trace_bundle(&output, &sample_trace("s2", "replacement")).unwrap_err();
+
+        assert!(err.to_string().contains("state"));
+        assert_eq!(fs::read(output.join("trace.json")).unwrap(), original_json);
+        assert_eq!(
+            fs::read_to_string(output.join("states/s1.txt")).unwrap(),
+            "original"
+        );
+        assert_eq!(
+            fs::read_to_string(output.join("states/notes.md")).unwrap(),
+            "keep me"
+        );
+    }
+
+    #[test]
+    fn stale_state_cleanup_failure_rolls_back_replacement() {
+        let dir = tempfile::tempdir().unwrap();
+        let output = dir.path().join("flow");
+        write_trace_bundle(&output, &sample_trace("s1", "original")).unwrap();
+        let original_json = fs::read(output.join("trace.json")).unwrap();
+        fs::create_dir(output.join("states/s9.txt")).unwrap();
+
+        let err = write_trace_bundle(&output, &sample_trace("s2", "replacement")).unwrap_err();
+
+        assert!(err.to_string().contains("stale state"));
+        assert_eq!(fs::read(output.join("trace.json")).unwrap(), original_json);
+        assert_eq!(
+            fs::read_to_string(output.join("states/s1.txt")).unwrap(),
+            "original"
+        );
+        assert!(!output.join("states/s2.txt").exists());
+        assert!(output.join("states/s9.txt").is_dir());
+    }
+
+    #[cfg(unix)]
+    #[test]
+    fn write_trace_bundle_rejects_symlinked_states_directory() {
+        use std::os::unix::fs::symlink;
+
+        let dir = tempfile::tempdir().unwrap();
+        let output = dir.path().join("flow");
+        let external_states = dir.path().join("external-states");
+        fs::create_dir_all(&output).unwrap();
+        fs::create_dir_all(&external_states).unwrap();
+        fs::write(external_states.join("s9.txt"), "external").unwrap();
+        symlink(&external_states, output.join("states")).unwrap();
+
+        let err = write_trace_bundle(&output, &sample_trace("s1", "replacement")).unwrap_err();
+
+        assert!(err.to_string().contains("symlink"));
+        assert_eq!(
+            fs::read_to_string(external_states.join("s9.txt")).unwrap(),
+            "external"
+        );
+        assert!(!external_states.join("s1.txt").exists());
+        assert!(!output.join("trace.json").exists());
+    }
+
+    #[cfg(unix)]
+    #[test]
+    fn write_trace_bundle_rejects_symlinked_output_directory() {
+        use std::os::unix::fs::symlink;
+
+        let dir = tempfile::tempdir().unwrap();
+        let external_output = dir.path().join("external-output");
+        let output = dir.path().join("flow");
+        fs::create_dir_all(&external_output).unwrap();
+        fs::write(external_output.join("notes.md"), "external").unwrap();
+        symlink(&external_output, &output).unwrap();
+
+        let err = write_trace_bundle(&output, &sample_trace("s1", "replacement")).unwrap_err();
+
+        assert!(err.to_string().contains("symlink"));
+        assert_eq!(
+            fs::read_to_string(external_output.join("notes.md")).unwrap(),
+            "external"
+        );
+        assert!(!external_output.join("trace.json").exists());
+        assert!(!external_output.join("states").exists());
+    }
+
+    #[test]
+    fn write_trace_v2_writes_single_json_and_prunes_stale_states() {
+        let dir = tempfile::tempdir().unwrap();
+        let output = dir.path().join("flow");
+        fs::create_dir_all(output.join("states")).unwrap();
+        fs::write(output.join("states/s1.txt"), "stale").unwrap();
+
+        let trace = TraceV2 {
+            recorded_at: "2026-08-10T02:12:55.360Z".into(),
+            started_at: None,
+            purpose: None,
+            entry: bsk_protocol::tools::TraceEntry {
+                start_url: "https://example.com/".into(),
+            },
+            pages: vec![PageRef {
+                id: "p1".into(),
+                url: "https://example.com/".into(),
+                title: None,
+            }],
+            steps: vec![],
+        };
+        write_trace_v2(&output, &trace).unwrap();
+        assert!(output.join("trace.json").exists());
+        assert!(!output.join("states/s1.txt").exists());
+    }
+
+    #[test]
+    fn write_trace_v2_stale_cleanup_failure_rolls_back_trace() {
+        let dir = tempfile::tempdir().unwrap();
+        let output = dir.path().join("flow");
+        write_trace_bundle(&output, &sample_trace("s1", "original")).unwrap();
+        let original_json = fs::read(output.join("trace.json")).unwrap();
+        fs::create_dir(output.join("states/s9.txt")).unwrap();
+        let trace = TraceV2 {
+            recorded_at: "2026-08-10T02:12:55.360Z".into(),
+            started_at: None,
+            purpose: None,
+            entry: bsk_protocol::tools::TraceEntry {
+                start_url: "https://example.com/".into(),
+            },
+            pages: vec![],
+            steps: vec![],
+        };
+
+        let err = write_trace_v2(&output, &trace).unwrap_err();
+
+        assert!(err.to_string().contains("stale state"));
+        assert_eq!(fs::read(output.join("trace.json")).unwrap(), original_json);
+        assert_eq!(
+            fs::read_to_string(output.join("states/s1.txt")).unwrap(),
+            "original"
+        );
+        assert!(output.join("states/s9.txt").is_dir());
+    }
+
+    #[test]
+    fn export_recorded_trace_accepts_v2() {
+        let dir = tempfile::tempdir().unwrap();
+        let output = dir.path().join("flow");
+        let trace = RecordedTrace::V2(TraceV2 {
+            recorded_at: "2026-08-10T02:12:55.360Z".into(),
+            started_at: None,
+            purpose: None,
+            entry: bsk_protocol::tools::TraceEntry {
+                start_url: "https://example.com/".into(),
+            },
+            pages: vec![],
+            steps: vec![],
+        });
+        let exported = export_recorded_trace(&output, &trace).unwrap();
+        assert_eq!(exported.trace_version, TRACE_VERSION_V2);
+        assert!(exported.states_dir.is_none());
+        assert!(output.join("trace.json").exists());
+    }
+
+    #[test]
+    fn existing_trace_json_file_is_rejected_with_bundle_usage() {
+        let dir = tempfile::tempdir().unwrap();
+        let output = dir.path().join("trace.json");
+        fs::write(&output, "{}\n").unwrap();
+
+        let err = validate_record_output(&output).unwrap_err();
+        let msg = err.to_string();
+        assert!(
+            msg.contains("JSON file") || msg.contains("json file"),
+            "should name the legacy JSON file usage: {msg}"
+        );
+        assert!(
+            msg.contains("states/") || msg.contains("--output trace"),
+            "should hint at the new bundle directory usage: {msg}"
+        );
+        assert_eq!(fs::read_to_string(&output).unwrap(), "{}\n");
+    }
+
+    #[test]
+    fn missing_trace_json_path_is_rejected_as_legacy_output() {
+        let err = validate_record_output(Path::new("trace.json")).unwrap_err();
+        let msg = err.to_string();
+        assert!(
+            msg.contains("JSON file") || msg.contains("json file"),
+            "{msg}"
+        );
+        assert!(
+            msg.contains("states/") || msg.contains("--output trace"),
+            "{msg}"
+        );
+    }
+
+    #[test]
+    fn export_failure_keeps_recovery_for_retry() {
+        let _lock = crate::cli::record_recovery::test_env_lock();
+        let tmp = tempfile::tempdir().unwrap();
+        unsafe {
+            std::env::set_var(crate::daemon::paths::BSK_HOME_ENV, tmp.path());
+        }
+
+        let dir = tempfile::tempdir().unwrap();
+        let bad_output = dir.path().join("trace.json");
+        fs::write(&bad_output, "{}\n").unwrap();
+        let trace = RecordedTrace::V3(sample_trace("s1", "@vom 1\nL1 page"));
+
+        let err = export_with_recovery(&bad_output, &trace).unwrap_err();
+        let msg = err.to_string();
+        assert!(
+            msg.contains("recover") || msg.contains("record stop"),
+            "export failure should tell the user how to recover: {msg}"
+        );
+        assert!(
+            crate::cli::record_recovery::exists(),
+            "recovery data must survive a failed bundle export"
+        );
+        let recovered = crate::cli::record_recovery::load()
+            .unwrap()
+            .expect("recovery file");
+        assert_eq!(recovered, trace);
+        assert_eq!(fs::read_to_string(&bad_output).unwrap(), "{}\n");
+
+        let good_output = dir.path().join("bundle");
+        export_with_recovery(&good_output, &recovered).unwrap();
+        assert!(good_output.join("trace.json").exists());
+        assert!(good_output.join("states/s1.txt").exists());
+        assert!(
+            !crate::cli::record_recovery::exists(),
+            "successful export must delete recovery data"
+        );
+
+        unsafe {
+            std::env::remove_var(crate::daemon::paths::BSK_HOME_ENV);
+        }
+    }
+}

crates/bsk-cli/src/cli/record_recovery.rs

@@ -0,0 +1,112 @@
+//! Persist a completed Trace so a failed bundle export can be retried.
+
+use std::fs;
+
+use anyhow::{Context, Result};
+use bsk_protocol::tools::RecordedTrace;
+
+use crate::daemon::paths;
+
+pub fn save(trace: &RecordedTrace) -> Result<()> {
+    let path = paths::record_recovery_path()?;
+    paths::ensure_bsk_home()?;
+    let json = serde_json::to_string_pretty(trace).context("serialize record recovery")?;
+    let tmp = path.with_extension("tmp");
+    fs::write(&tmp, format!("{json}\n")).with_context(|| format!("write {}", tmp.display()))?;
+    fs::rename(&tmp, &path).with_context(|| format!("commit {}", path.display()))?;
+    Ok(())
+}
+
+pub fn load() -> Result<Option<RecordedTrace>> {
+    let path = paths::record_recovery_path()?;
+    if !path.exists() {
+        return Ok(None);
+    }
+    let raw = fs::read_to_string(&path).with_context(|| format!("read {}", path.display()))?;
+    let trace = serde_json::from_str(&raw).with_context(|| format!("parse {}", path.display()))?;
+    Ok(Some(trace))
+}
+
+pub fn exists() -> bool {
+    paths::record_recovery_path()
+        .map(|path| path.exists())
+        .unwrap_or(false)
+}
+
+pub fn clear() {
+    if let Ok(path) = paths::record_recovery_path() {
+        let _ = fs::remove_file(path);
+    }
+}
+
+#[cfg(test)]
+pub(crate) fn test_env_lock() -> std::sync::MutexGuard<'static, ()> {
+    crate::daemon::paths::test_env_lock()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use bsk_protocol::tools::{
+        RecorderInfo, StopReason, TRACE_VERSION_V3, TraceEntry, TraceStateV3, TraceV3,
+        VOM_FORMAT_VERSION,
+    };
+
+    fn sample_trace() -> RecordedTrace {
+        RecordedTrace::V3(TraceV3 {
+            version: TRACE_VERSION_V3,
+            recorded_at: "2026-08-10T02:12:55.360Z".into(),
+            stopped_by: StopReason::UserFinish,
+            entry: TraceEntry {
+                start_url: "https://example.com/".into(),
+            },
+            recorder: RecorderInfo {
+                bsk: "0.1.10".into(),
+                vom: VOM_FORMAT_VERSION,
+            },
+            states: vec![TraceStateV3 {
+                id: "s1".into(),
+                url: "https://example.com/".into(),
+                title: Some("Example".into()),
+                body: "@vom 1\nL1 page".into(),
+                truncated: false,
+            }],
+            steps: vec![],
+            purpose: None,
+            started_at: None,
+        })
+    }
+
+    fn with_temp_home<F: FnOnce()>(f: F) {
+        let _lock = test_env_lock();
+        let tmp = tempfile::tempdir().unwrap();
+        unsafe {
+            std::env::set_var(crate::daemon::paths::BSK_HOME_ENV, tmp.path());
+        }
+        f();
+        clear();
+        unsafe {
+            std::env::remove_var(crate::daemon::paths::BSK_HOME_ENV);
+        }
+    }
+
+    #[test]
+    fn save_load_round_trips() {
+        with_temp_home(|| {
+            let trace = sample_trace();
+            save(&trace).unwrap();
+            assert!(exists());
+            assert_eq!(load().unwrap(), Some(trace));
+        });
+    }
+
+    #[test]
+    fn clear_removes_recovery_file() {
+        with_temp_home(|| {
+            save(&sample_trace()).unwrap();
+            clear();
+            assert!(!exists());
+            assert_eq!(load().unwrap(), None);
+        });
+    }
+}

crates/bsk-cli/src/cli/record_state.rs

@@ -59,15 +59,9 @@ fn started_at_unix_ms() -> String {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use std::sync::Mutex;
-
-    fn env_guard() -> &'static Mutex<()> {
-        static GUARD: Mutex<()> = Mutex::new(());
-        &GUARD
-    }
 
     fn with_temp_home<F: FnOnce()>(f: F) {
-        let _lock = env_guard().lock().unwrap_or_else(|e| e.into_inner());
+        let _lock = crate::daemon::paths::test_env_lock();
         let tmp = tempfile::tempdir().unwrap();
         unsafe {
             std::env::set_var(crate::daemon::paths::BSK_HOME_ENV, tmp.path());

crates/bsk-cli/src/cli/render_error.rs

@@ -43,13 +43,29 @@ pub mod reason {
     pub const REF_NOT_FOUND: &str = "ref_not_found";
     pub const SELECTOR_NOT_FOUND: &str = "selector_not_found";
     pub const TARGET_NOT_FILLABLE: &str = "target_not_fillable";
+    pub const FILL_VALUE_INVALID: &str = "fill_value_invalid";
+    pub const FILL_TARGET_CHANGED: &str = "fill_target_changed";
+    pub const FILL_FOCUS_LOST: &str = "fill_focus_lost";
+    pub const FILL_VALUE_MISMATCH: &str = "fill_value_mismatch";
+    pub const FILL_FAILED: &str = "fill_failed";
     pub const TARGET_NOT_SELECT: &str = "target_not_select";
     pub const OPTION_NOT_FOUND: &str = "option_not_found";
     pub const SINGLE_SELECT_VALUE_COUNT: &str = "single_select_value_count";
     pub const TAB_NOT_ACTIVE: &str = "tab_not_active";
+    pub const CDP_EXTENSION_ACCESS_DENIED: &str = "cdp_extension_access_denied";
     pub const BORROW_CONFLICT: &str = "borrow_conflict";
     pub const SCREENSHOT_CAPTURE_FAILED: &str = "screenshot_capture_failed";
+    pub const FILE_INPUT_PROBE_FAILED: &str = "file_input_probe_failed";
+    pub const FILE_INPUT_NOT_ACTIVATED: &str = "file_input_not_activated";
+    pub const SET_FILE_INPUT_FAILED: &str = "set_file_input_failed";
+    pub const UPLOAD_MECHANISM_UNSUPPORTED: &str = "upload_mechanism_unsupported";
+    pub const FILE_DROP_TARGET_UNAVAILABLE: &str = "file_drop_target_unavailable";
+    pub const FILE_DROP_FAILED: &str = "file_drop_failed";
+    pub const DOWNLOAD_CAPTURE_FAILED: &str = "download_capture_failed";
+    pub const TRANSFER_OUTCOME_UNKNOWN: &str = "transfer_outcome_unknown";
+    pub const TRANSFER_TIMEOUT: &str = "transfer_timeout";
     pub const SESSION_BUSY: &str = crate::rpc_reason::SESSION_BUSY;
+    pub const RECORD_START_PAGE_UNREACHABLE: &str = "record_start_page_unreachable";
 }
 
 /// Per-code rendering metadata. The `summary` is what the CLI prints
@@ -196,6 +212,27 @@ pub fn info_for_error(code: ErrorCode, data: Option<&serde_json::Value>) -> Rend
         return base;
     };
     match (code, reason) {
+        (ErrorCode::CdpFailed, reason::CDP_EXTENSION_ACCESS_DENIED) => RenderInfo {
+            summary: "Chrome blocked CDP access to another extension's content in this tab",
+            hint: Some(
+                "a web page can contain a restricted extension frame; disable the conflicting extension and reload, or use `bsk navigate <url>` to leave this page; reconnecting BrowserSkill alone does not remove the restriction",
+            ),
+            exit_code: base.exit_code,
+        },
+        (_, reason::TRANSFER_OUTCOME_UNKNOWN) => RenderInfo {
+            summary: "the file transfer outcome could not be confirmed",
+            hint: Some(
+                "do not retry the transfer: the browser may already have applied it; inspect the page or stop safely",
+            ),
+            exit_code: base.exit_code,
+        },
+        (ErrorCode::Timeout, reason::TRANSFER_TIMEOUT) => RenderInfo {
+            summary: "the file transfer timed out after browser dispatch",
+            hint: Some(
+                "do not retry when effect_state is unknown or committed; inspect the page before taking another action",
+            ),
+            exit_code: base.exit_code,
+        },
         (ErrorCode::PermissionDenied, reason::ELEMENT_NOT_VISIBLE) => RenderInfo {
             summary: "target element has no visible geometry",
             hint: Some(
@@ -223,7 +260,42 @@ pub fn info_for_error(code: ErrorCode, data: Option<&serde_json::Value>) -> Rend
         (ErrorCode::InvalidParams, reason::TARGET_NOT_FILLABLE) => RenderInfo {
             summary: "target element is not fillable",
             hint: Some(
-                "choose an input, textarea, or contenteditable element from the latest snapshot",
+                "observe the page and choose an enabled, editable text input, textarea, or contenteditable; use the appropriate interaction for other control types",
+            ),
+            exit_code: base.exit_code,
+        },
+        (ErrorCode::InvalidParams, reason::FILL_VALUE_INVALID) => RenderInfo {
+            summary: "the requested text does not fit the field constraints",
+            hint: Some(
+                "check the field's input type and maxlength; use a compatible value without silently truncating or changing the user's intended data",
+            ),
+            exit_code: base.exit_code,
+        },
+        (ErrorCode::CdpFailed, reason::FILL_TARGET_CHANGED) => RenderInfo {
+            summary: "the fill target changed during the action",
+            hint: Some(
+                "observe the current page and field value; if the intended result is missing, use a fresh editable target before retrying",
+            ),
+            exit_code: base.exit_code,
+        },
+        (ErrorCode::CdpFailed, reason::FILL_FOCUS_LOST) => RenderInfo {
+            summary: "the page moved focus away from the fill target",
+            hint: Some(
+                "observe the page for a dialog or rerender and resolve it before retrying; fill manages field focus and does not require bringing the browser to the foreground",
+            ),
+            exit_code: base.exit_code,
+        },
+        (ErrorCode::CdpFailed, reason::FILL_VALUE_MISMATCH) => RenderInfo {
+            summary: "the fill result could not be confirmed",
+            hint: Some(
+                "observe the field before retrying; the page may have formatted the value. Continue if the visible result satisfies the user's intent; otherwise correct the remaining difference. Do not blindly repeat fill or immediately request human help",
+            ),
+            exit_code: base.exit_code,
+        },
+        (ErrorCode::CdpFailed, reason::FILL_FAILED) => RenderInfo {
+            summary: "fill encountered a browser or page-script error",
+            hint: Some(
+                "observe the page before retrying because the field may already have changed; retry only if the intended result is missing and the target is still available",
             ),
             exit_code: base.exit_code,
         },
@@ -269,6 +341,70 @@ pub fn info_for_error(code: ErrorCode, data: Option<&serde_json::Value>) -> Rend
             ),
             exit_code: base.exit_code,
         },
+        (
+            ErrorCode::CdpFailed | ErrorCode::Timeout | ErrorCode::InvalidParams,
+            reason::RECORD_START_PAGE_UNREACHABLE,
+        ) => RenderInfo {
+            summary: "default start page https://example.com/ could not be opened",
+            hint: Some(
+                "that site may be unreachable from your network. Pass --url with a page you can open, e.g. `bsk record start --url https://<site>/`",
+            ),
+            exit_code: base.exit_code,
+        },
+        (ErrorCode::Unsupported, reason::FILE_INPUT_NOT_ACTIVATED) => RenderInfo {
+            summary: "the upload trigger did not activate a file input",
+            hint: Some(
+                "the page may use a non-input picker such as showOpenFilePicker(); when effect_state is none, re-observe and use `--mode drop` once only for a reliably identified attachment-receiving area — otherwise use `bsk request-help` with the exact original local path",
+            ),
+            exit_code: base.exit_code,
+        },
+        (ErrorCode::Timeout | ErrorCode::CdpFailed, reason::FILE_INPUT_PROBE_FAILED) => {
+            RenderInfo {
+                summary: "the browser upload transaction could not be established",
+                hint: Some(
+                    "do not retry the same upload blindly; use `bsk request-help` when human help is available",
+                ),
+                exit_code: base.exit_code,
+            }
+        }
+        (ErrorCode::Timeout | ErrorCode::CdpFailed, reason::SET_FILE_INPUT_FAILED) => RenderInfo {
+            summary: "the browser could not attach the staged file to the input",
+            hint: Some(
+                "check that BrowserSkill has Chrome's 'Allow access to file URLs' permission; otherwise use `bsk request-help`",
+            ),
+            exit_code: base.exit_code,
+        },
+        (ErrorCode::Unsupported, reason::UPLOAD_MECHANISM_UNSUPPORTED) => RenderInfo {
+            summary: "the browser does not support native file-drop upload",
+            hint: Some(
+                "do not retry the same drop; use a standard file-input target or `bsk request-help`",
+            ),
+            exit_code: base.exit_code,
+        },
+        (
+            ErrorCode::PermissionDenied | ErrorCode::Timeout | ErrorCode::CdpFailed,
+            reason::FILE_DROP_TARGET_UNAVAILABLE,
+        ) => RenderInfo {
+            summary: "the file-drop target could not be safely resolved",
+            hint: Some(
+                "rerun observe and select the attachment-receiving editor or drop zone itself; do not guess another target — no file was delivered when effect_state is none",
+            ),
+            exit_code: base.exit_code,
+        },
+        (ErrorCode::Timeout | ErrorCode::CdpFailed, reason::FILE_DROP_FAILED) => RenderInfo {
+            summary: "the browser could not complete the native file drop",
+            hint: Some(
+                "do not retry when effect_state is unknown; observe the page first, otherwise use `bsk request-help`",
+            ),
+            exit_code: base.exit_code,
+        },
+        (ErrorCode::CdpFailed, reason::DOWNLOAD_CAPTURE_FAILED) => RenderInfo {
+            summary: "the browser download could not be attributed or completed",
+            hint: Some(
+                "do not retry blindly; the target browser may not expose a reliable download intent signal",
+            ),
+            exit_code: base.exit_code,
+        },
         _ => base,
     }
 }
@@ -354,6 +490,56 @@ mod tests {
         ErrorCode::NoBrowserConnected,
     ];
 
+    #[test]
+    fn fill_reasons_provide_recovery_without_changing_exit_codes() {
+        for (code, reason, hint_fragment) in [
+            (
+                ErrorCode::InvalidParams,
+                reason::TARGET_NOT_FILLABLE,
+                "enabled, editable",
+            ),
+            (
+                ErrorCode::InvalidParams,
+                reason::FILL_VALUE_INVALID,
+                "maxlength",
+            ),
+            (
+                ErrorCode::CdpFailed,
+                reason::FILL_TARGET_CHANGED,
+                "fresh editable target",
+            ),
+            (
+                ErrorCode::CdpFailed,
+                reason::FILL_FOCUS_LOST,
+                "does not require bringing the browser",
+            ),
+            (
+                ErrorCode::CdpFailed,
+                reason::FILL_VALUE_MISMATCH,
+                "Continue if the visible result satisfies",
+            ),
+            (
+                ErrorCode::CdpFailed,
+                reason::FILL_FAILED,
+                "may already have changed",
+            ),
+        ] {
+            let data = serde_json::json!({ "reason": reason });
+            let info = info_for_error(code, Some(&data));
+            assert!(info.hint.unwrap().contains(hint_fragment), "{reason}");
+            assert_eq!(info.exit_code, info_for(code).exit_code);
+        }
+        let mismatch = serde_json::json!({ "reason": reason::FILL_VALUE_MISMATCH });
+        let info = info_for_error(ErrorCode::CdpFailed, Some(&mismatch));
+        assert_eq!(info.summary, "the fill result could not be confirmed");
+        assert!(info.hint.unwrap().contains("Do not blindly repeat fill"));
+        let unknown = serde_json::json!({ "reason": "future_fill_reason" });
+        assert_eq!(
+            info_for_error(ErrorCode::CdpFailed, Some(&unknown)),
+            info_for(ErrorCode::CdpFailed)
+        );
+    }
+
     #[test]
     fn element_not_visible_overrides_permission_denied_copy() {
         let data = serde_json::json!({ "reason": reason::ELEMENT_NOT_VISIBLE });
@@ -366,6 +552,18 @@ mod tests {
         assert!(!info.summary.contains("sandbox"));
     }
 
+    #[test]
+    fn extension_access_denied_explains_the_frame_restriction_and_navigation_recovery() {
+        let data = serde_json::json!({ "reason": reason::CDP_EXTENSION_ACCESS_DENIED });
+        let info = info_for_error(ErrorCode::CdpFailed, Some(&data));
+        assert!(info.summary.contains("another extension"));
+        let hint = info.hint.unwrap();
+        assert!(hint.contains("restricted extension frame"));
+        assert!(hint.contains("bsk navigate <url>"));
+        assert!(hint.contains("disable the conflicting extension"));
+        assert_eq!(info.exit_code, 3);
+    }
+
     #[test]
     fn session_busy_overrides_timeout_copy() {
         let data = serde_json::json!({ "reason": reason::SESSION_BUSY });
@@ -378,6 +576,22 @@ mod tests {
         assert_eq!(info.exit_code, 4);
     }
 
+    #[test]
+    fn record_start_page_unreachable_overrides_cdp_failed_copy() {
+        let data = serde_json::json!({ "reason": reason::RECORD_START_PAGE_UNREACHABLE });
+        let info = info_for_error(ErrorCode::CdpFailed, Some(&data));
+        assert_eq!(
+            info.summary,
+            "default start page https://example.com/ could not be opened"
+        );
+        assert!(
+            info.hint.unwrap().contains("--url"),
+            "expected --url hint, got {:?}",
+            info.hint
+        );
+        assert_eq!(info.exit_code, 3);
+    }
+
     #[test]
     fn screenshot_capture_failed_overrides_cdp_failed_copy() {
         let data = serde_json::json!({ "reason": reason::SCREENSHOT_CAPTURE_FAILED });
@@ -392,6 +606,64 @@ mod tests {
         assert_eq!(info.exit_code, 3);
     }
 
+    #[test]
+    fn file_transfer_reasons_render_actionable_fallbacks() {
+        let unsupported = serde_json::json!({ "reason": reason::FILE_INPUT_NOT_ACTIVATED });
+        let info = info_for_error(ErrorCode::Unsupported, Some(&unsupported));
+        assert_eq!(
+            info.summary,
+            "the upload trigger did not activate a file input"
+        );
+        let hint = info.hint.unwrap();
+        assert!(hint.contains("--mode drop"));
+        assert!(hint.contains("reliably identified attachment-receiving area"));
+        assert!(hint.contains("exact original local path"));
+
+        let control = serde_json::json!({ "reason": reason::FILE_INPUT_PROBE_FAILED });
+        let info = info_for_error(ErrorCode::Timeout, Some(&control));
+        assert!(
+            info.summary
+                .contains("transaction could not be established")
+        );
+        assert!(info.hint.unwrap().contains("do not retry"));
+
+        let unsupported_drop =
+            serde_json::json!({ "reason": reason::UPLOAD_MECHANISM_UNSUPPORTED });
+        let info = info_for_error(ErrorCode::Unsupported, Some(&unsupported_drop));
+        assert_eq!(
+            info.summary,
+            "the browser does not support native file-drop upload"
+        );
+        assert!(info.hint.unwrap().contains("standard file-input"));
+
+        let unavailable_target =
+            serde_json::json!({ "reason": reason::FILE_DROP_TARGET_UNAVAILABLE });
+        let info = info_for_error(ErrorCode::PermissionDenied, Some(&unavailable_target));
+        assert_eq!(
+            info.summary,
+            "the file-drop target could not be safely resolved"
+        );
+        assert!(info.hint.unwrap().contains("effect_state is none"));
+
+        let failed_drop = serde_json::json!({ "reason": reason::FILE_DROP_FAILED });
+        let info = info_for_error(ErrorCode::Timeout, Some(&failed_drop));
+        assert!(info.summary.contains("native file drop"));
+        assert!(info.hint.unwrap().contains("effect_state is unknown"));
+
+        let download = serde_json::json!({ "reason": reason::DOWNLOAD_CAPTURE_FAILED });
+        let info = info_for_error(ErrorCode::CdpFailed, Some(&download));
+        assert!(info.summary.contains("download could not be attributed"));
+
+        let unknown = serde_json::json!({ "reason": reason::TRANSFER_OUTCOME_UNKNOWN });
+        let info = info_for_error(ErrorCode::ProtocolError, Some(&unknown));
+        assert!(info.summary.contains("outcome could not be confirmed"));
+        assert!(info.hint.unwrap().contains("do not retry"));
+
+        let timeout = serde_json::json!({ "reason": reason::TRANSFER_TIMEOUT });
+        let info = info_for_error(ErrorCode::Timeout, Some(&timeout));
+        assert!(info.summary.contains("timed out after browser dispatch"));
+    }
+
     #[test]
     fn ref_not_found_overrides_not_found_copy() {
         let data = serde_json::json!({ "reason": reason::REF_NOT_FOUND });

crates/bsk-cli/src/cli/status.rs

@@ -65,7 +65,7 @@ fn render_human(s: &StatusResult) {
         let bold = "\x1b[1m";
         let reset = "\x1b[0m";
         eprintln!(
-            "{bold}{yellow}warning:{reset} {} browser(s) have protocol version drift from the daemon{reset}",
+            "{bold}{yellow}note:{reset} {} browser(s) report a different protocol version; still usable — continue, and run `bsk update` soon{reset}",
             s.version_skew_browsers.len()
         );
         for skew in &s.version_skew_browsers {
@@ -75,7 +75,7 @@ fn render_human(s: &StatusResult) {
                 skew.label.clone()
             };
             eprintln!(
-                "  {} ({}) — protocol ext {} vs daemon {} (app ext v{}, daemon v{}) — please align protocol versions",
+                "  {} ({}) — protocol ext {} vs CLI {} (app ext v{}, CLI v{}) — still usable; please upgrade",
                 skew.instance_id,
                 label,
                 display_protocol(&skew.client_protocol_version),

crates/bsk-cli/src/cli/update.rs

@@ -6,9 +6,7 @@ use std::path::{Path, PathBuf};
 use std::time::{Duration, SystemTime};
 
 #[cfg(windows)]
-use std::os::windows::process::CommandExt;
-#[cfg(windows)]
-use std::process::{Command, Stdio};
+mod windows;
 
 use anyhow::{Context, Result, bail};
 use clap::Args;
@@ -363,7 +361,8 @@ fn install_candidate_with_client(
         crate::daemon::start::run_stop().context("stop bsk daemon before update")?;
     }
 
-    let action = replace_binary_for_update(&target, &binary, daemon_was_running)?;
+    let restart_args = daemon_was_running.then(StartArgs::default);
+    let action = replace_binary_for_update(&target, &binary, restart_args.as_ref())?;
 
     if daemon_was_running && matches!(action, InstallAction::Replaced) {
         crate::daemon::start::run_start(StartArgs::default())
@@ -394,8 +393,8 @@ pub(crate) fn download_candidate_binary(
 
 /// Daemon-side install: download, verify, and replace the executable at
 /// `target` (on Windows: stage the replacement next to it). Unlike the
-/// CLI path this never stops or starts the daemon — the daemon drives
-/// its own restart once the binary is replaced.
+/// CLI path this never stops the daemon. On Windows the helper starts
+/// its replacement after exit; elsewhere the daemon drives its own restart.
 ///
 /// `target` must be captured *before* any replacement happens: on Linux
 /// `std::env::current_exe` starts returning a ` (deleted)`-suffixed
@@ -403,10 +402,11 @@ pub(crate) fn download_candidate_binary(
 pub(crate) fn self_install_candidate(
     candidate: &UpdateCandidate,
     target: &Path,
+    restart_args: &StartArgs,
 ) -> Result<InstallAction> {
     let client = update_http_client(ARCHIVE_FETCH_TIMEOUT)?;
     let binary = download_candidate_binary(candidate, &client)?;
-    replace_binary_at_path(target, &binary)
+    replace_binary_for_update(target, &binary, Some(restart_args))
 }
 
 /// Outcome of one daemon auto-update step (see [`auto_update_step`]).
@@ -423,8 +423,8 @@ pub(crate) enum AutoUpdateOutcome {
     /// The new binary replaced the old one; the daemon should now
     /// restart into it.
     Replaced { latest: String },
-    /// Windows staged the replacement next to the running binary; a
-    /// daemon/terminal restart finishes it.
+    /// Windows started a helper that will replace the executable and
+    /// restart the daemon once the current process exits.
     Staged { latest: String },
 }
 
@@ -668,6 +668,7 @@ fn render_report(format: Format, report: &UpdateReport) -> Result<()> {
             }
             if matches!(report.install_action, Some("staged")) {
                 println!("the detached update helper will apply the replacement after exit");
+                println!("if it fails, see the *.update-*.log file next to the executable");
             }
         }
         Format::Json => {
@@ -688,22 +689,22 @@ pub fn extract_bsk_binary(archive_bytes: &[u8], kind: ArchiveKind) -> Result<Vec
 }
 
 pub fn replace_binary_at_path(target: &Path, binary: &[u8]) -> Result<InstallAction> {
-    replace_binary_for_update(target, binary, false)
+    replace_binary_for_update(target, binary, None)
 }
 
 fn replace_binary_for_update(
     target: &Path,
     binary: &[u8],
-    restart_daemon: bool,
+    restart_args: Option<&StartArgs>,
 ) -> Result<InstallAction> {
     #[cfg(windows)]
     {
-        stage_windows_replacement(target, binary, restart_daemon)
+        stage_windows_replacement(target, binary, restart_args)
     }
 
     #[cfg(not(windows))]
     {
-        let _ = restart_daemon;
+        let _ = restart_args;
         replace_binary_atomically(target, binary)
     }
 }
@@ -749,33 +750,69 @@ fn replace_binary_atomically(target: &Path, binary: &[u8]) -> Result<InstallActi
 fn stage_windows_replacement(
     target: &Path,
     binary: &[u8],
-    restart_daemon: bool,
+    restart_args: Option<&StartArgs>,
 ) -> Result<InstallAction> {
-    let paths = staged_replacement_paths(target, std::process::id())?;
-    std::fs::write(&paths.binary_path, binary)
-        .with_context(|| format!("write {}", paths.binary_path.display()))?;
-    std::fs::write(
-        &paths.script_path,
-        windows_replacement_script(target, &paths.binary_path, restart_daemon),
-    )
-    .with_context(|| format!("write {}", paths.script_path.display()))?;
+    // The helper owns replacement/restart after this process exits. Do not
+    // report Staged until it has actually begun executing the script.
+    let _child = spawn_windows_replacement(target, binary, restart_args, 120)?;
+    Ok(InstallAction::Staged)
+}
 
-    const CREATE_NO_WINDOW: u32 = 0x0800_0000;
-    let command = format!("\"{}\"", paths.script_path.display());
-    if let Err(err) = Command::new("cmd.exe")
-        .args(["/D", "/S", "/C"])
-        .arg(command)
-        .stdin(Stdio::null())
-        .stdout(Stdio::null())
-        .stderr(Stdio::null())
-        .creation_flags(CREATE_NO_WINDOW)
-        .spawn()
-    {
+#[cfg(windows)]
+fn spawn_windows_replacement(
+    target: &Path,
+    binary: &[u8],
+    restart_args: Option<&StartArgs>,
+    attempts: u32,
+) -> Result<windows::Helper> {
+    let paths = staged_replacement_paths(target, std::process::id())?;
+    let ready_path = paths.script_path.with_extension("ready");
+    let log_path = paths.script_path.with_extension("log");
+    let result = (|| {
+        // A previous failed attempt by this process must not acknowledge a
+        // new helper. All paths are private to this target and process id.
+        match std::fs::remove_file(&ready_path) {
+            Ok(()) => {}
+            Err(err) if err.kind() == std::io::ErrorKind::NotFound => {}
+            Err(err) => return Err(err.into()),
+        }
+        std::fs::write(&paths.binary_path, binary)
+            .with_context(|| format!("write {}", paths.binary_path.display()))?;
+        std::fs::write(
+            &paths.script_path,
+            windows_replacement_script(restart_args, attempts),
+        )
+        .with_context(|| format!("write {}", paths.script_path.display()))?;
+        let mut child = windows::spawn(
+            &paths.script_path,
+            &paths.binary_path,
+            target,
+            &ready_path,
+            &log_path,
+        )
+        .context("launch detached Windows update helper")?;
+        let deadline = std::time::Instant::now() + Duration::from_secs(5);
+        loop {
+            if std::fs::remove_file(&ready_path).is_ok() {
+                return Ok(child);
+            }
+            if let Some(status) = child.try_wait()? {
+                bail!("Windows update helper exited before becoming ready: {status}");
+            }
+            if std::time::Instant::now() >= deadline {
+                let _ = child.kill();
+                let _ = child.wait();
+                bail!("Windows update helper did not become ready within 5 seconds");
+            }
+            std::thread::sleep(Duration::from_millis(20));
+        }
+    })();
+    if result.is_err() {
         let _ = std::fs::remove_file(&paths.binary_path);
         let _ = std::fs::remove_file(&paths.script_path);
-        return Err(err).context("launch detached Windows update helper");
+        let _ = std::fs::remove_file(&ready_path);
     }
-    Ok(InstallAction::Staged)
+    result.with_context(|| format!("Windows update helper failed; see {}", log_path.display()))
 }
 
 pub fn staged_replacement_paths(target: &Path, pid: u32) -> Result<StagedReplacementPaths> {
@@ -792,25 +829,43 @@ pub fn staged_replacement_paths(target: &Path, pid: u32) -> Result<StagedReplace
 }
 
 #[cfg(any(windows, test))]
-fn windows_replacement_script(target: &Path, staged_binary: &Path, restart_daemon: bool) -> String {
-    let target = target.display().to_string().replace('%', "%%");
-    let staged_binary = staged_binary.display().to_string().replace('%', "%%");
-    let restart = if restart_daemon {
-        format!("start \"\" /B \"{target}\" daemon start >nul 2>nul\r\n")
-    } else {
-        String::new()
-    };
+fn windows_replacement_script(restart_args: Option<&StartArgs>, attempts: u32) -> String {
+    // Only fixed switches and numeric values go into the ASCII script. Paths
+    // stay in Unicode environment variables, including %, ! and shell symbols.
+    let restart = restart_args.map_or_else(String::new, |args| {
+        format!(
+            "\"%BSK_UPDATE_TARGET%\" daemon start --port {} --session-idle {}ms --daemon-idle {}ms\r\nif errorlevel 1 goto failed_restart\r\n",
+            args.resolved_port(),
+            args.resolved_session_idle().as_millis(),
+            args.resolved_daemon_idle().as_millis(),
+        )
+    });
     format!(
         "@echo off\r\n\
-         setlocal\r\n\
+         setlocal DisableDelayedExpansion\r\n\
+         > \"%BSK_UPDATE_READY%\" echo ready\r\n\
+         if errorlevel 1 exit /b 1\r\n\
+         set /a attempts=0\r\n\
          :retry\r\n\
-         move /Y \"{staged_binary}\" \"{target}\" >nul 2>nul\r\n\
-         if errorlevel 1 (\r\n\
-           timeout /t 1 /nobreak >nul\r\n\
-           goto retry\r\n\
-         )\r\n\
+         move /Y \"%BSK_UPDATE_SOURCE%\" \"%BSK_UPDATE_TARGET%\" >nul\r\n\
+         if not errorlevel 1 goto replaced\r\n\
+         set /a attempts+=1\r\n\
+         if %attempts% geq {attempts} goto failed_replace\r\n\
+         if not exist \"%BSK_UPDATE_SOURCE%\" goto failed_replace\r\n\
+         ping -n 2 127.0.0.1 >nul\r\n\
+         goto retry\r\n\
+         :failed_replace\r\n\
+         echo Failed to replace the executable after %attempts% attempts.\r\n\
+         goto failed\r\n\
+         :replaced\r\n\
          {restart}\
-         del /F /Q \"%~f0\" >nul 2>nul\r\n"
+         del /F /Q \"%BSK_UPDATE_LOG%\" >nul 2>nul\r\n\
+         (del /F /Q \"%BSK_UPDATE_SCRIPT%\" >nul 2>nul & exit 0)\r\n\
+         :failed_restart\r\n\
+         echo Replaced the executable but failed to restart the daemon.\r\n\
+         :failed\r\n\
+         del /F /Q \"%BSK_UPDATE_SOURCE%\" >nul 2>nul\r\n\
+         (del /F /Q \"%BSK_UPDATE_SCRIPT%\" >nul 2>nul & exit 1)\r\n"
     )
 }
 
@@ -964,6 +1019,7 @@ mod tests {
         assert_eq!(extracted, b"windows binary");
     }
 
+    #[cfg(not(windows))]
     #[test]
     fn replaces_binary_at_path() {
         let tmp = tempfile::TempDir::new().unwrap();
@@ -998,29 +1054,139 @@ mod tests {
     }
 
     #[test]
-    fn windows_replacement_script_retries_restarts_and_cleans_itself() {
-        let script = windows_replacement_script(
-            Path::new(r"C:\Program Files\bsk.exe"),
-            Path::new(r"C:\Program Files\bsk.exe.update-42"),
-            true,
+    fn windows_replacement_script_preserves_restart_config() {
+        let args = StartArgs {
+            port: Some(54321),
+            session_idle: Some(Duration::from_millis(1234)),
+            daemon_idle: Some(Duration::from_secs(75)),
+            ..Default::default()
+        };
+        let script = windows_replacement_script(Some(&args), 120);
+        assert!(script.is_ascii());
+        assert!(
+            script
+                .contains("daemon start --port 54321 --session-idle 1234ms --daemon-idle 75000ms")
         );
-
-        assert!(script.contains(":retry"));
-        assert!(script.contains("move /Y"));
-        assert!(script.contains(r#"start "" /B "C:\Program Files\bsk.exe" daemon start"#));
-        assert!(script.contains(r#"del /F /Q "%~f0""#));
+        assert!(!windows_replacement_script(None, 120).contains("daemon start"));
     }
 
-    #[test]
-    fn windows_replacement_script_omits_restart_when_daemon_was_not_running() {
-        let script = windows_replacement_script(
-            Path::new(r"C:\bsk.exe"),
-            Path::new(r"C:\bsk.exe.update-42"),
-            false,
-        );
+    #[cfg(windows)]
+    fn wait_for_helper(mut child: windows::Helper) -> std::process::ExitStatus {
+        let deadline = std::time::Instant::now() + Duration::from_secs(10);
+        loop {
+            if let Some(status) = child.try_wait().unwrap() {
+                return status;
+            }
+            if std::time::Instant::now() >= deadline {
+                let _ = child.kill();
+                let _ = child.wait();
+                panic!("Windows update helper did not exit");
+            }
+            std::thread::sleep(Duration::from_millis(20));
+        }
+    }
 
-        assert!(!script.contains("daemon start"));
-        assert!(script.contains(r#"del /F /Q "%~f0""#));
+    #[cfg(windows)]
+    #[test]
+    fn windows_helper_replaces_in_unicode_spaces_and_shell_symbol_paths() {
+        let tmp = tempfile::TempDir::new().unwrap();
+        for name in [
+            "ascii",
+            "with space",
+            "中文目录",
+            "literal %PATH% ! & (folder)",
+        ] {
+            let dir = tmp.path().join(name);
+            std::fs::create_dir(&dir).unwrap();
+            let target = dir.join("bsk.exe");
+            std::fs::write(&target, b"old binary").unwrap();
+            let child = spawn_windows_replacement(&target, b"new binary", None, 5).unwrap();
+            let paths = staged_replacement_paths(&target, std::process::id()).unwrap();
+            let status = wait_for_helper(child);
+            assert!(
+                status.success(),
+                "{name}: {:?}",
+                std::fs::read_to_string(paths.script_path.with_extension("log"))
+            );
+            assert_eq!(std::fs::read(&target).unwrap(), b"new binary", "{name}");
+            // Successful updates leave no helper, staged binary, ready file or log.
+            assert_eq!(std::fs::read_dir(&dir).unwrap().count(), 1, "{name}");
+        }
+    }
+
+    #[cfg(windows)]
+    #[test]
+    fn windows_helper_waits_for_unlock_then_replaces() {
+        use std::os::windows::fs::OpenOptionsExt;
+        let tmp = tempfile::TempDir::new().unwrap();
+        let target = tmp.path().join("bsk.exe");
+        std::fs::write(&target, b"old binary").unwrap();
+        // Deny delete sharing, like a running Windows executable.
+        let lock = std::fs::OpenOptions::new()
+            .read(true)
+            .share_mode(1)
+            .open(&target)
+            .unwrap();
+        let mut child = spawn_windows_replacement(&target, b"new binary", None, 5).unwrap();
+        std::thread::sleep(Duration::from_millis(150));
+        assert!(child.try_wait().unwrap().is_none());
+        assert_eq!(std::fs::read(&target).unwrap(), b"old binary");
+        drop(lock);
+        assert!(wait_for_helper(child).success());
+        assert_eq!(std::fs::read(&target).unwrap(), b"new binary");
+    }
+
+    #[cfg(windows)]
+    #[test]
+    fn windows_helper_bounds_retries_and_preserves_old_binary_on_failure() {
+        use std::os::windows::fs::OpenOptionsExt;
+        let tmp = tempfile::TempDir::new().unwrap();
+        let target = tmp.path().join("bsk.exe");
+        std::fs::write(&target, b"old binary").unwrap();
+        let _lock = std::fs::OpenOptions::new()
+            .read(true)
+            .share_mode(1)
+            .open(&target)
+            .unwrap();
+        let child = spawn_windows_replacement(&target, b"new binary", None, 2).unwrap();
+        assert!(!wait_for_helper(child).success());
+        let paths = staged_replacement_paths(&target, std::process::id()).unwrap();
+        assert_eq!(std::fs::read(&target).unwrap(), b"old binary");
+        assert!(!paths.binary_path.exists());
+        assert!(!paths.script_path.exists());
+        let log_bytes = std::fs::read(paths.script_path.with_extension("log")).unwrap();
+        let log = String::from_utf8_lossy(&log_bytes);
+        assert!(
+            log.contains("Failed to replace the executable after 2 attempts."),
+            "{log}"
+        );
+    }
+
+    #[cfg(windows)]
+    #[test]
+    fn windows_helper_reports_restart_failure_without_losing_replacement() {
+        let tmp = tempfile::TempDir::new().unwrap();
+        let target = tmp.path().join("bsk.exe");
+        std::fs::write(&target, b"old binary").unwrap();
+        // Not a valid PE executable: replacement succeeds but restart fails.
+        let child = spawn_windows_replacement(
+            &target,
+            b"invalid executable",
+            Some(&StartArgs::default()),
+            2,
+        )
+        .unwrap();
+        assert!(!wait_for_helper(child).success());
+        let paths = staged_replacement_paths(&target, std::process::id()).unwrap();
+        let bytes = std::fs::read(paths.script_path.with_extension("log")).unwrap();
+        let log = String::from_utf8_lossy(&bytes);
+        assert!(
+            log.contains("Replaced the executable but failed to restart the daemon."),
+            "{log}"
+        );
+        assert_eq!(std::fs::read(&target).unwrap(), b"invalid executable");
+        assert!(!paths.script_path.exists());
+        assert!(!paths.binary_path.exists());
     }
 
     #[test]
@@ -1279,10 +1445,9 @@ mod tests {
     }
 
     #[test]
-    fn auto_update_step_staged_outcome_requests_no_restart() {
-        // The Windows shape: the replacement is staged next to the
-        // running binary, so the outcome must not ask for the immediate
-        // self-restart that `Replaced` triggers.
+    fn auto_update_step_reports_staged_handoff() {
+        // The helper, rather than this process, starts the new executable.
+        // The daemon must exit to let that staged handoff finish.
         let candidate = test_candidate();
         let outcome =
             auto_update_step(Some(&candidate), true, 0, |_| Ok(InstallAction::Staged)).unwrap();

crates/bsk-cli/src/cli/update/windows.rs

@@ -0,0 +1,203 @@
+//! Launch the update helper with only its stdio handles inherited.
+//!
+//! std::process::Command on Windows also inherits other inheritable handles.
+//! A daemon's listening socket must not survive in the helper: it would keep
+//! the port occupied after the daemon exits and prevent its replacement starting.
+
+use std::ffi::OsStr;
+use std::fs::File;
+use std::io;
+use std::os::windows::ffi::OsStrExt;
+use std::os::windows::io::{AsRawHandle, FromRawHandle, OwnedHandle};
+use std::os::windows::process::ExitStatusExt;
+use std::path::Path;
+use std::process::ExitStatus;
+
+use windows_sys::Win32::Foundation::{
+    HANDLE, HANDLE_FLAG_INHERIT, SetHandleInformation, WAIT_OBJECT_0, WAIT_TIMEOUT,
+};
+use windows_sys::Win32::System::Threading::{
+    CREATE_NEW_PROCESS_GROUP, CREATE_NO_WINDOW, CREATE_UNICODE_ENVIRONMENT, CreateProcessW,
+    DeleteProcThreadAttributeList, EXTENDED_STARTUPINFO_PRESENT, GetExitCodeProcess,
+    InitializeProcThreadAttributeList, PROC_THREAD_ATTRIBUTE_HANDLE_LIST, PROCESS_INFORMATION,
+    STARTF_USESTDHANDLES, STARTUPINFOEXW, TerminateProcess, UpdateProcThreadAttribute,
+    WaitForSingleObject,
+};
+
+pub(super) struct Helper(OwnedHandle);
+
+impl Helper {
+    pub(super) fn try_wait(&mut self) -> io::Result<Option<ExitStatus>> {
+        // SAFETY: the owned process handle remains valid for both calls.
+        match unsafe { WaitForSingleObject(self.0.as_raw_handle(), 0) } {
+            WAIT_OBJECT_0 => {
+                let mut code = 0;
+                if unsafe { GetExitCodeProcess(self.0.as_raw_handle(), &mut code) } == 0 {
+                    return Err(io::Error::last_os_error());
+                }
+                Ok(Some(ExitStatus::from_raw(code)))
+            }
+            WAIT_TIMEOUT => Ok(None),
+            _ => Err(io::Error::last_os_error()),
+        }
+    }
+
+    pub(super) fn kill(&mut self) -> io::Result<()> {
+        // SAFETY: this handle belongs to the helper we created.
+        if unsafe { TerminateProcess(self.0.as_raw_handle(), 1) } == 0 {
+            return Err(io::Error::last_os_error());
+        }
+        Ok(())
+    }
+
+    pub(super) fn wait(&mut self) -> io::Result<()> {
+        // Only used after kill(); do not leave a live helper on startup failure.
+        if unsafe { WaitForSingleObject(self.0.as_raw_handle(), 5000) } != WAIT_OBJECT_0 {
+            return Err(io::Error::new(
+                io::ErrorKind::TimedOut,
+                "update helper did not exit",
+            ));
+        }
+        Ok(())
+    }
+}
+
+// The opaque attribute list needs pointer-aligned storage and explicit teardown.
+struct AttributeList(Vec<usize>);
+
+impl Drop for AttributeList {
+    fn drop(&mut self) {
+        // SAFETY: this wrapper is constructed only after successful initialization.
+        unsafe { DeleteProcThreadAttributeList(self.0.as_mut_ptr().cast()) };
+    }
+}
+
+pub(super) fn spawn(
+    script: &Path,
+    source: &Path,
+    target: &Path,
+    ready: &Path,
+    log_path: &Path,
+) -> io::Result<Helper> {
+    let root = std::env::var_os("SystemRoot")
+        .ok_or_else(|| io::Error::new(io::ErrorKind::NotFound, "SystemRoot is not set"))?;
+    let application = wide(Path::new(&root).join("System32/cmd.exe").as_os_str());
+    // /S /C strips one outer pair of quotes. Environment expansion keeps paths
+    // Unicode and avoids CRT escaping, batch-file encoding, and CALL expansion.
+    let mut command = wide(OsStr::new(
+        "cmd.exe /D /V:OFF /S /C \"\"%BSK_UPDATE_SCRIPT%\"\"",
+    ));
+    let overrides = [
+        ("BSK_UPDATE_SCRIPT", script.as_os_str()),
+        ("BSK_UPDATE_SOURCE", source.as_os_str()),
+        ("BSK_UPDATE_TARGET", target.as_os_str()),
+        ("BSK_UPDATE_READY", ready.as_os_str()),
+        ("BSK_UPDATE_LOG", log_path.as_os_str()),
+    ];
+    let mut env: Vec<_> = std::env::vars_os()
+        .filter(|(key, _)| {
+            let key = key.to_string_lossy();
+            !overrides
+                .iter()
+                .any(|(name, _)| key.eq_ignore_ascii_case(name))
+                && !key.eq_ignore_ascii_case(crate::daemon::start::DAEMONIZED_ENV)
+                && !key.eq_ignore_ascii_case(crate::daemon::start::DAEMON_REPLACEMENT_WAIT_ENV)
+        })
+        .collect();
+    env.extend(
+        overrides
+            .into_iter()
+            .map(|(key, value)| (key.into(), value.to_owned())),
+    );
+    env.sort_by_key(|(key, _)| key.to_string_lossy().to_uppercase());
+    let mut environment = Vec::new();
+    for (key, value) in env {
+        environment.extend(key.encode_wide());
+        environment.push(b'=' as u16);
+        environment.extend(value.encode_wide());
+        environment.push(0);
+    }
+    environment.push(0);
+
+    let input = File::open("NUL")?;
+    let log = File::create(log_path)?;
+    let handles: [HANDLE; 2] = [input.as_raw_handle(), log.as_raw_handle()];
+    for &handle in &handles {
+        // SAFETY: these are dedicated helper stdio handles, held until spawn
+        // finishes. The explicit handle list excludes every other parent handle.
+        if unsafe { SetHandleInformation(handle, HANDLE_FLAG_INHERIT, HANDLE_FLAG_INHERIT) } == 0 {
+            return Err(io::Error::last_os_error());
+        }
+    }
+
+    let mut size = 0;
+    // SAFETY: the first call queries the allocation size, as required by Win32.
+    unsafe { InitializeProcThreadAttributeList(std::ptr::null_mut(), 1, 0, &mut size) };
+    if size == 0 {
+        return Err(io::Error::last_os_error());
+    }
+    let mut storage = vec![0usize; size.div_ceil(std::mem::size_of::<usize>())];
+    // SAFETY: storage has the requested size and pointer alignment.
+    if unsafe { InitializeProcThreadAttributeList(storage.as_mut_ptr().cast(), 1, 0, &mut size) }
+        == 0
+    {
+        return Err(io::Error::last_os_error());
+    }
+    let mut attributes = AttributeList(storage);
+    // SAFETY: the handle array and attribute storage outlive CreateProcessW.
+    if unsafe {
+        UpdateProcThreadAttribute(
+            attributes.0.as_mut_ptr().cast(),
+            0,
+            PROC_THREAD_ATTRIBUTE_HANDLE_LIST as usize,
+            handles.as_ptr().cast(),
+            std::mem::size_of_val(&handles),
+            std::ptr::null_mut(),
+            std::ptr::null(),
+        )
+    } == 0
+    {
+        return Err(io::Error::last_os_error());
+    }
+    // SAFETY: these Win32 structs permit zero initialization; required fields
+    // are populated below before the process is created.
+    let mut startup: STARTUPINFOEXW = unsafe { std::mem::zeroed() };
+    startup.StartupInfo.cb = std::mem::size_of::<STARTUPINFOEXW>() as u32;
+    startup.StartupInfo.dwFlags = STARTF_USESTDHANDLES;
+    startup.StartupInfo.hStdInput = handles[0];
+    startup.StartupInfo.hStdOutput = handles[1];
+    startup.StartupInfo.hStdError = handles[1];
+    startup.lpAttributeList = attributes.0.as_mut_ptr().cast();
+    let mut process: PROCESS_INFORMATION = unsafe { std::mem::zeroed() };
+    // SAFETY: strings are NUL-terminated UTF-16, the environment is double-NUL
+    // terminated, and all buffers/handles remain alive for the duration of spawn.
+    if unsafe {
+        CreateProcessW(
+            application.as_ptr(),
+            command.as_mut_ptr(),
+            std::ptr::null(),
+            std::ptr::null(),
+            1,
+            CREATE_NO_WINDOW
+                | CREATE_NEW_PROCESS_GROUP
+                | CREATE_UNICODE_ENVIRONMENT
+                | EXTENDED_STARTUPINFO_PRESENT,
+            environment.as_ptr().cast(),
+            std::ptr::null(),
+            &startup.StartupInfo,
+            &mut process,
+        )
+    } == 0
+    {
+        return Err(io::Error::last_os_error());
+    }
+    // SAFETY: successful CreateProcessW transfers these two handles to us.
+    let _thread = unsafe { OwnedHandle::from_raw_handle(process.hThread) };
+    Ok(Helper(unsafe {
+        OwnedHandle::from_raw_handle(process.hProcess)
+    }))
+}
+
+fn wide(value: &OsStr) -> Vec<u16> {
+    value.encode_wide().chain(Some(0)).collect()
+}

crates/bsk-cli/src/cli/upload.rs

@@ -0,0 +1,202 @@
+//! `bsk upload` — stage caller-readable files and attach them to a page.
+
+use std::fs::File;
+use std::io::Read;
+use std::path::{Path, PathBuf};
+use std::time::Duration;
+
+use anyhow::Context;
+use base64::Engine;
+use bsk_protocol::Method;
+use bsk_protocol::tools::{
+    TransferBeginParams, TransferBeginResult, TransferChunkParams, TransferChunkResult,
+    TransferIdParams, TransferReadyResult, UploadFile, UploadMode, UploadParams, UploadResult,
+};
+use clap::{Args, ValueEnum};
+
+use crate::cli::ensure_daemon::ensure_daemon;
+use crate::cli::error::{CliError, Format};
+use crate::cli::interaction::split_target;
+use crate::cli::navigate::parse_timeout_ms;
+
+#[derive(Debug, Default, Clone, Copy, PartialEq, Eq, ValueEnum)]
+pub enum UploadModeArg {
+    #[default]
+    Input,
+    Drop,
+}
+
+impl From<UploadModeArg> for UploadMode {
+    fn from(value: UploadModeArg) -> Self {
+        match value {
+            UploadModeArg::Input => Self::Input,
+            UploadModeArg::Drop => Self::Drop,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Args)]
+pub struct UploadArgs {
+    /// Snapshot ref (`@e3`) or CSS selector for the chooser trigger / drop target.
+    pub target: Option<String>,
+    #[arg(long = "ref")]
+    pub ref_: Option<String>,
+    #[arg(long = "selector")]
+    pub selector: Option<String>,
+    /// Local file to upload. Repeat for a multiple-file input.
+    #[arg(long = "file", required = true)]
+    pub files: Vec<PathBuf>,
+    /// Browser mechanism used to deliver the staged files.
+    #[arg(long, value_enum, default_value_t)]
+    pub mode: UploadModeArg,
+    #[arg(long)]
+    pub session: String,
+    #[arg(long = "tab-id")]
+    pub tab_id: Option<i64>,
+    #[arg(long, default_value = "2m", value_parser = parse_timeout_ms)]
+    pub timeout: u32,
+}
+
+pub fn dispatch(args: UploadArgs, format: Format) -> Result<(), CliError> {
+    if args.files.len() > crate::daemon::file_transfer::MAX_UPLOAD_FILES {
+        return Err(CliError::Local(anyhow::anyhow!(
+            "upload accepts at most {} files",
+            crate::daemon::file_transfer::MAX_UPLOAD_FILES
+        )));
+    }
+    let info = ensure_daemon().context("ensure daemon is running")?;
+    let (ref_, selector) = split_target(args.target, args.ref_, args.selector)?;
+    let mut staged = Vec::new();
+    for path in &args.files {
+        if let Err(err) = stage_file(&info.sock_path, &args.session, path, &mut staged) {
+            let _ = crate::cli::business_rpc::release_transfers(
+                &info.sock_path,
+                staged.iter().map(|(id, _)| id.as_str()),
+            );
+            return Err(err);
+        }
+    }
+    let params = UploadParams {
+        session_id: args.session,
+        ref_,
+        selector,
+        tab_id: args.tab_id,
+        files: staged
+            .iter()
+            .map(|(id, name)| UploadFile {
+                transfer_id: id.clone(),
+                name: name.clone(),
+                staged_path: None,
+            })
+            .collect(),
+        mode: args.mode.into(),
+        timeout_ms: Some(args.timeout),
+    };
+    let result = crate::cli::business_rpc::call::<_, UploadResult>(
+        info.sock_path.clone(),
+        "upload",
+        Method::ToolUpload,
+        Some(params),
+        ipc_timeout(args.timeout),
+    );
+    // The helper can reject cancellation before handing any request to the
+    // daemon. In that case staging still belongs to this CLI.
+    if result.as_ref().is_err_and(|err| {
+        err.data()
+            .is_some_and(|data| data["reason"] == "cancelled_before_dispatch")
+    }) {
+        let _ = crate::cli::business_rpc::release_transfers(
+            &info.sock_path,
+            staged.iter().map(|(id, _)| id.as_str()),
+        );
+    }
+    // Once tool.upload is dispatched, staging ownership belongs to the
+    // session. A transport timeout cannot prove that Chrome did not attach
+    // the file, so releasing here could invalidate a late successful attach.
+    // Session teardown remains the single cleanup boundary after dispatch.
+    let reply = result?;
+    match format {
+        Format::Json => println!("{}", serde_json::to_string_pretty(&reply).unwrap()),
+        Format::Human => println!(
+            "upload ok tab={} files={}",
+            reply.tab_id,
+            reply.file_names.join(", ")
+        ),
+    }
+    Ok(())
+}
+
+fn stage_file(
+    sock: &Path,
+    session: &str,
+    path: &PathBuf,
+    staged: &mut Vec<(String, String)>,
+) -> Result<(), CliError> {
+    let mut file = File::open(path)
+        .with_context(|| format!("open upload file {}", path.display()))
+        .map_err(CliError::Local)?;
+    let meta = file
+        .metadata()
+        .with_context(|| format!("read upload file metadata {}", path.display()))
+        .map_err(CliError::Local)?;
+    if !meta.is_file() {
+        return Err(CliError::Local(anyhow::anyhow!(
+            "upload source is not a regular file: {}",
+            path.display()
+        )));
+    }
+    let name = path
+        .file_name()
+        .and_then(|v| v.to_str())
+        .filter(|v| !v.is_empty())
+        .ok_or_else(|| CliError::Local(anyhow::anyhow!("upload file has no valid basename")))?
+        .to_string();
+    let begin: TransferBeginResult = crate::cli::business_rpc::call(
+        sock.to_path_buf(),
+        "transfer-begin",
+        Method::TransferBegin,
+        Some(TransferBeginParams {
+            session_id: session.to_string(),
+            name: name.clone(),
+            byte_size: meta.len(),
+        }),
+        Duration::from_secs(10),
+    )?;
+    // Record ownership immediately so one error path releases both completed
+    // files and the current partially staged file.
+    staged.push((begin.transfer_id.clone(), name));
+    let mut offset = 0u64;
+    let mut buf = vec![0u8; begin.chunk_size as usize];
+    loop {
+        let n = file.read(&mut buf).map_err(|e| CliError::Local(e.into()))?;
+        if n == 0 {
+            break;
+        }
+        let reply: TransferChunkResult = crate::cli::business_rpc::call(
+            sock.to_path_buf(),
+            "transfer-chunk",
+            Method::TransferChunk,
+            Some(TransferChunkParams {
+                transfer_id: begin.transfer_id.clone(),
+                offset,
+                data_base64: base64::engine::general_purpose::STANDARD.encode(&buf[..n]),
+            }),
+            Duration::from_secs(30),
+        )?;
+        offset = reply.next_offset;
+    }
+    let _: TransferReadyResult = crate::cli::business_rpc::call(
+        sock.to_path_buf(),
+        "transfer-finish",
+        Method::TransferFinish,
+        Some(TransferIdParams {
+            transfer_id: begin.transfer_id.clone(),
+        }),
+        Duration::from_secs(10),
+    )?;
+    Ok(())
+}
+
+fn ipc_timeout(timeout_ms: u32) -> Duration {
+    Duration::from_millis(u64::from(timeout_ms) + 5_000)
+}

crates/bsk-cli/src/daemon/file_transfer.rs

@@ -0,0 +1,728 @@
+//! Session-scoped, content-transparent staging for upload/download tools.
+//!
+//! The registry deliberately exposes no arbitrary-path operations. Upload
+//! bytes arrive from the invoking CLI in bounded chunks; browser downloads
+//! land in a daemon-minted directory and are read back by the invoking CLI.
+
+use std::collections::HashMap;
+use std::fs::{self, File, OpenOptions};
+use std::io::{Read, Seek, SeekFrom, Write};
+use std::path::{Component, Path, PathBuf};
+use std::sync::Mutex;
+
+use base64::Engine;
+use bsk_protocol::tools::{
+    TransferBeginParams, TransferBeginResult, TransferChunkParams, TransferChunkResult,
+    TransferIdParams, TransferReadyResult, TransferReleaseResult,
+};
+use bsk_protocol::{ErrorCode, RpcError};
+use uuid::Uuid;
+
+use super::paths;
+
+pub const TRANSFER_CHUNK_SIZE: u32 = 512 * 1024;
+pub const MAX_TRANSFER_BYTES: u64 = 512 * 1024 * 1024;
+pub const MAX_UPLOAD_FILES: usize = 20;
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+enum Direction {
+    Upload,
+    Download,
+}
+
+#[derive(Debug)]
+struct Entry {
+    session_id: String,
+    direction: Direction,
+    path: PathBuf,
+    expected_size: Option<u64>,
+    written: u64,
+    ready: bool,
+}
+
+/// Browser download path after the daemon has verified that it belongs to
+/// the capability minted for this transfer. From this point onward the daemon
+/// owns cleanup on every error path; unvalidated arbitrary paths are never
+/// removed.
+#[derive(Debug)]
+struct ValidatedBrowserFile {
+    path: PathBuf,
+    parent: PathBuf,
+    cleanup_armed: bool,
+}
+
+impl ValidatedBrowserFile {
+    fn validate(reported_path: &Path, transfer_id: &str) -> Result<Self, RpcError> {
+        if !reported_path.is_absolute() {
+            return Err(permission(
+                "download path is outside its browser capability",
+            ));
+        }
+        let expected_parent = Path::new("BrowserSkill").join(transfer_id);
+        let parent = reported_path
+            .parent()
+            .ok_or_else(|| permission("download path has no parent directory"))?;
+        if !parent.ends_with(&expected_parent) {
+            return Err(permission(
+                "download escaped its browser capability directory",
+            ));
+        }
+        for path in [reported_path, parent] {
+            if fs::symlink_metadata(path)
+                .map_err(io_error)?
+                .file_type()
+                .is_symlink()
+            {
+                return Err(permission("download capability path contains a symlink"));
+            }
+        }
+        Ok(Self {
+            path: reported_path.to_path_buf(),
+            parent: parent.to_path_buf(),
+            cleanup_armed: true,
+        })
+    }
+
+    fn remove_source(mut self) -> Result<(), RpcError> {
+        remove_if_present(&self.path).map_err(io_error)?;
+        remove_dir_if_present(&self.parent).map_err(io_error)?;
+        self.cleanup_armed = false;
+        Ok(())
+    }
+}
+
+impl Drop for ValidatedBrowserFile {
+    fn drop(&mut self) {
+        if !self.cleanup_armed {
+            return;
+        }
+        let _ = remove_if_present(&self.path);
+        let _ = remove_dir_if_present(&self.parent);
+    }
+}
+
+#[derive(Debug)]
+pub struct DownloadStaging {
+    pub transfer_id: String,
+    pub browser_relative_dir: String,
+}
+
+#[derive(Debug)]
+pub struct TransferRegistry {
+    root: PathBuf,
+    initialized: Mutex<bool>,
+    entries: Mutex<HashMap<String, Entry>>,
+}
+
+impl TransferRegistry {
+    pub fn new() -> anyhow::Result<Self> {
+        Ok(Self {
+            root: paths::bsk_home()?.join("run").join("transfers"),
+            initialized: Mutex::new(false),
+            entries: Mutex::new(HashMap::new()),
+        })
+    }
+
+    #[cfg(test)]
+    fn at_root(root: PathBuf) -> anyhow::Result<Self> {
+        let registry = Self {
+            root,
+            initialized: Mutex::new(false),
+            entries: Mutex::new(HashMap::new()),
+        };
+        registry.ensure_root()?;
+        Ok(registry)
+    }
+
+    pub fn initialize(&self) -> anyhow::Result<()> {
+        self.ensure_root()
+    }
+
+    fn ensure_root(&self) -> anyhow::Result<()> {
+        let mut initialized = self.initialized.lock().unwrap();
+        if *initialized {
+            return Ok(());
+        }
+        if self.root.exists() {
+            // Transfers never survive a daemon; removing stale directories on
+            // startup gives crash cleanup without a second persistence model.
+            let _ = fs::remove_dir_all(&self.root);
+        }
+        fs::create_dir_all(&self.root)?;
+        #[cfg(unix)]
+        {
+            use std::os::unix::fs::PermissionsExt;
+            fs::set_permissions(&self.root, fs::Permissions::from_mode(0o700))?;
+        }
+        *initialized = true;
+        Ok(())
+    }
+
+    pub fn begin_upload(&self, p: TransferBeginParams) -> Result<TransferBeginResult, RpcError> {
+        self.ensure_root()
+            .map_err(|err| protocol_error(err.to_string()))?;
+        let basename = safe_upload_basename(&p.name)?;
+        if p.byte_size > MAX_TRANSFER_BYTES {
+            return Err(invalid(format!(
+                "file size {} exceeds transfer limit {}",
+                p.byte_size, MAX_TRANSFER_BYTES
+            )));
+        }
+        let mut entries = self.entries.lock().unwrap();
+        let session_bytes: u64 = entries
+            .values()
+            .filter(|entry| entry.session_id == p.session_id)
+            .map(|entry| entry.expected_size.unwrap_or(entry.written))
+            .sum();
+        if session_bytes.saturating_add(p.byte_size) > MAX_TRANSFER_BYTES {
+            return Err(invalid(format!(
+                "session transfer staging exceeds limit {}",
+                MAX_TRANSFER_BYTES
+            )));
+        }
+        let id = new_id();
+        let dir = self.root.join(&id);
+        fs::create_dir(&dir).map_err(io_error)?;
+        set_private_dir(&dir).map_err(io_error)?;
+        let path = dir.join(basename);
+        let file = OpenOptions::new()
+            .write(true)
+            .create_new(true)
+            .open(&path)
+            .map_err(io_error)?;
+        set_private_file(&file).map_err(io_error)?;
+        entries.insert(
+            id.clone(),
+            Entry {
+                session_id: p.session_id,
+                direction: Direction::Upload,
+                path,
+                expected_size: Some(p.byte_size),
+                written: 0,
+                ready: false,
+            },
+        );
+        Ok(TransferBeginResult {
+            transfer_id: id,
+            chunk_size: TRANSFER_CHUNK_SIZE,
+        })
+    }
+
+    pub fn write_chunk(&self, p: TransferChunkParams) -> Result<TransferChunkResult, RpcError> {
+        let bytes = base64::engine::general_purpose::STANDARD
+            .decode(p.data_base64)
+            .map_err(|e| invalid(format!("invalid transfer chunk: {e}")))?;
+        if bytes.len() > TRANSFER_CHUNK_SIZE as usize {
+            return Err(invalid("transfer chunk exceeds negotiated size"));
+        }
+        let mut entries = self.entries.lock().unwrap();
+        let entry = entries
+            .get_mut(&p.transfer_id)
+            .ok_or_else(|| not_found("transfer not found"))?;
+        if entry.direction != Direction::Upload || entry.ready {
+            return Err(invalid("transfer is not writable"));
+        }
+        if p.offset != entry.written {
+            return Err(invalid(format!(
+                "non-sequential transfer offset: expected {}, got {}",
+                entry.written, p.offset
+            )));
+        }
+        let next = entry.written.saturating_add(bytes.len() as u64);
+        if next > entry.expected_size.unwrap_or(MAX_TRANSFER_BYTES) {
+            return Err(invalid("transfer exceeds declared byte size"));
+        }
+        let mut file = OpenOptions::new()
+            .append(true)
+            .open(&entry.path)
+            .map_err(io_error)?;
+        file.write_all(&bytes).map_err(io_error)?;
+        entry.written = next;
+        Ok(TransferChunkResult {
+            next_offset: next,
+            eof: false,
+            data_base64: None,
+        })
+    }
+
+    pub fn finish_upload(&self, p: TransferIdParams) -> Result<TransferReadyResult, RpcError> {
+        let mut entries = self.entries.lock().unwrap();
+        let entry = entries
+            .get_mut(&p.transfer_id)
+            .ok_or_else(|| not_found("transfer not found"))?;
+        if entry.direction != Direction::Upload {
+            return Err(invalid("transfer is not an upload"));
+        }
+        if entry.written != entry.expected_size.unwrap_or(entry.written) {
+            return Err(invalid(format!(
+                "incomplete transfer: expected {} bytes, received {}",
+                entry.expected_size.unwrap_or(0),
+                entry.written
+            )));
+        }
+        entry.ready = true;
+        Ok(TransferReadyResult {
+            transfer_id: p.transfer_id,
+            byte_size: entry.written,
+        })
+    }
+
+    pub fn resolve_uploads(
+        &self,
+        session_id: &str,
+        ids: &[String],
+    ) -> Result<Vec<PathBuf>, RpcError> {
+        let entries = self.entries.lock().unwrap();
+        ids.iter()
+            .map(|id| {
+                let entry = entries
+                    .get(id)
+                    .ok_or_else(|| not_found("upload transfer not found"))?;
+                if entry.session_id != session_id
+                    || entry.direction != Direction::Upload
+                    || !entry.ready
+                {
+                    return Err(permission(
+                        "upload transfer is outside this session or not ready",
+                    ));
+                }
+                Ok(entry.path.clone())
+            })
+            .collect()
+    }
+
+    pub fn begin_download(&self, session_id: &str) -> Result<DownloadStaging, RpcError> {
+        self.ensure_root()
+            .map_err(|err| protocol_error(err.to_string()))?;
+        let id = new_id();
+        let dir = self.root.join(&id);
+        fs::create_dir(&dir).map_err(io_error)?;
+        set_private_dir(&dir).map_err(io_error)?;
+        self.entries.lock().unwrap().insert(
+            id.clone(),
+            Entry {
+                session_id: session_id.to_string(),
+                direction: Direction::Download,
+                path: dir.clone(),
+                expected_size: None,
+                written: 0,
+                ready: false,
+            },
+        );
+        Ok(DownloadStaging {
+            browser_relative_dir: format!("BrowserSkill/{id}"),
+            transfer_id: id,
+        })
+    }
+
+    pub fn import_download(&self, id: &str, reported_path: &Path) -> Result<u64, RpcError> {
+        let mut entries = self.entries.lock().unwrap();
+        let entry = entries
+            .get_mut(id)
+            .ok_or_else(|| not_found("download transfer not found"))?;
+        if entry.direction != Direction::Download {
+            return Err(permission("transfer is not a download capability"));
+        }
+        let browser_file = ValidatedBrowserFile::validate(reported_path, id)?;
+        let canonical = browser_file.path.canonicalize().map_err(io_error)?;
+        let meta = fs::metadata(&canonical).map_err(io_error)?;
+        if !meta.is_file() || meta.len() > MAX_TRANSFER_BYTES {
+            return Err(invalid(
+                "download is not a regular file or exceeds the transfer limit",
+            ));
+        }
+
+        let destination = entry.path.join("payload");
+        let source = File::open(&canonical).map_err(io_error)?;
+        let mut destination_file = OpenOptions::new()
+            .write(true)
+            .create_new(true)
+            .open(&destination)
+            .map_err(io_error)?;
+        let copied = std::io::copy(
+            &mut source.take(MAX_TRANSFER_BYTES.saturating_add(1)),
+            &mut destination_file,
+        )
+        .map_err(io_error)?;
+        if copied > MAX_TRANSFER_BYTES {
+            let _ = fs::remove_file(&destination);
+            return Err(invalid("download exceeds the transfer limit"));
+        }
+        entry.path = destination;
+        entry.written = copied;
+        entry.expected_size = Some(copied);
+        entry.ready = true;
+        browser_file.remove_source()?;
+        Ok(copied)
+    }
+
+    pub fn read_chunk(&self, p: TransferChunkParams) -> Result<TransferChunkResult, RpcError> {
+        let entries = self.entries.lock().unwrap();
+        let entry = entries
+            .get(&p.transfer_id)
+            .ok_or_else(|| not_found("transfer not found"))?;
+        if entry.direction != Direction::Download || !entry.ready {
+            return Err(invalid("transfer is not readable"));
+        }
+        if p.offset > entry.written {
+            return Err(invalid("read offset is beyond transfer length"));
+        }
+        let mut file = File::open(&entry.path).map_err(io_error)?;
+        file.seek(SeekFrom::Start(p.offset)).map_err(io_error)?;
+        let mut buf = vec![0u8; TRANSFER_CHUNK_SIZE as usize];
+        let count = file.read(&mut buf).map_err(io_error)?;
+        buf.truncate(count);
+        let next = p.offset + count as u64;
+        Ok(TransferChunkResult {
+            next_offset: next,
+            eof: next >= entry.written,
+            data_base64: Some(base64::engine::general_purpose::STANDARD.encode(buf)),
+        })
+    }
+
+    pub fn release(&self, p: TransferIdParams) -> TransferReleaseResult {
+        let entry = self.entries.lock().unwrap().remove(&p.transfer_id);
+        if let Some(entry) = entry {
+            let dir = if entry.path.is_dir() {
+                entry.path
+            } else {
+                entry.path.parent().unwrap_or(&self.root).to_path_buf()
+            };
+            let _ = fs::remove_dir_all(dir);
+            TransferReleaseResult { released: true }
+        } else {
+            TransferReleaseResult { released: false }
+        }
+    }
+
+    pub fn release_session(&self, session_id: &str) {
+        let ids: Vec<String> = self
+            .entries
+            .lock()
+            .unwrap()
+            .iter()
+            .filter(|(_, entry)| entry.session_id == session_id)
+            .map(|(id, _)| id.clone())
+            .collect();
+        for id in ids {
+            self.release(TransferIdParams { transfer_id: id });
+        }
+    }
+}
+
+fn new_id() -> String {
+    format!("tr_{}", Uuid::new_v4().simple())
+}
+
+fn set_private_dir(path: &Path) -> std::io::Result<()> {
+    #[cfg(unix)]
+    {
+        use std::os::unix::fs::PermissionsExt;
+        fs::set_permissions(path, fs::Permissions::from_mode(0o700))?;
+    }
+    Ok(())
+}
+
+fn set_private_file(file: &File) -> std::io::Result<()> {
+    #[cfg(unix)]
+    {
+        use std::os::unix::fs::PermissionsExt;
+        file.set_permissions(fs::Permissions::from_mode(0o600))?;
+    }
+    Ok(())
+}
+
+fn remove_if_present(path: &Path) -> std::io::Result<()> {
+    match fs::remove_file(path) {
+        Ok(()) => Ok(()),
+        Err(err) if err.kind() == std::io::ErrorKind::NotFound => Ok(()),
+        Err(err) => Err(err),
+    }
+}
+
+fn remove_dir_if_present(path: &Path) -> std::io::Result<()> {
+    match fs::remove_dir(path) {
+        Ok(()) => Ok(()),
+        Err(err) if err.kind() == std::io::ErrorKind::NotFound => Ok(()),
+        Err(err) => Err(err),
+    }
+}
+
+fn safe_upload_basename(name: &str) -> Result<&str, RpcError> {
+    if name.is_empty() || name.contains(['/', '\\', '\0']) {
+        return Err(invalid("upload name must be a safe basename"));
+    }
+    let mut components = Path::new(name).components();
+    match (components.next(), components.next()) {
+        (Some(Component::Normal(_)), None) => Ok(name),
+        _ => Err(invalid("upload name must be a safe basename")),
+    }
+}
+
+fn invalid(message: impl Into<String>) -> RpcError {
+    RpcError {
+        code: ErrorCode::InvalidParams,
+        message: message.into(),
+        data: None,
+    }
+}
+fn not_found(message: impl Into<String>) -> RpcError {
+    RpcError {
+        code: ErrorCode::NotFound,
+        message: message.into(),
+        data: None,
+    }
+}
+fn permission(message: impl Into<String>) -> RpcError {
+    RpcError {
+        code: ErrorCode::PermissionDenied,
+        message: message.into(),
+        data: None,
+    }
+}
+fn io_error(err: std::io::Error) -> RpcError {
+    RpcError {
+        code: ErrorCode::ProtocolError,
+        message: err.to_string(),
+        data: None,
+    }
+}
+
+fn protocol_error(message: impl Into<String>) -> RpcError {
+    RpcError {
+        code: ErrorCode::ProtocolError,
+        message: message.into(),
+        data: None,
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn registry() -> (tempfile::TempDir, TransferRegistry) {
+        let temp = tempfile::tempdir().unwrap();
+        let registry = TransferRegistry::at_root(temp.path().join("transfers")).unwrap();
+        (temp, registry)
+    }
+
+    #[test]
+    fn upload_is_not_resolvable_until_complete_or_from_another_session() {
+        let (_temp, registry) = registry();
+        let begin = registry
+            .begin_upload(TransferBeginParams {
+                session_id: "s1".into(),
+                name: "image.png".into(),
+                byte_size: 3,
+            })
+            .unwrap();
+        assert!(
+            registry
+                .resolve_uploads("s1", std::slice::from_ref(&begin.transfer_id))
+                .is_err()
+        );
+        registry
+            .write_chunk(TransferChunkParams {
+                transfer_id: begin.transfer_id.clone(),
+                offset: 0,
+                data_base64: base64::engine::general_purpose::STANDARD.encode(b"abc"),
+            })
+            .unwrap();
+        registry
+            .finish_upload(TransferIdParams {
+                transfer_id: begin.transfer_id.clone(),
+            })
+            .unwrap();
+        assert!(
+            registry
+                .resolve_uploads("s2", std::slice::from_ref(&begin.transfer_id))
+                .is_err()
+        );
+        let [path] = registry
+            .resolve_uploads("s1", &[begin.transfer_id])
+            .unwrap()
+            .try_into()
+            .unwrap();
+        assert_eq!(path.file_name().unwrap(), "image.png");
+        assert_eq!(fs::read(path).unwrap(), b"abc");
+    }
+
+    #[test]
+    fn upload_name_must_be_one_safe_path_component() {
+        let (_temp, registry) = registry();
+        for name in ["", ".", "..", "../secret", "folder/file", "folder\\file"] {
+            let result = registry.begin_upload(TransferBeginParams {
+                session_id: "s1".into(),
+                name: name.into(),
+                byte_size: 0,
+            });
+            assert!(result.is_err(), "accepted unsafe upload name");
+        }
+        assert!(registry.entries.lock().unwrap().is_empty());
+        assert!(fs::read_dir(&registry.root).unwrap().next().is_none());
+    }
+
+    #[test]
+    fn same_upload_names_are_isolated_by_transfer_directory() {
+        let (_temp, registry) = registry();
+        let first = registry
+            .begin_upload(TransferBeginParams {
+                session_id: "s1".into(),
+                name: "image.png".into(),
+                byte_size: 0,
+            })
+            .unwrap();
+        let second = registry
+            .begin_upload(TransferBeginParams {
+                session_id: "s1".into(),
+                name: "image.png".into(),
+                byte_size: 0,
+            })
+            .unwrap();
+        let entries = registry.entries.lock().unwrap();
+        let first_path = &entries[&first.transfer_id].path;
+        let second_path = &entries[&second.transfer_id].path;
+        assert_ne!(first_path.parent(), second_path.parent());
+        assert_eq!(first_path.file_name().unwrap(), "image.png");
+        assert_eq!(second_path.file_name().unwrap(), "image.png");
+    }
+
+    #[cfg(unix)]
+    #[test]
+    fn upload_staging_is_private() {
+        use std::os::unix::fs::PermissionsExt;
+
+        let (_temp, registry) = registry();
+        let begin = registry
+            .begin_upload(TransferBeginParams {
+                session_id: "s1".into(),
+                name: "private.png".into(),
+                byte_size: 0,
+            })
+            .unwrap();
+        let entries = registry.entries.lock().unwrap();
+        let path = &entries[&begin.transfer_id].path;
+        assert_eq!(
+            fs::metadata(path.parent().unwrap())
+                .unwrap()
+                .permissions()
+                .mode()
+                & 0o777,
+            0o700
+        );
+        assert_eq!(
+            fs::metadata(path).unwrap().permissions().mode() & 0o777,
+            0o600
+        );
+    }
+
+    #[test]
+    fn download_must_import_from_its_browser_capability_directory() {
+        let (temp, registry) = registry();
+        let staging = registry.begin_download("s1").unwrap();
+        let outside = temp.path().join("outside.bin");
+        fs::write(&outside, b"secret").unwrap();
+        assert!(
+            registry
+                .import_download(&staging.transfer_id, &outside)
+                .is_err()
+        );
+        assert_eq!(fs::read(&outside).unwrap(), b"secret");
+
+        let browser_dir = temp
+            .path()
+            .join("Downloads")
+            .join("BrowserSkill")
+            .join(&staging.transfer_id);
+        fs::create_dir_all(&browser_dir).unwrap();
+        let inside = browser_dir.join("result.bin");
+        fs::write(&inside, b"result").unwrap();
+        assert_eq!(
+            registry
+                .import_download(&staging.transfer_id, &inside)
+                .unwrap(),
+            6
+        );
+        assert!(!inside.exists());
+        let chunk = registry
+            .read_chunk(TransferChunkParams {
+                transfer_id: staging.transfer_id,
+                offset: 0,
+                data_base64: String::new(),
+            })
+            .unwrap();
+        assert_eq!(
+            base64::engine::general_purpose::STANDARD
+                .decode(chunk.data_base64.unwrap())
+                .unwrap(),
+            b"result"
+        );
+    }
+
+    #[test]
+    fn validated_oversized_browser_download_is_removed_on_rejection() {
+        let (temp, registry) = registry();
+        let staging = registry.begin_download("s1").unwrap();
+        let browser_dir = temp
+            .path()
+            .join("Downloads")
+            .join("BrowserSkill")
+            .join(&staging.transfer_id);
+        fs::create_dir_all(&browser_dir).unwrap();
+        let inside = browser_dir.join("oversized.bin");
+        File::create(&inside)
+            .unwrap()
+            .set_len(MAX_TRANSFER_BYTES + 1)
+            .unwrap();
+
+        assert!(
+            registry
+                .import_download(&staging.transfer_id, &inside)
+                .is_err()
+        );
+        assert!(!inside.exists());
+        assert!(!browser_dir.exists());
+    }
+
+    #[test]
+    fn releasing_a_session_removes_all_staging() {
+        let (_temp, registry) = registry();
+        let first = registry.begin_download("s1").unwrap();
+        let second = registry.begin_download("s2").unwrap();
+        let first_directory = registry.root.join(&first.transfer_id);
+        let second_directory = registry.root.join(&second.transfer_id);
+        registry.release_session("s1");
+        assert!(!first_directory.exists());
+        assert!(second_directory.exists());
+    }
+
+    #[test]
+    fn upload_staging_is_bounded_across_a_session() {
+        let (_temp, registry) = registry();
+        registry
+            .begin_upload(TransferBeginParams {
+                session_id: "s1".into(),
+                name: "large.bin".into(),
+                byte_size: MAX_TRANSFER_BYTES,
+            })
+            .unwrap();
+        assert!(
+            registry
+                .begin_upload(TransferBeginParams {
+                    session_id: "s1".into(),
+                    name: "one-more-byte.bin".into(),
+                    byte_size: 1,
+                })
+                .is_err()
+        );
+        assert!(
+            registry
+                .begin_upload(TransferBeginParams {
+                    session_id: "s2".into(),
+                    name: "other-session.bin".into(),
+                    byte_size: 1,
+                })
+                .is_ok()
+        );
+    }
+}

crates/bsk-cli/src/daemon/ipc.rs

@@ -27,7 +27,8 @@ use bsk_protocol::system::{
     VersionSkewEntry,
 };
 use bsk_protocol::tools::{
-    HelpOutcome, RequestHelpResult, ReturnFailure, WaitMsParams, WaitMsResult,
+    DownloadParams, DownloadResult, HelpOutcome, RequestHelpResult, ReturnFailure,
+    TransferBeginParams, TransferIdParams, UploadParams, WaitMsParams, WaitMsResult,
 };
 use bsk_protocol::{
     CancelParams, CancelResult, ErrorCode, Method, PingResult, ResponseBody, RpcError, RpcId,
@@ -61,6 +62,10 @@ pub type RpcHandler = Arc<
 >;
 
 const DEFAULT_RPC_TIMEOUT: Duration = Duration::from_secs(15);
+// The upload transaction owns its operation deadline and may need a bounded
+// cleanup before it can return a useful structured error. Keep only that
+// transport alive slightly longer so it does not replace the result.
+const EXTENSION_RESPONSE_GRACE: Duration = Duration::from_secs(2);
 /// Upper bound on `wait_for_browser_ms` accepted over IPC.
 const MAX_BROWSER_WAIT: Duration = Duration::from_secs(60);
 // `session.stop` fast-fails while another tool is active; this budget
@@ -206,15 +211,15 @@ pub fn full_handler(status: DaemonStatus, state: Arc<DaemonState>) -> RpcHandler
                     Ok(v) => ResponseBody::Ok(v),
                     Err(e) => ResponseBody::Err(e),
                 },
-                Method::SessionStart => match handle_session_start(&state, params).await {
+                Method::SessionStart => match handle_session_start(&state, rpc_id, params).await {
                     Ok(v) => ResponseBody::Ok(v),
                     Err(e) => ResponseBody::Err(e),
                 },
-                Method::SessionStop => match handle_session_stop(&state, params).await {
+                Method::SessionStop => match handle_session_stop(&state, rpc_id, params).await {
                     Ok(v) => ResponseBody::Ok(v),
                     Err(e) => ResponseBody::Err(e),
                 },
-                Method::SessionStopAll => match handle_session_stop_all(&state).await {
+                Method::SessionStopAll => match handle_session_stop_all_rpc(&state, rpc_id).await {
                     Ok(v) => ResponseBody::Ok(v),
                     Err(e) => ResponseBody::Err(e),
                 },
@@ -223,6 +228,11 @@ pub fn full_handler(status: DaemonStatus, state: Arc<DaemonState>) -> RpcHandler
                     Ok(v) => ResponseBody::Ok(v),
                     Err(e) => ResponseBody::Err(e),
                 },
+                Method::TransferBegin => handle_transfer_begin(&state, params),
+                Method::TransferChunk => handle_transfer_chunk(&state, params),
+                Method::TransferFinish => handle_transfer_finish(&state, params),
+                Method::TransferRead => handle_transfer_read(&state, params),
+                Method::TransferRelease => handle_transfer_release(&state, params),
                 Method::ToolTabList
                 | Method::ToolTabCreate
                 | Method::ToolTabClose
@@ -248,6 +258,8 @@ pub fn full_handler(status: DaemonStatus, state: Arc<DaemonState>) -> RpcHandler
                 | Method::ToolFill
                 | Method::ToolPress
                 | Method::ToolSelect
+                | Method::ToolUpload
+                | Method::ToolDownload
                 | Method::ToolEvaluate
                 | Method::ToolWaitForNavigation
                 | Method::ToolRequestHelp
@@ -326,7 +338,7 @@ async fn handle_tool_dispatch(
             data: None,
         });
     }
-    let timeout = match tool_dispatch_timeout(&params) {
+    let timeout = match tool_dispatch_transport_timeout(&method, &params) {
         Ok(timeout) => timeout,
         Err(err) => return ResponseBody::Err(err),
     };
@@ -340,24 +352,180 @@ async fn handle_tool_dispatch(
             });
         }
     };
+    let mut params = params;
+    let mut download_transfer_id: Option<String> = None;
+    if method == Method::ToolUpload {
+        let mut upload: UploadParams = match serde_json::from_value(params) {
+            Ok(v) => v,
+            Err(err) => return ResponseBody::Err(invalid_params(err.to_string())),
+        };
+        if upload.files.is_empty() || upload.files.len() > super::file_transfer::MAX_UPLOAD_FILES {
+            return ResponseBody::Err(invalid_params(format!(
+                "upload requires 1..={} files",
+                super::file_transfer::MAX_UPLOAD_FILES
+            )));
+        }
+        let ids: Vec<String> = upload.files.iter().map(|f| f.transfer_id.clone()).collect();
+        let paths = match state.transfers.resolve_uploads(&session_id.0, &ids) {
+            Ok(v) => v,
+            Err(err) => return ResponseBody::Err(err),
+        };
+        for (file, path) in upload.files.iter_mut().zip(paths) {
+            file.staged_path = Some(path.to_string_lossy().into_owned());
+        }
+        params = serde_json::to_value(upload).unwrap_or(Value::Null);
+    } else if method == Method::ToolDownload {
+        let mut download: DownloadParams = match serde_json::from_value(params) {
+            Ok(v) => v,
+            Err(err) => return ResponseBody::Err(invalid_params(err.to_string())),
+        };
+        let staging = match state.transfers.begin_download(&session_id.0) {
+            Ok(v) => v,
+            Err(err) => return ResponseBody::Err(err),
+        };
+        download.browser_relative_dir = Some(staging.browser_relative_dir);
+        download.max_byte_size = Some(super::file_transfer::MAX_TRANSFER_BYTES);
+        download_transfer_id = Some(staging.transfer_id);
+        params = serde_json::to_value(download).unwrap_or(Value::Null);
+    }
     let entry = inflight_guard.entry();
     // `record_stop` must reach the extension while `record_await` holds the
     // serial busy lock — finishing the recording unblocks await.
     let outcome = if method == Method::ToolRecordStop {
         state
             .tool_queues
-            .dispatch_unlocked(&session_id, method, params, timeout, Some(entry))
+            .dispatch_unlocked(&session_id, method.clone(), params, timeout, Some(entry))
             .await
     } else {
         state
             .tool_queues
-            .dispatch(&session_id, method, params, timeout, Some(entry))
+            .dispatch(&session_id, method.clone(), params, timeout, Some(entry))
             .await
     };
     drop(inflight_guard);
     match outcome {
+        Ok(v) if method == Method::ToolDownload => {
+            let id = download_transfer_id.expect("download transfer allocated");
+            let mut result: DownloadResult = match serde_json::from_value(v) {
+                Ok(v) => v,
+                Err(err) => {
+                    state
+                        .transfers
+                        .release(TransferIdParams { transfer_id: id });
+                    return ResponseBody::Err(RpcError {
+                        code: ErrorCode::ProtocolError,
+                        message: format!("invalid tool.download result: {err}"),
+                        data: None,
+                    });
+                }
+            };
+            let Some(path) = result.browser_path.take() else {
+                state
+                    .transfers
+                    .release(TransferIdParams { transfer_id: id });
+                return ResponseBody::Err(RpcError {
+                    code: ErrorCode::ProtocolError,
+                    message: "tool.download returned no browser_path".into(),
+                    data: None,
+                });
+            };
+            match state
+                .transfers
+                .import_download(&id, std::path::Path::new(&path))
+            {
+                Ok(size) => {
+                    result.byte_size = size;
+                    result.transfer_id = Some(id);
+                    ResponseBody::Ok(serde_json::to_value(result).unwrap_or(Value::Null))
+                }
+                Err(err) => {
+                    state
+                        .transfers
+                        .release(TransferIdParams { transfer_id: id });
+                    ResponseBody::Err(err)
+                }
+            }
+        }
         Ok(v) => ResponseBody::Ok(v),
-        Err(err) => ResponseBody::Err(err.into_rpc()),
+        Err(err) => {
+            if let Some(id) = download_transfer_id {
+                state
+                    .transfers
+                    .release(TransferIdParams { transfer_id: id });
+            }
+            ResponseBody::Err(err.into_rpc())
+        }
+    }
+}
+
+fn handle_transfer_begin(state: &Arc<DaemonState>, params: Value) -> ResponseBody {
+    let p: TransferBeginParams = match serde_json::from_value(params) {
+        Ok(v) => v,
+        Err(err) => return ResponseBody::Err(invalid_params(err.to_string())),
+    };
+    if state
+        .sessions
+        .get(&SessionId(p.session_id.clone()))
+        .is_none()
+    {
+        return ResponseBody::Err(RpcError {
+            code: ErrorCode::NotFound,
+            message: format!("session {} unknown", p.session_id),
+            data: None,
+        });
+    }
+    match state.transfers.begin_upload(p) {
+        Ok(v) => ResponseBody::Ok(serde_json::to_value(v).unwrap_or(Value::Null)),
+        Err(err) => ResponseBody::Err(err),
+    }
+}
+
+fn handle_transfer_chunk(state: &Arc<DaemonState>, params: Value) -> ResponseBody {
+    let p = match serde_json::from_value(params) {
+        Ok(v) => v,
+        Err(err) => return ResponseBody::Err(invalid_params(err.to_string())),
+    };
+    match state.transfers.write_chunk(p) {
+        Ok(v) => ResponseBody::Ok(serde_json::to_value(v).unwrap_or(Value::Null)),
+        Err(err) => ResponseBody::Err(err),
+    }
+}
+
+fn handle_transfer_finish(state: &Arc<DaemonState>, params: Value) -> ResponseBody {
+    let p = match serde_json::from_value(params) {
+        Ok(v) => v,
+        Err(err) => return ResponseBody::Err(invalid_params(err.to_string())),
+    };
+    match state.transfers.finish_upload(p) {
+        Ok(v) => ResponseBody::Ok(serde_json::to_value(v).unwrap_or(Value::Null)),
+        Err(err) => ResponseBody::Err(err),
+    }
+}
+
+fn handle_transfer_read(state: &Arc<DaemonState>, params: Value) -> ResponseBody {
+    let p = match serde_json::from_value(params) {
+        Ok(v) => v,
+        Err(err) => return ResponseBody::Err(invalid_params(err.to_string())),
+    };
+    match state.transfers.read_chunk(p) {
+        Ok(v) => ResponseBody::Ok(serde_json::to_value(v).unwrap_or(Value::Null)),
+        Err(err) => ResponseBody::Err(err),
+    }
+}
+
+fn handle_transfer_release(state: &Arc<DaemonState>, params: Value) -> ResponseBody {
+    let p: TransferIdParams = match serde_json::from_value(params) {
+        Ok(v) => v,
+        Err(err) => return ResponseBody::Err(invalid_params(err.to_string())),
+    };
+    ResponseBody::Ok(serde_json::to_value(state.transfers.release(p)).unwrap_or(Value::Null))
+}
+
+fn invalid_params(message: impl Into<String>) -> RpcError {
+    RpcError {
+        code: ErrorCode::InvalidParams,
+        message: message.into(),
+        data: None,
     }
 }
 
@@ -436,8 +604,9 @@ async fn handle_wait_ms(
 /// cancellation surfaces (M10.2 + review C2):
 ///
 /// 1. [`AbortRegistry`] — answers daemon-local cancellable runners
-///    (currently `tool.wait_ms`). If a token is registered we trip it
-///    and stop.
+///    (`tool.wait_ms` and `session.*` lifecycle calls). If a token is
+///    registered we trip it and stop. Lifecycle handlers forward their
+///    own WS cancel after preserving request-before-cancel ordering.
 /// 2. [`super::inflight::ToolInflightRegistry`] — every IPC-tracked
 ///    `tool.*` RPC, registered the moment the IPC handler accepts the
 ///    request. Trips the entry's cancel token regardless of whether
@@ -534,6 +703,16 @@ fn tool_dispatch_timeout(params: &Value) -> Result<Duration, RpcError> {
     Ok(Duration::from_millis(u64::from(ms)))
 }
 
+fn tool_dispatch_transport_timeout(method: &Method, params: &Value) -> Result<Duration, RpcError> {
+    tool_dispatch_timeout(params).map(|timeout| {
+        if matches!(method, Method::ToolUpload | Method::ToolDownload) {
+            timeout.saturating_add(EXTENSION_RESPONSE_GRACE)
+        } else {
+            timeout
+        }
+    })
+}
+
 // Local CLI-facing shapes. Intentionally distinct from
 // `bsk_protocol::tools::SessionStart*` (which describes the WS-facing
 // `tool.session_*` round-trip with the extension) so the two sides can
@@ -636,7 +815,20 @@ fn clamp_browser_wait(wait_ms: Option<u64>) -> Option<Duration> {
     ))
 }
 
-async fn handle_session_start(state: &Arc<DaemonState>, params: Value) -> Result<Value, RpcError> {
+async fn handle_session_start(
+    state: &Arc<DaemonState>,
+    rpc_id: RpcId,
+    params: Value,
+) -> Result<Value, RpcError> {
+    let abort_guard = state
+        .abort_registry
+        .register(rpc_id)
+        .map_err(|err| RpcError {
+            code: ErrorCode::ProtocolError,
+            message: format!("session.start cancellation registration failed: {err:?}"),
+            data: None,
+        })?;
+    let cancel = abort_guard.token().clone();
     let params: CliSessionStartParams = if params.is_null() {
         CliSessionStartParams {
             browser_instance_id: None,
@@ -675,6 +867,7 @@ async fn handle_session_start(state: &Arc<DaemonState>, params: Value) -> Result
         },
         state.config.extension_connect_wait,
         DEFAULT_RPC_TIMEOUT,
+        Some(cancel),
     )
     .await
     {
@@ -698,6 +891,8 @@ fn map_start_error(err: StartSessionError) -> RpcError {
         StartSessionError::AmbiguousBrowserLabel { .. } => ErrorCode::InvalidParams,
         StartSessionError::IdExhausted => ErrorCode::ProtocolError,
         StartSessionError::Timeout => ErrorCode::Timeout,
+        StartSessionError::Cancelled => ErrorCode::Cancelled,
+        StartSessionError::CleanupFailed { .. } => ErrorCode::ProtocolError,
         StartSessionError::TransportClosed => ErrorCode::ProtocolError,
         StartSessionError::ExtensionError(inner) => inner.code,
     };
@@ -713,6 +908,16 @@ fn map_start_error(err: StartSessionError) -> RpcError {
             "label": label,
             "instance_ids": instance_ids,
         })),
+        StartSessionError::CleanupFailed {
+            session_id,
+            agent_window_id,
+            ..
+        } => Some(serde_json::json!({
+            "reason": "cleanup_failed",
+            "session_id": session_id.0,
+            "agent_window_id": agent_window_id,
+        })),
+        StartSessionError::ExtensionError(inner) => inner.data.clone(),
         _ => None,
     };
     RpcError {
@@ -722,7 +927,20 @@ fn map_start_error(err: StartSessionError) -> RpcError {
     }
 }
 
-async fn handle_session_stop(state: &Arc<DaemonState>, params: Value) -> Result<Value, RpcError> {
+async fn handle_session_stop(
+    state: &Arc<DaemonState>,
+    rpc_id: RpcId,
+    params: Value,
+) -> Result<Value, RpcError> {
+    let abort_guard = state
+        .abort_registry
+        .register(rpc_id)
+        .map_err(|err| RpcError {
+            code: ErrorCode::ProtocolError,
+            message: format!("session.stop cancellation registration failed: {err:?}"),
+            data: None,
+        })?;
+    let cancel = abort_guard.token().clone();
     let params: CliSessionStopParams = if params.is_null() {
         CliSessionStopParams {
             session_id: None,
@@ -736,7 +954,7 @@ async fn handle_session_stop(state: &Arc<DaemonState>, params: Value) -> Result<
         })?
     };
     if params.all {
-        return handle_session_stop_all(state).await;
+        return handle_session_stop_all(state, Some(cancel)).await;
     }
     let session_id = match params.session_id {
         Some(s) => SessionId(s),
@@ -755,10 +973,12 @@ async fn handle_session_stop(state: &Arc<DaemonState>, params: Value) -> Result<
         &state.session_interrupts,
         &session_id,
         DEFAULT_SESSION_STOP_TIMEOUT,
+        Some(cancel),
     )
     .await
     {
         Ok(stop) => {
+            state.transfers.release_session(&session_id.0);
             let result = CliSessionStopResult {
                 stopped: vec![session_id.0],
                 failed: Vec::new(),
@@ -806,7 +1026,25 @@ fn map_stop_error(err: StopSessionError) -> RpcError {
     }
 }
 
-async fn handle_session_stop_all(state: &Arc<DaemonState>) -> Result<Value, RpcError> {
+async fn handle_session_stop_all_rpc(
+    state: &Arc<DaemonState>,
+    rpc_id: RpcId,
+) -> Result<Value, RpcError> {
+    let abort_guard = state
+        .abort_registry
+        .register(rpc_id)
+        .map_err(|err| RpcError {
+            code: ErrorCode::ProtocolError,
+            message: format!("session.stop_all cancellation registration failed: {err:?}"),
+            data: None,
+        })?;
+    handle_session_stop_all(state, Some(abort_guard.token().clone())).await
+}
+
+async fn handle_session_stop_all(
+    state: &Arc<DaemonState>,
+    cancel: Option<super::abort::AbortToken>,
+) -> Result<Value, RpcError> {
     let ids: Vec<SessionId> = state
         .sessions
         .snapshot()
@@ -818,6 +1056,16 @@ async fn handle_session_stop_all(state: &Arc<DaemonState>) -> Result<Value, RpcE
     let mut returned_tab_ids = Vec::new();
     let mut return_failures = Vec::new();
     for id in ids {
+        if cancel
+            .as_ref()
+            .is_some_and(super::abort::AbortToken::is_cancelled)
+        {
+            return Err(RpcError {
+                code: ErrorCode::Cancelled,
+                message: "session.stop_all was cancelled".into(),
+                data: None,
+            });
+        }
         match stop_session(
             &state.browsers,
             &state.sessions,
@@ -825,10 +1073,12 @@ async fn handle_session_stop_all(state: &Arc<DaemonState>) -> Result<Value, RpcE
             &state.session_interrupts,
             &id,
             DEFAULT_SESSION_STOP_TIMEOUT,
+            cancel.clone(),
         )
         .await
         {
             Ok(stop) => {
+                state.transfers.release_session(&id.0);
                 stopped.push(id.0);
                 returned_tab_ids.extend(stop.returned_tab_ids);
                 return_failures.extend(stop.return_failures);
@@ -850,6 +1100,15 @@ async fn handle_session_stop_all(state: &Arc<DaemonState>) -> Result<Value, RpcE
                 returned_tab_ids.extend(stop.returned_tab_ids);
                 return_failures.extend(stop.return_failures);
             }
+            Err(err)
+                if matches!(
+                    &err,
+                    StopSessionError::ExtensionError(inner)
+                        if inner.code == ErrorCode::Cancelled
+                ) =>
+            {
+                return Err(map_stop_error(err));
+            }
             Err(err) => {
                 debug!(session = %id, ?err, "session.stop_all: failure (continuing)");
                 let rpc = map_stop_error(err);
@@ -1142,6 +1401,28 @@ mod windows {
                 connected = pipe.connect() => {
                     match connected {
                         Ok(()) => {
+                            // Keep the connected instance alive until its replacement
+                            // exists. Otherwise a fast handler can close the last
+                            // instance and make new clients fail with NotFound.
+                            loop {
+                                match ServerOptions::new()
+                                    .access_inbound(true)
+                                    .access_outbound(true)
+                                    .create(&listener.pipe_name)
+                                {
+                                    Ok(next) => {
+                                        listener.first = Some(next);
+                                        break;
+                                    }
+                                    Err(err) => {
+                                        warn!(?err, "create next named-pipe instance failed");
+                                        tokio::select! {
+                                            _ = &mut shutdown => return,
+                                            _ = tokio::time::sleep(std::time::Duration::from_millis(50)) => {}
+                                        }
+                                    }
+                                }
+                            }
                             on_open();
                             let handler = handler.clone();
                             let on_act = on_activity.clone();
@@ -1210,6 +1491,118 @@ mod windows {
         }
         Ok(())
     }
+    #[cfg(test)]
+    mod tests {
+        use super::*;
+        use std::sync::Mutex;
+        use std::time::Duration;
+        use tokio::net::windows::named_pipe::ClientOptions;
+        use tokio::sync::oneshot;
+
+        fn isolated_listener() -> NamedPipeListener {
+            let pipe_name = format!(r"\\.\pipe\bsk-test-{}", uuid::Uuid::new_v4());
+            let first = ServerOptions::new()
+                .first_pipe_instance(true)
+                .create(&pipe_name)
+                .unwrap();
+            NamedPipeListener {
+                pipe_name,
+                first: Some(first),
+            }
+        }
+
+        #[tokio::test]
+        async fn next_instance_exists_before_connection_is_handed_off() {
+            let listener = isolated_listener();
+            let name = listener.pipe_name.clone();
+            let probe_name = name.clone();
+            let (probe_tx, probe_rx) = oneshot::channel();
+            let probe_tx = Mutex::new(Some(probe_tx));
+            let (stop_tx, stop_rx) = oneshot::channel();
+            let server = tokio::spawn(serve(
+                listener,
+                crate::daemon::ipc::default_ping_handler(),
+                move || {
+                    if let Some(tx) = probe_tx.lock().unwrap().take() {
+                        // This callback runs before the connection task can finish.
+                        // A second client must already have an instance to open.
+                        let result = ClientOptions::new().open(&probe_name).map(drop);
+                        let _ = tx.send(result);
+                    }
+                },
+                || {},
+                || {},
+                async {
+                    let _ = stop_rx.await;
+                },
+            ));
+            let client = ClientOptions::new().open(&name).unwrap();
+            let result = tokio::time::timeout(Duration::from_secs(5), probe_rx).await;
+            drop(client);
+            let _ = stop_tx.send(());
+            server.await.unwrap();
+            result
+                .expect("accept callback ran")
+                .unwrap()
+                .expect("next pipe instance is ready");
+        }
+
+        #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+        async fn rapid_disconnects_and_concurrent_rpc_connections() {
+            let listener = isolated_listener();
+            let name = listener.pipe_name.clone();
+            let (stop_tx, stop_rx) = oneshot::channel();
+            let server = tokio::spawn(serve(
+                listener,
+                crate::daemon::ipc::default_ping_handler(),
+                || {},
+                || {},
+                || {},
+                async {
+                    let _ = stop_rx.await;
+                },
+            ));
+            let exercise = async {
+                // Include clients that close without sending any request.
+                for _ in 0..64 {
+                    drop(
+                        crate::ipc_client::Client::connect_path(name.clone().into())
+                            .await
+                            .unwrap(),
+                    );
+                }
+                let mut clients = tokio::task::JoinSet::new();
+                for _ in 0..8 {
+                    let name = name.clone();
+                    clients.spawn(async move {
+                        for _ in 0..32 {
+                            let mut client =
+                                crate::ipc_client::Client::connect_path(name.clone().into())
+                                    .await
+                                    .unwrap();
+                            let reply: bsk_protocol::PingResult = client
+                                .call(
+                                    bsk_protocol::Method::SystemPing,
+                                    &serde_json::json!({}),
+                                    Duration::from_secs(5),
+                                )
+                                .await
+                                .unwrap()
+                                .unwrap();
+                            assert!(reply.pong);
+                        }
+                    });
+                }
+                while let Some(result) = clients.join_next().await {
+                    result.unwrap();
+                }
+            };
+            let result = tokio::time::timeout(Duration::from_secs(30), exercise).await;
+            let _ = stop_tx.send(());
+            server.await.unwrap();
+            result.expect("connection exercise completed");
+        }
+    }
 }
 
 #[cfg(windows)]
@@ -1252,6 +1645,30 @@ mod tests {
         assert_eq!(got, std::time::Duration::from_millis(300_000));
     }
 
+    #[test]
+    fn extension_transport_outlives_the_operation_deadline() {
+        let params = serde_json::json!({
+            "session_id": "abcd",
+            "timeout_ms": 60_000,
+        });
+        let dispatch_timeout =
+            tool_dispatch_transport_timeout(&Method::ToolUpload, &params).unwrap();
+
+        assert_eq!(dispatch_timeout, Duration::from_secs(62));
+    }
+
+    #[test]
+    fn extension_response_grace_does_not_change_other_tools() {
+        let params = serde_json::json!({
+            "session_id": "abcd",
+            "timeout_ms": 60_000,
+        });
+        let dispatch_timeout =
+            tool_dispatch_transport_timeout(&Method::ToolClick, &params).unwrap();
+
+        assert_eq!(dispatch_timeout, Duration::from_secs(60));
+    }
+
     #[test]
     fn tool_dispatch_timeout_rejects_zero_timeout_ms() {
         let params = serde_json::json!({

crates/bsk-cli/src/daemon/lockfile.rs

@@ -90,17 +90,21 @@ pub fn pid_alive(pid: u32) -> bool {
 
     #[cfg(windows)]
     {
-        use windows_sys::Win32::Foundation::CloseHandle;
+        use windows_sys::Win32::Foundation::{CloseHandle, WAIT_TIMEOUT};
         use windows_sys::Win32::System::Threading::{
-            OpenProcess, PROCESS_QUERY_LIMITED_INFORMATION,
+            OpenProcess, PROCESS_SYNCHRONIZE, WaitForSingleObject,
         };
         unsafe {
-            let handle = OpenProcess(PROCESS_QUERY_LIMITED_INFORMATION, 0, pid);
+            let handle = OpenProcess(PROCESS_SYNCHRONIZE, 0, pid);
             if handle.is_null() {
                 return false;
             }
+            // An exited process can still be opened while another handle
+            // keeps its kernel object alive. Poll its termination signal;
+            // unlike checking STILL_ACTIVE, this also handles exit code 259.
+            let alive = WaitForSingleObject(handle, 0) == WAIT_TIMEOUT;
             CloseHandle(handle);
-            true
+            alive
         }
     }
 
@@ -110,3 +114,39 @@ pub fn pid_alive(pid: u32) -> bool {
         false
     }
 }
+
+#[cfg(all(test, windows))]
+mod tests {
+    use super::pid_alive;
+    use std::os::windows::process::CommandExt;
+    use std::process::Command;
+
+    #[test]
+    fn current_process_is_alive() {
+        assert!(pid_alive(std::process::id()));
+    }
+
+    #[test]
+    fn invalid_process_is_not_alive() {
+        assert!(!pid_alive(0));
+        assert!(!pid_alive(u32::MAX));
+    }
+
+    #[test]
+    fn exited_process_with_retained_handle_is_not_alive() {
+        for exit_code in [0, 259] {
+            let mut child = Command::new("cmd.exe")
+                .args(["/D", "/C", &format!("exit {exit_code}")])
+                .creation_flags(0x0800_0000) // CREATE_NO_WINDOW
+                .spawn()
+                .expect("spawn short-lived process");
+            assert_eq!(child.wait().unwrap().code(), Some(exit_code));
+            // Keep Child (and its process handle) alive during the probe.
+            assert!(
+                !pid_alive(child.id()),
+                "exited process with code {exit_code}"
+            );
+            drop(child);
+        }
+    }
+}

crates/bsk-cli/src/daemon/mod.rs

@@ -3,6 +3,7 @@
 pub mod abort;
 pub mod browsers;
 mod cancel_forward;
+pub mod file_transfer;
 pub mod inflight;
 pub mod info;
 pub mod ipc;

crates/bsk-cli/src/daemon/paths.rs

@@ -100,6 +100,14 @@ pub fn record_session_path() -> Result<PathBuf> {
     Ok(bsk_home()?.join("record-session.json"))
 }
 
+/// Path to a completed Trace saved before bundle export (`record-recovery.json`).
+///
+/// Kept until export succeeds so a failed `--output` write cannot drop the
+/// recording the extension already returned.
+pub fn record_recovery_path() -> Result<PathBuf> {
+    Ok(bsk_home()?.join("record-recovery.json"))
+}
+
 /// Windows named-pipe name. Include the resolved `BSK_HOME` path in the
 /// token so test homes and custom installs do not share a predictable
 /// per-username pipe.
@@ -119,6 +127,13 @@ pub fn pipe_name() -> String {
     render_pipe_name(&user, &home)
 }
 
+/// Process-wide lock for tests that mutate `BSK_HOME`.
+#[cfg(test)]
+pub(crate) fn test_env_lock() -> std::sync::MutexGuard<'static, ()> {
+    static GUARD: std::sync::Mutex<()> = std::sync::Mutex::new(());
+    GUARD.lock().unwrap_or_else(|e| e.into_inner())
+}
+
 /// Render the Windows named-pipe name for a user/home pair. Kept
 /// platform-agnostic so unit tests on any host can pin the output
 /// character set.
@@ -136,19 +151,12 @@ fn render_pipe_name(user: &str, home: &str) -> String {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use std::sync::Mutex;
     use tempfile::TempDir;
 
-    /// Serialise tests that mutate the global `BSK_HOME` env var.
-    fn env_guard() -> &'static Mutex<()> {
-        static GUARD: Mutex<()> = Mutex::new(());
-        &GUARD
-    }
-
     fn with_temp_home<F: FnOnce(&Path)>(f: F) {
-        let _lock = env_guard().lock().unwrap_or_else(|e| e.into_inner());
+        let _lock = test_env_lock();
         let tmp = TempDir::new().unwrap();
-        // SAFETY: serialised by env_guard above.
+        // SAFETY: serialised by test_env_lock above.
         unsafe {
             std::env::set_var(BSK_HOME_ENV, tmp.path().join("bsk"));
         }
@@ -187,6 +195,10 @@ mod tests {
                 record_session_path().unwrap(),
                 home.join("record-session.json")
             );
+            assert_eq!(
+                record_recovery_path().unwrap(),
+                home.join("record-recovery.json")
+            );
         });
     }
 

crates/bsk-cli/src/daemon/queue.rs

@@ -43,6 +43,7 @@ use serde_json::Value;
 use tokio::sync::{mpsc, oneshot};
 use tracing::{debug, warn};
 
+use super::abort::AbortToken;
 use super::browsers::BrowserRegistry;
 use super::inflight::{PromoteOutcome, ToolInflightEntry};
 use super::sessions::{SessionId, SessionRegistry};
@@ -83,6 +84,14 @@ pub struct ToolJob {
     /// IPC request id (e.g. `session.stop`'s queued teardown call —
     /// those already carry their own bespoke retry / abort path).
     pub inflight: Option<Arc<ToolInflightEntry>>,
+    /// Cancellation token for session-lifecycle work registered in the
+    /// daemon-local abort registry. Unlike `inflight`, lifecycle callers
+    /// forward their own WS cancel after the request has been queued.
+    pub lifecycle_cancel: Option<AbortToken>,
+    /// How long cancellation keeps the worker busy while the extension
+    /// finishes compensation. Session teardown uses its full RPC budget
+    /// so daemon state is reconciled even after the CLI exits.
+    pub cancel_cleanup_timeout: Duration,
     /// Owning session — used by `ToolInflightRegistry::cancel_session`
     /// to drain queued jobs that have not yet been promoted to
     /// "forwarded". Worker callers also have access to it via the
@@ -322,6 +331,8 @@ impl ToolQueueRegistry {
             timeout,
             false,
             inflight,
+            None,
+            CANCEL_CLEANUP_TIMEOUT,
         )
         .await;
         // A long-running request may outlive the idle threshold. Touching
@@ -360,6 +371,8 @@ impl ToolQueueRegistry {
             timeout,
             respond: respond_tx,
             inflight,
+            lifecycle_cancel: None,
+            cancel_cleanup_timeout: CANCEL_CLEANUP_TIMEOUT,
             session_id: sid.clone(),
         };
         match forward_one(sid, &self.browsers, &self.sessions, &job).await {
@@ -378,6 +391,7 @@ impl ToolQueueRegistry {
         method: Method,
         params: Value,
         timeout: Duration,
+        cancel: Option<AbortToken>,
     ) -> Result<Value, DispatchError> {
         let (sender, state) = {
             let mut guard = self.queues.lock().expect("tool queue registry poisoned");
@@ -404,6 +418,8 @@ impl ToolQueueRegistry {
             timeout,
             true,
             None,
+            cancel,
+            timeout,
         )
         .await
     }
@@ -419,7 +435,10 @@ async fn dispatch_with_sender(
     timeout: Duration,
     wait_for_capacity: bool,
     inflight: Option<Arc<ToolInflightEntry>>,
+    lifecycle_cancel: Option<AbortToken>,
+    cancel_cleanup_timeout: Duration,
 ) -> Result<Value, DispatchError> {
+    let effect_aware_transfer = is_effect_aware_transfer(&method);
     let (respond_tx, respond_rx) = oneshot::channel();
     let job = ToolJob {
         method,
@@ -427,8 +446,11 @@ async fn dispatch_with_sender(
         timeout,
         respond: respond_tx,
         inflight,
+        lifecycle_cancel,
+        cancel_cleanup_timeout,
         session_id,
     };
+    let lifecycle_cancellable = job.lifecycle_cancel.is_some();
     if wait_for_capacity {
         let waited = tokio::time::timeout(
             timeout.saturating_add(Duration::from_secs(1)),
@@ -453,8 +475,21 @@ async fn dispatch_with_sender(
             mpsc::error::TrySendError::Closed(_) => DispatchError::QueueClosed,
         });
     }
-    let waited =
-        tokio::time::timeout(timeout.saturating_add(Duration::from_secs(1)), respond_rx).await;
+    // Lifecycle teardown and effect-aware transfers both keep the worker busy
+    // for bounded compensation after their original deadline. Keep the outer
+    // waiter alive for the same grace period so it cannot abandon reconciliation.
+    let response_grace = if lifecycle_cancellable || effect_aware_transfer {
+        cancel_cleanup_timeout
+    } else {
+        Duration::ZERO
+    };
+    let waited = tokio::time::timeout(
+        timeout
+            .saturating_add(response_grace)
+            .saturating_add(Duration::from_secs(1)),
+        respond_rx,
+    )
+    .await;
     match waited {
         Ok(Ok(Ok(v))) => Ok(v),
         Ok(Ok(Err(rpc))) => Err(DispatchError::Rpc(rpc)),
@@ -492,11 +527,17 @@ async fn forward_one(
     // / browser resolution work, and before any WS frame leaves the
     // daemon (review C2). We keep the `tokio::select!` later so
     // cancels arriving mid-WS-hop also unblock the worker.
-    if let Some(entry) = job.inflight.as_ref()
-        && entry.is_cancelled()
+    if job
+        .inflight
+        .as_ref()
+        .is_some_and(|entry| entry.is_cancelled())
+        || job
+            .lifecycle_cancel
+            .as_ref()
+            .is_some_and(AbortToken::is_cancelled)
     {
         return Err(cancelled_error(
-            Some(entry),
+            job.inflight.as_deref(),
             "tool dispatch cancelled before forwarding",
         ));
     }
@@ -560,11 +601,20 @@ async fn forward_one(
             }
         }
         None => {
-            // Daemon-internal callers (e.g. queued session.stop drain)
-            // don't carry an inflight entry; they cannot be cancelled
-            // through the IPC `cancel` surface, so a plain sink send
-            // is sufficient and keeps the existing transport-error
-            // behaviour.
+            // Daemon-internal callers do not carry a tool-inflight entry.
+            // Session lifecycle jobs instead use the abort-registry token
+            // below; other internal callers keep the plain send path.
+            if job
+                .lifecycle_cancel
+                .as_ref()
+                .is_some_and(AbortToken::is_cancelled)
+            {
+                client.pending.lock().unwrap().cancel(&rpc_id);
+                return Err(cancelled_error(
+                    None,
+                    "session lifecycle cancelled before forwarding",
+                ));
+            }
             if client.sink.send(request).is_err() {
                 client.pending.lock().unwrap().cancel(&rpc_id);
                 return Err(RpcError {
@@ -573,46 +623,161 @@ async fn forward_one(
                     data: None,
                 });
             }
-            None
+            job.lifecycle_cancel.clone()
         }
     };
+    let forward_lifecycle_cancel = || {
+        if job.lifecycle_cancel.is_none() {
+            return;
+        }
+        let cancel = Frame::Request(RequestFrame {
+            id: format!("cancel-{rpc_id}"),
+            method: Method::Cancel,
+            params: Some(serde_json::json!({ "rpc_id": rpc_id })),
+        });
+        if let Err(err) = client.sink.send(cancel) {
+            warn!(%rpc_id, ?err, "failed to forward session lifecycle cancel");
+        }
+    };
+    let on_abort: Option<&(dyn Fn() + Sync)> = job
+        .lifecycle_cancel
+        .as_ref()
+        .map(|_| &forward_lifecycle_cancel as &(dyn Fn() + Sync));
+    let send_deadline_cancel = || {
+        client
+            .sink
+            .send(Frame::Request(RequestFrame {
+                id: format!("deadline-cancel-{rpc_id}"),
+                method: Method::Cancel,
+                params: Some(serde_json::json!({ "rpc_id": rpc_id })),
+            }))
+            .is_ok()
+    };
+    let deadline_cancel: Option<&(dyn Fn() -> bool + Sync)> =
+        is_effect_aware_transfer(&job.method).then_some(&send_deadline_cancel);
     let waited = await_with_optional_cancel(
         job.timeout,
-        CANCEL_CLEANUP_TIMEOUT,
+        job.cancel_cleanup_timeout,
         waiter,
         cancel_token.as_ref(),
+        on_abort,
+        deadline_cancel,
     )
     .await;
     let response = match waited {
         WaitOutcome::Response(resp) => resp,
         WaitOutcome::CancelledAfterResponse(resp) => {
-            // Cancellation wins the external verdict, but only after the
-            // extension's original RPC has settled. Preserve any non-cancel
-            // error so compensation failures remain explicit instead of
-            // being hidden behind a generic cancelled result.
-            if let ResponseBody::Err(err) = resp.body
-                && !matches!(err.code, ErrorCode::Cancelled | ErrorCode::UserAborted)
+            if job.method == Method::ToolSessionStop
+                && let ResponseBody::Ok(value) = &resp.body
+            {
+                // Teardown crossed its final irreversible boundary before
+                // the cancel landed. Commit the real extension result so the
+                // daemon does not retain a session whose window is gone.
+                return Ok(value.clone());
+            }
+            // File transfer commits are irreversible. A late cancel cannot
+            // overwrite a confirmed success, and an unknown transfer effect
+            // must remain explicit so callers do not retry or release upload
+            // staging as though nothing happened.
+            if is_effect_aware_transfer(&job.method) {
+                match &resp.body {
+                    ResponseBody::Ok(_) => resp,
+                    ResponseBody::Err(err)
+                        if transfer_effect(err) == Some("unknown")
+                            || transfer_effect(err) == Some("committed") =>
+                    {
+                        return Err(err.clone());
+                    }
+                    ResponseBody::Err(err)
+                        if !matches!(err.code, ErrorCode::Cancelled | ErrorCode::UserAborted) =>
+                    {
+                        return Err(err.clone());
+                    }
+                    ResponseBody::Err(_) => {
+                        return Err(cancelled_error(
+                            job.inflight.as_deref(),
+                            "tool dispatch cancelled after extension cleanup",
+                        ));
+                    }
+                }
+            } else {
+                // For ordinary tools cancellation keeps the existing verdict,
+                // while non-cancel errors still expose compensation failures.
+                if let ResponseBody::Err(err) = resp.body
+                    && !matches!(err.code, ErrorCode::Cancelled | ErrorCode::UserAborted)
+                {
+                    return Err(err);
+                }
+                return Err(cancelled_error(
+                    job.inflight.as_deref(),
+                    "tool dispatch cancelled after extension cleanup",
+                ));
+            }
+        }
+        WaitOutcome::TimedOutAfterResponse(resp) => match resp.body {
+            ResponseBody::Ok(value)
+                if job.method == Method::ToolSessionStop
+                    || is_effect_aware_transfer(&job.method) =>
+            {
+                // The close crossed its irreversible boundary during the
+                // timeout cleanup grace, or the transfer committed before its
+                // deadline cancel settled. Preserve the irreversible result.
+                return Ok(value);
+            }
+            ResponseBody::Err(err)
+                if !matches!(err.code, ErrorCode::Cancelled | ErrorCode::UserAborted) =>
             {
                 return Err(err);
             }
-            return Err(cancelled_error(
-                job.inflight.as_deref(),
-                "tool dispatch cancelled after extension cleanup",
-            ));
-        }
+            ResponseBody::Err(err) if is_effect_aware_transfer(&job.method) => {
+                return Err(timed_out_transfer_error(&err));
+            }
+            _ => {
+                return Err(RpcError {
+                    code: ErrorCode::Timeout,
+                    message: format!("tool RPC timed out after {:?}", job.timeout),
+                    data: None,
+                });
+            }
+        },
         WaitOutcome::CleanupTimeout => {
             client.pending.lock().unwrap().cancel(&rpc_id);
+            if is_effect_aware_transfer(&job.method) {
+                return Err(unknown_transfer_error(
+                    ErrorCode::Timeout,
+                    "cancelled file transfer did not confirm its outcome before cleanup timed out",
+                    "cleanup",
+                    true,
+                ));
+            }
             return Err(RpcError {
                 code: ErrorCode::Timeout,
                 message: format!(
                     "cancelled tool did not finish cleanup within {:?}",
-                    CANCEL_CLEANUP_TIMEOUT
+                    job.cancel_cleanup_timeout
                 ),
                 data: Some(serde_json::json!({ "reason": "cancel_cleanup_timeout" })),
             });
         }
+        WaitOutcome::TimeoutCleanupFailed => {
+            client.pending.lock().unwrap().cancel(&rpc_id);
+            return Err(unknown_transfer_error(
+                ErrorCode::Timeout,
+                "file transfer timed out and cleanup could not be confirmed",
+                "cleanup",
+                true,
+            ));
+        }
         WaitOutcome::WaiterClosed => {
             client.pending.lock().unwrap().cancel(&rpc_id);
+            if is_effect_aware_transfer(&job.method) {
+                return Err(unknown_transfer_error(
+                    ErrorCode::ProtocolError,
+                    "file transfer transport closed after dispatch; outcome is unknown",
+                    "transport",
+                    false,
+                ));
+            }
             return Err(RpcError {
                 code: ErrorCode::ProtocolError,
                 message: "transport closed mid-call".into(),
@@ -621,6 +786,14 @@ async fn forward_one(
         }
         WaitOutcome::Timeout => {
             client.pending.lock().unwrap().cancel(&rpc_id);
+            if is_effect_aware_transfer(&job.method) {
+                return Err(unknown_transfer_error(
+                    ErrorCode::Timeout,
+                    "file transfer timed out after dispatch; outcome is unknown",
+                    "transport",
+                    false,
+                ));
+            }
             return Err(RpcError {
                 code: ErrorCode::Timeout,
                 message: format!("tool RPC timed out after {:?}", job.timeout),
@@ -634,11 +807,65 @@ async fn forward_one(
     }
 }
 
+fn is_effect_aware_transfer(method: &Method) -> bool {
+    matches!(method, Method::ToolUpload | Method::ToolDownload)
+}
+
+fn transfer_effect(err: &RpcError) -> Option<&str> {
+    err.data.as_ref()?.get("effect_state")?.as_str()
+}
+
+fn unknown_transfer_error(
+    code: ErrorCode,
+    message: impl Into<String>,
+    phase: &str,
+    cleanup_failed: bool,
+) -> RpcError {
+    let mut data = serde_json::json!({
+        "reason": "transfer_outcome_unknown",
+        "effect_state": "unknown",
+        "phase": phase,
+    });
+    if cleanup_failed {
+        data["cleanup_state"] = serde_json::json!("failed");
+    }
+    RpcError {
+        code,
+        message: message.into(),
+        data: Some(data),
+    }
+}
+
+fn timed_out_transfer_error(err: &RpcError) -> RpcError {
+    let data = err.data.as_ref();
+    RpcError {
+        code: ErrorCode::Timeout,
+        message: "file transfer timed out after dispatch".into(),
+        data: Some(serde_json::json!({
+            "reason": "transfer_timeout",
+            "effect_state": data
+                .and_then(|value| value.get("effect_state"))
+                .and_then(Value::as_str)
+                .unwrap_or("unknown"),
+            "phase": data
+                .and_then(|value| value.get("phase"))
+                .and_then(Value::as_str)
+                .unwrap_or("cleanup"),
+            "cleanup_state": data
+                .and_then(|value| value.get("cleanup_state"))
+                .and_then(Value::as_str)
+                .unwrap_or("complete"),
+        })),
+    }
+}
+
 #[derive(Debug)]
 enum WaitOutcome {
     Response(bsk_protocol::ResponseFrame),
     CancelledAfterResponse(bsk_protocol::ResponseFrame),
+    TimedOutAfterResponse(bsk_protocol::ResponseFrame),
     CleanupTimeout,
+    TimeoutCleanupFailed,
     WaiterClosed,
     Timeout,
 }
@@ -648,6 +875,8 @@ async fn await_with_optional_cancel(
     cleanup_timeout: Duration,
     mut waiter: oneshot::Receiver<bsk_protocol::ResponseFrame>,
     cancel: Option<&super::abort::AbortToken>,
+    on_abort: Option<&(dyn Fn() + Sync)>,
+    on_deadline: Option<&(dyn Fn() -> bool + Sync)>,
 ) -> WaitOutcome {
     match cancel {
         Some(token) => {
@@ -659,6 +888,9 @@ async fn await_with_optional_cancel(
                 // after compensation or the cleanup deadline expires.
                 biased;
                 _ = token.cancelled() => {
+                    if let Some(on_abort) = on_abort {
+                        on_abort();
+                    }
                     match tokio::time::timeout(cleanup_timeout, &mut waiter).await {
                         Ok(Ok(resp)) => WaitOutcome::CancelledAfterResponse(resp),
                         Ok(Err(_)) => WaitOutcome::WaiterClosed,
@@ -669,13 +901,42 @@ async fn await_with_optional_cancel(
                     Ok(resp) => WaitOutcome::Response(resp),
                     Err(_) => WaitOutcome::WaiterClosed,
                 },
-                _ = &mut deadline => WaitOutcome::Timeout,
+                _ = &mut deadline => {
+                    if let Some(send_cancel) = on_deadline {
+                        return if send_cancel() {
+                            match tokio::time::timeout(cleanup_timeout, &mut waiter).await {
+                                Ok(Ok(resp)) => WaitOutcome::TimedOutAfterResponse(resp),
+                                Ok(Err(_)) | Err(_) => WaitOutcome::TimeoutCleanupFailed,
+                            }
+                        } else {
+                            WaitOutcome::TimeoutCleanupFailed
+                        };
+                    }
+                    let Some(on_abort) = on_abort else {
+                        return WaitOutcome::Timeout;
+                    };
+                    on_abort();
+                    match tokio::time::timeout(cleanup_timeout, &mut waiter).await {
+                        Ok(Ok(resp)) => WaitOutcome::TimedOutAfterResponse(resp),
+                        Ok(Err(_)) => WaitOutcome::WaiterClosed,
+                        Err(_) => WaitOutcome::CleanupTimeout,
+                    }
+                },
             }
         }
         None => match tokio::time::timeout(timeout, &mut waiter).await {
             Ok(Ok(resp)) => WaitOutcome::Response(resp),
             Ok(Err(_)) => WaitOutcome::WaiterClosed,
-            Err(_) => WaitOutcome::Timeout,
+            Err(_) => match on_deadline {
+                Some(send_cancel) if send_cancel() => {
+                    match tokio::time::timeout(cleanup_timeout, &mut waiter).await {
+                        Ok(Ok(resp)) => WaitOutcome::TimedOutAfterResponse(resp),
+                        Ok(Err(_)) | Err(_) => WaitOutcome::TimeoutCleanupFailed,
+                    }
+                }
+                Some(_) => WaitOutcome::TimeoutCleanupFailed,
+                None => WaitOutcome::Timeout,
+            },
         },
     }
 }
@@ -759,6 +1020,8 @@ mod await_with_optional_cancel_tests {
             Duration::from_secs(1),
             rx,
             Some(&token),
+            None,
+            None,
         )
         .await;
         assert!(
@@ -781,6 +1044,8 @@ mod await_with_optional_cancel_tests {
             Duration::from_secs(1),
             rx,
             Some(&token),
+            None,
+            None,
         )
         .await;
         match outcome {
@@ -795,9 +1060,15 @@ mod await_with_optional_cancel_tests {
     async fn no_cancel_token_still_returns_response() {
         let (tx, rx) = oneshot::channel();
         tx.send(dummy_response()).unwrap();
-        let outcome =
-            await_with_optional_cancel(Duration::from_secs(10), Duration::from_secs(1), rx, None)
-                .await;
+        let outcome = await_with_optional_cancel(
+            Duration::from_secs(10),
+            Duration::from_secs(1),
+            rx,
+            None,
+            None,
+            None,
+        )
+        .await;
         assert!(matches!(outcome, WaitOutcome::Response(_)));
     }
 
@@ -812,6 +1083,8 @@ mod await_with_optional_cancel_tests {
                 Duration::from_secs(1),
                 rx,
                 Some(&token),
+                None,
+                None,
             )
             .await
         });
@@ -835,10 +1108,75 @@ mod await_with_optional_cancel_tests {
             Duration::from_millis(10),
             rx,
             Some(&token),
+            None,
+            None,
         )
         .await;
         assert!(matches!(outcome, WaitOutcome::CleanupTimeout));
     }
+
+    #[tokio::test]
+    async fn lifecycle_timeout_forwards_abort_and_waits_for_cleanup_response() {
+        use std::sync::atomic::{AtomicBool, Ordering};
+
+        let token = AbortToken::new();
+        let (tx, rx) = oneshot::channel();
+        let abort_forwarded = AtomicBool::new(false);
+        let forward_abort = || abort_forwarded.store(true, Ordering::SeqCst);
+        tokio::spawn(async move {
+            tokio::time::sleep(Duration::from_millis(20)).await;
+            tx.send(dummy_response()).unwrap();
+        });
+
+        let outcome = await_with_optional_cancel(
+            Duration::from_millis(5),
+            Duration::from_millis(100),
+            rx,
+            Some(&token),
+            Some(&forward_abort),
+            None,
+        )
+        .await;
+
+        assert!(abort_forwarded.load(Ordering::SeqCst));
+        assert!(matches!(outcome, WaitOutcome::TimedOutAfterResponse(_)));
+    }
+
+    #[tokio::test]
+    async fn transfer_deadline_sends_cancel_and_waits_for_cleanup_response() {
+        let (tx, rx) = oneshot::channel();
+        tokio::spawn(async move {
+            tokio::time::sleep(Duration::from_millis(5)).await;
+            tx.send(dummy_response()).unwrap();
+        });
+        let send_cancel = || true;
+        let outcome = await_with_optional_cancel(
+            Duration::from_millis(1),
+            Duration::from_secs(1),
+            rx,
+            None,
+            None,
+            Some(&send_cancel),
+        )
+        .await;
+        assert!(matches!(outcome, WaitOutcome::TimedOutAfterResponse(_)));
+    }
+
+    #[tokio::test]
+    async fn transfer_deadline_is_unknown_when_cancel_cannot_be_sent() {
+        let (_tx, rx) = oneshot::channel();
+        let send_cancel = || false;
+        let outcome = await_with_optional_cancel(
+            Duration::from_millis(1),
+            Duration::from_secs(1),
+            rx,
+            None,
+            None,
+            Some(&send_cancel),
+        )
+        .await;
+        assert!(matches!(outcome, WaitOutcome::TimeoutCleanupFailed));
+    }
 }
 
 #[cfg(test)]
@@ -901,6 +1239,8 @@ mod tool_job_session_id_tests {
             timeout: Duration::from_secs(1),
             respond: tx,
             inflight: None,
+            lifecycle_cancel: None,
+            cancel_cleanup_timeout: CANCEL_CLEANUP_TIMEOUT,
             session_id: SessionId("sess-A".into()),
         };
         assert_eq!(job.session_id.0, "sess-A");

crates/bsk-cli/src/daemon/sessions.rs

@@ -12,13 +12,14 @@ use bsk_protocol::system::{BrowserStatusEntry, SessionStatusEntry};
 use bsk_protocol::tools::{
     SessionStartParams, SessionStartResult, SessionStopParams, SessionStopResult,
 };
-use bsk_protocol::{Frame, RequestFrame, ResponseBody, RpcError, RpcId};
+use bsk_protocol::{ErrorCode, Frame, Method, RequestFrame, ResponseBody, RpcError, RpcId};
 use rand::Rng;
 use serde::{Deserialize, Serialize};
 use tokio::time::{Duration, timeout};
 
+use super::abort::AbortToken;
 use super::browsers::{BrowserClient, BrowserId, BrowserRegistry, SelectError};
-use super::queue::{DispatchError, ToolQueueRegistry};
+use super::queue::{CANCEL_CLEANUP_TIMEOUT, DispatchError, ToolQueueRegistry};
 use super::session_interrupt::SessionInterruptRegistry;
 
 #[derive(Debug, Clone, PartialEq, Eq, Hash, Serialize, Deserialize)]
@@ -339,6 +340,14 @@ pub enum StartSessionError {
     IdExhausted,
     #[error("session creation timed out waiting for extension")]
     Timeout,
+    #[error("session creation was cancelled")]
+    Cancelled,
+    #[error("session creation cancellation cleanup failed for {session_id}: {message}")]
+    CleanupFailed {
+        session_id: SessionId,
+        agent_window_id: Option<i64>,
+        message: String,
+    },
     #[error("extension rejected tool.session_start: {0:?}")]
     ExtensionError(RpcError),
     #[error("transport closed while waiting for extension response")]
@@ -354,6 +363,8 @@ impl StartSessionError {
             StartSessionError::AmbiguousBrowserLabel { .. } => "invalid_params",
             StartSessionError::IdExhausted => "protocol_error",
             StartSessionError::Timeout => "timeout",
+            StartSessionError::Cancelled => "cancelled",
+            StartSessionError::CleanupFailed { .. } => "protocol_error",
             StartSessionError::TransportClosed => "protocol_error",
             StartSessionError::ExtensionError(err) => match err.code {
                 bsk_protocol::ErrorCode::Timeout => "timeout",
@@ -407,6 +418,7 @@ pub struct AgentWindowOptions {
 /// On success the matching per-session dispatch queue (M6.5,
 /// design §5) is also spawned so that subsequent `tool.*` RPCs serialise
 /// against this session's Agent Window / ref-store.
+#[allow(clippy::too_many_arguments)]
 pub async fn start_session(
     registry: &Arc<BrowserRegistry>,
     sessions: &Arc<SessionRegistry>,
@@ -415,24 +427,32 @@ pub async fn start_session(
     window: AgentWindowOptions,
     connect_wait: Duration,
     timeout_dur: Duration,
+    cancel: Option<AbortToken>,
 ) -> Result<Session, StartSessionError> {
-    let client: Arc<BrowserClient> = registry
-        .select_with_connect_wait(requested, connect_wait)
-        .await
-        .map_err(|e| match e {
-            SelectError::NoBrowserConnected => StartSessionError::NoBrowserConnected,
-            SelectError::MultipleBrowsersOnline => StartSessionError::MultipleBrowsersOnline {
-                browsers: snapshot_status_entries(registry, sessions),
-            },
-            SelectError::NotFound => StartSessionError::BrowserNotFound,
-            SelectError::AmbiguousLabel {
-                label,
-                instance_ids,
-            } => StartSessionError::AmbiguousBrowserLabel {
-                label,
-                instance_ids,
-            },
-        })?;
+    let selection = registry.select_with_connect_wait(requested, connect_wait);
+    tokio::pin!(selection);
+    let selected = match cancel.as_ref() {
+        Some(token) => tokio::select! {
+            biased;
+            _ = token.cancelled() => return Err(StartSessionError::Cancelled),
+            selected = &mut selection => selected,
+        },
+        None => selection.await,
+    };
+    let client: Arc<BrowserClient> = selected.map_err(|e| match e {
+        SelectError::NoBrowserConnected => StartSessionError::NoBrowserConnected,
+        SelectError::MultipleBrowsersOnline => StartSessionError::MultipleBrowsersOnline {
+            browsers: snapshot_status_entries(registry, sessions),
+        },
+        SelectError::NotFound => StartSessionError::BrowserNotFound,
+        SelectError::AmbiguousLabel {
+            label,
+            instance_ids,
+        } => StartSessionError::AmbiguousBrowserLabel {
+            label,
+            instance_ids,
+        },
+    })?;
     let session_id = sessions
         .reserve_id(client.id.clone(), SESSION_ID_MAX_RESERVE_ATTEMPTS, now_ms)
         .ok_or(StartSessionError::IdExhausted)?;
@@ -453,22 +473,71 @@ pub async fn start_session(
         let mut pending = client.pending.lock().unwrap();
         pending.register(rpc_id.clone())
     };
+    if cancel.as_ref().is_some_and(AbortToken::is_cancelled) {
+        sessions.cancel_reservation(&session_id);
+        client.pending.lock().unwrap().cancel(&rpc_id);
+        return Err(StartSessionError::Cancelled);
+    }
     if client.sink.send(Frame::Request(request)).is_err() {
         sessions.cancel_reservation(&session_id);
         client.pending.lock().unwrap().cancel(&rpc_id);
         return Err(StartSessionError::TransportClosed);
     }
-    let response = match timeout(timeout_dur, waiter).await {
-        Ok(Ok(resp)) => resp,
-        Ok(Err(_)) => {
+    let mut waiter = waiter;
+    enum StartWaitOutcome {
+        Response(bsk_protocol::ResponseFrame),
+        WaiterClosed,
+        Cancelled,
+        Timeout,
+    }
+    let wait_outcome = match cancel.as_ref() {
+        Some(token) => {
+            let deadline = tokio::time::sleep(timeout_dur);
+            tokio::pin!(deadline);
+            tokio::select! {
+                biased;
+                _ = token.cancelled() => StartWaitOutcome::Cancelled,
+                response = &mut waiter => match response {
+                    Ok(response) => StartWaitOutcome::Response(response),
+                    Err(_) => StartWaitOutcome::WaiterClosed,
+                },
+                _ = &mut deadline => StartWaitOutcome::Timeout,
+            }
+        }
+        None => match timeout(timeout_dur, &mut waiter).await {
+            Ok(Ok(response)) => StartWaitOutcome::Response(response),
+            Ok(Err(_)) => StartWaitOutcome::WaiterClosed,
+            Err(_) => StartWaitOutcome::Timeout,
+        },
+    };
+    let response = match wait_outcome {
+        StartWaitOutcome::Response(response) => response,
+        StartWaitOutcome::WaiterClosed => {
             sessions.cancel_reservation(&session_id);
             client.pending.lock().unwrap().cancel(&rpc_id);
             return Err(StartSessionError::TransportClosed);
         }
-        Err(_) => {
-            sessions.cancel_reservation(&session_id);
-            client.pending.lock().unwrap().cancel(&rpc_id);
-            return Err(StartSessionError::Timeout);
+        StartWaitOutcome::Cancelled => {
+            return Err(finish_aborted_start(
+                &client,
+                sessions,
+                &session_id,
+                &rpc_id,
+                waiter,
+                StartAbortReason::Cancelled,
+            )
+            .await);
+        }
+        StartWaitOutcome::Timeout => {
+            return Err(finish_aborted_start(
+                &client,
+                sessions,
+                &session_id,
+                &rpc_id,
+                waiter,
+                StartAbortReason::Timeout,
+            )
+            .await);
         }
     };
     let agent_window_id = match response.body {
@@ -504,6 +573,124 @@ pub async fn start_session(
     Ok(session)
 }
 
+#[derive(Clone, Copy)]
+enum StartAbortReason {
+    Cancelled,
+    Timeout,
+}
+
+impl StartAbortReason {
+    fn error(self) -> StartSessionError {
+        match self {
+            Self::Cancelled => StartSessionError::Cancelled,
+            Self::Timeout => StartSessionError::Timeout,
+        }
+    }
+}
+
+async fn finish_aborted_start(
+    client: &Arc<BrowserClient>,
+    sessions: &Arc<SessionRegistry>,
+    session_id: &SessionId,
+    rpc_id: &RpcId,
+    mut waiter: tokio::sync::oneshot::Receiver<bsk_protocol::ResponseFrame>,
+    reason: StartAbortReason,
+) -> StartSessionError {
+    sessions.cancel_reservation(session_id);
+    let cancel = Frame::Request(RequestFrame {
+        id: format!("cancel-{rpc_id}"),
+        method: Method::Cancel,
+        params: Some(serde_json::json!({ "rpc_id": rpc_id })),
+    });
+    if client.sink.send(cancel).is_err() {
+        client.pending.lock().unwrap().cancel(rpc_id);
+        return StartSessionError::TransportClosed;
+    }
+
+    match timeout(CANCEL_CLEANUP_TIMEOUT, &mut waiter).await {
+        Ok(Ok(response)) => match response.body {
+            ResponseBody::Err(err)
+                if matches!(err.code, ErrorCode::Cancelled | ErrorCode::UserAborted) =>
+            {
+                reason.error()
+            }
+            ResponseBody::Err(err) => StartSessionError::ExtensionError(err),
+            ResponseBody::Ok(value) => {
+                let agent_window_id = serde_json::from_value::<SessionStartResult>(value)
+                    .ok()
+                    .and_then(|result| result.agent_window_id);
+                match rollback_extension_session(client, session_id).await {
+                    Ok(()) => reason.error(),
+                    Err(message) => StartSessionError::CleanupFailed {
+                        session_id: session_id.clone(),
+                        agent_window_id,
+                        message,
+                    },
+                }
+            }
+        },
+        Ok(Err(_)) => {
+            client.pending.lock().unwrap().cancel(rpc_id);
+            StartSessionError::TransportClosed
+        }
+        Err(_) => {
+            // The cancel frame is already queued behind session_start. Even
+            // if chrome.windows.create is still blocked, the extension will
+            // observe the aborted signal and compensate before replying.
+            client.pending.lock().unwrap().cancel(rpc_id);
+            reason.error()
+        }
+    }
+}
+
+async fn rollback_extension_session(
+    client: &Arc<BrowserClient>,
+    session_id: &SessionId,
+) -> Result<(), String> {
+    let rollback_id = next_rpc_id("sess-start-rollback");
+    let waiter = {
+        let mut pending = client.pending.lock().unwrap();
+        pending.register(rollback_id.clone())
+    };
+    let request = RequestFrame {
+        id: rollback_id.clone(),
+        method: Method::ToolSessionStop,
+        params: Some(
+            serde_json::to_value(SessionStopParams {
+                session_id: session_id.0.clone(),
+            })
+            .expect("serialize session_start rollback params"),
+        ),
+    };
+    if client.sink.send(Frame::Request(request)).is_err() {
+        client.pending.lock().unwrap().cancel(&rollback_id);
+        return Err("browser disconnected before rollback was queued".into());
+    }
+    let response = match timeout(CANCEL_CLEANUP_TIMEOUT, waiter).await {
+        Ok(Ok(response)) => response,
+        Ok(Err(_)) => return Err("browser disconnected during rollback".into()),
+        Err(_) => {
+            client.pending.lock().unwrap().cancel(&rollback_id);
+            return Err(format!(
+                "extension did not finish rollback within {CANCEL_CLEANUP_TIMEOUT:?}"
+            ));
+        }
+    };
+    match response.body {
+        ResponseBody::Ok(value) => {
+            let result: SessionStopResult = serde_json::from_value(value)
+                .map_err(|err| format!("invalid rollback response: {err}"))?;
+            if result.return_failures.is_empty() {
+                Ok(())
+            } else {
+                Err("rollback reported borrowed-tab return failures".into())
+            }
+        }
+        ResponseBody::Err(err) if err.code == ErrorCode::NotFound => Ok(()),
+        ResponseBody::Err(err) => Err(format!("rollback failed: {} ({:?})", err.message, err.code)),
+    }
+}
+
 /// Tear down a single session id (round-trips `tool.session_stop` to the
 /// owning extension and unregisters on success). Also tears down the
 /// per-session dispatch queue so the worker task exits.
@@ -514,6 +701,7 @@ pub async fn stop_session(
     interrupts: &Arc<SessionInterruptRegistry>,
     session_id: &SessionId,
     timeout_dur: Duration,
+    cancel: Option<AbortToken>,
 ) -> Result<SessionStopResult, StopSessionError> {
     let session = sessions.get(session_id).ok_or(StopSessionError::NotFound)?;
     if registry.get(&session.browser_id).is_none() {
@@ -524,12 +712,17 @@ pub async fn stop_session(
         session_id: session_id.0.clone(),
     })
     .unwrap();
+    // Internal callers (idle reaping and shutdown) do not have a CLI-owned
+    // token, but still need timeout-driven WS cancellation so a late window
+    // close cannot leave daemon state pointing at a vanished session.
+    let lifecycle_cancel = Some(cancel.unwrap_or_default());
     match queues
         .dispatch_after_closing(
             session_id,
             bsk_protocol::Method::ToolSessionStop,
             params,
             timeout_dur,
+            lifecycle_cancel,
         )
         .await
     {
@@ -567,6 +760,10 @@ pub async fn stop_session(
                 drop_session_local(queues, interrupts, session_id);
                 return Ok(SessionStopResult::default());
             }
+            if err.code == bsk_protocol::ErrorCode::Timeout {
+                queues.reopen(session_id);
+                return Err(StopSessionError::Timeout);
+            }
             queues.reopen(session_id);
             Err(StopSessionError::ExtensionError(err))
         }

crates/bsk-cli/src/daemon/start.rs

@@ -15,7 +15,6 @@ use std::net::{IpAddr, Ipv4Addr, SocketAddr};
 use std::path::Path;
 use std::sync::Arc;
 use std::sync::Mutex;
-use std::sync::atomic::{AtomicUsize, Ordering};
 use std::time::{Duration, Instant};
 
 use anyhow::{Context, Result};
@@ -27,7 +26,7 @@ use crate::daemon::{
     browsers::{BROWSER_LIVENESS_TICK, BROWSER_LIVENESS_TIMEOUT, EXTENSION_CONNECT_WAIT},
     info as daemon_info, ipc, lockfile, paths,
     sessions::{StopSessionError, forget_session, stop_session},
-    state::DaemonState,
+    state::{DaemonState, PROTOCOL_VERSION},
     ws,
 };
 
@@ -226,11 +225,15 @@ pub fn run_foreground(cfg: DaemonConfig) -> Result<()> {
             .with_context(|| format!("bind IPC endpoint {}", sock_path.display()))?;
 
         let state = Arc::new(DaemonState::new(cfg.clone()));
+        state
+            .transfers
+            .initialize()
+            .context("initialize transfer staging")?;
         let session_idle_task = spawn_session_idle_reaper(Arc::clone(&state));
         let browser_liveness_task = spawn_browser_liveness_reaper(Arc::clone(&state));
         // Fired by the update check task after a successful auto-update:
-        // the replacement daemon has already been spawned, so this
-        // process should shut down and let it take over.
+        // the replacement daemon (or Windows update helper) is ready, so
+        // this process should shut down and let it take over.
         let restart_notify = Arc::new(tokio::sync::Notify::new());
         let update_check_task =
             spawn_update_check_task(Arc::clone(&state), Arc::clone(&restart_notify));
@@ -291,15 +294,14 @@ pub fn run_foreground(cfg: DaemonConfig) -> Result<()> {
         });
 
         let started_at = Instant::now();
-        let activity: Arc<Mutex<Instant>> = Arc::new(Mutex::new(started_at));
-        let active_ipc_connections = Arc::new(AtomicUsize::new(0));
+        let activity = Arc::new(Mutex::new(IpcActivityState::new(started_at)));
         let (ipc_shutdown_tx, ipc_shutdown_rx) = tokio::sync::oneshot::channel::<()>();
         let status = ipc::DaemonStatus {
             started_at,
             ws_port,
             sock_path: sock_path.clone(),
             daemon_version: env!("CARGO_PKG_VERSION"),
-            protocol_version: "1.0",
+            protocol_version: PROTOCOL_VERSION,
         };
         let handler = ipc::full_handler(status, Arc::clone(&state));
 
@@ -314,18 +316,14 @@ pub fn run_foreground(cfg: DaemonConfig) -> Result<()> {
             let handler = handler.clone();
             let ipc_open = {
                 let activity = activity.clone();
-                let active = active_ipc_connections.clone();
                 move || {
-                    active.fetch_add(1, Ordering::SeqCst);
-                    record_activity(&activity);
+                    record_ipc_open(&activity);
                 }
             };
             let ipc_close = {
                 let activity = activity.clone();
-                let active = active_ipc_connections.clone();
                 move || {
-                    active.fetch_sub(1, Ordering::SeqCst);
-                    record_activity(&activity);
+                    record_ipc_close(&activity);
                 }
             };
             tokio::spawn(ipc::serve(
@@ -344,7 +342,6 @@ pub fn run_foreground(cfg: DaemonConfig) -> Result<()> {
         let idle_task = {
             let activity = activity.clone();
             let state = Arc::clone(&state);
-            let active_ipc_connections = active_ipc_connections.clone();
             let daemon_idle = cfg.daemon_idle;
             tokio::spawn(async move {
                 let tick = (daemon_idle / 4).max(Duration::from_millis(250));
@@ -352,22 +349,19 @@ pub fn run_foreground(cfg: DaemonConfig) -> Result<()> {
                 ticker.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
                 loop {
                     ticker.tick().await;
-                    let last = match activity.lock() {
-                        Ok(a) => *a,
-                        Err(p) => *p.into_inner(),
-                    };
-                    if last.elapsed() < daemon_idle {
+                    let ipc_is_idle =
+                        lock_activity(&activity).should_exit_for_idle(Instant::now(), daemon_idle);
+                    if !ipc_is_idle {
                         continue;
                     }
-                    // Bridge from M2/M3 (which only knew about connection
-                    // counts) to M4/M5 (browser + session registries):
-                    // hold the daemon alive while any IPC client is
-                    // connected, any browser is paired, or any session
-                    // is live (design §3.2).
-                    if active_ipc_connections.load(Ordering::SeqCst) > 0
-                        || !state.browsers.is_empty()
-                        || !state.sessions.is_empty()
-                    {
+                    // IPC liveness (open-connection count + idle window) is already
+                    // enforced above by `should_exit_for_idle`, which reads
+                    // `IpcActivityState` under one lock and returns true only when
+                    // there are zero connections AND the idle interval has elapsed.
+                    // Here we additionally hold the daemon alive while any browser
+                    // is paired or any session is live (design §3.2, M4/M5
+                    // registries).
+                    if !state.browsers.is_empty() || !state.sessions.is_empty() {
                         continue;
                     }
                     info!(
@@ -458,6 +452,7 @@ pub(crate) fn spawn_browser_liveness_reaper(
                     for s in state.sessions.purge_browser(&client.id) {
                         state.tool_queues.remove(&s.id);
                         state.session_interrupts.drop_session(&s.id);
+                        state.transfers.release_session(&s.id.0);
                         debug!(session = %s.id, "purged session on browser liveness timeout");
                     }
                 }
@@ -491,10 +486,14 @@ pub(crate) fn spawn_session_idle_reaper(state: Arc<DaemonState>) -> tokio::task:
                     &state.session_interrupts,
                     &session_id,
                     Duration::from_secs(10),
+                    None,
                 )
                 .await
                 {
-                    Ok(_) => info!(session = %session_id, "idle session stopped"),
+                    Ok(_) => {
+                        state.transfers.release_session(&session_id.0);
+                        info!(session = %session_id, "idle session stopped");
+                    }
                     Err(StopSessionError::SessionBusy | StopSessionError::Stopping) => {
                         debug!(session = %session_id, "idle session still active; retrying later");
                     }
@@ -505,6 +504,7 @@ pub(crate) fn spawn_session_idle_reaper(state: Arc<DaemonState>) -> tokio::task:
                             &state.session_interrupts,
                             &session_id,
                         );
+                        state.transfers.release_session(&session_id.0);
                     }
                     Err(err) => {
                         warn!(session = %session_id, error = %err, "failed to stop idle session");
@@ -539,8 +539,8 @@ pub(crate) fn spawn_session_idle_reaper(state: Arc<DaemonState>) -> tokio::task:
 /// task spawns the replacement daemon (see
 /// [`DAEMON_REPLACEMENT_WAIT_ENV`]) and fires `restart` so this process
 /// shuts down and the new version takes over; on Windows the
-/// replacement can only be staged, so it just logs that a restart is
-/// needed.
+/// helper waits for this process to exit, replaces the binary, and starts
+/// the new daemon with the same configuration.
 pub(crate) fn spawn_update_check_task(
     state: Arc<DaemonState>,
     restart: Arc<tokio::sync::Notify>,
@@ -605,7 +605,11 @@ pub(crate) fn spawn_update_check_task(
                         |candidate| {
                             let target =
                                 exe_path.as_deref().context("current executable unknown")?;
-                            update::self_install_candidate(candidate, target)
+                            update::self_install_candidate(
+                                candidate,
+                                target,
+                                &restart_start_args(&state.config),
+                            )
                         },
                     )
                 })
@@ -635,10 +639,11 @@ pub(crate) fn spawn_update_check_task(
                     sessions,
                     "auto-update postponed: agent session(s) active; will retry on the next tick"
                 ),
-                update::AutoUpdateOutcome::Staged { latest } => warn!(
-                    %latest,
-                    "auto-update staged the new binary but cannot replace the running daemon in place; restart the daemon (or terminal) to finish the upgrade"
-                ),
+                update::AutoUpdateOutcome::Staged { latest } => {
+                    info!(%latest, "update helper ready; exiting so it can replace and restart the daemon");
+                    restart.notify_one();
+                    return;
+                }
                 update::AutoUpdateOutcome::Replaced { latest } => {
                     info!(
                         current = env!("CARGO_PKG_VERSION"),
@@ -681,10 +686,55 @@ fn restart_start_args(cfg: &DaemonConfig) -> StartArgs {
     }
 }
 
-fn record_activity(activity: &Arc<Mutex<Instant>>) {
-    if let Ok(mut a) = activity.lock() {
-        *a = Instant::now();
+#[derive(Debug)]
+struct IpcActivityState {
+    last_activity: Instant,
+    active_connections: usize,
+}
+
+impl IpcActivityState {
+    fn new(now: Instant) -> Self {
+        Self {
+            last_activity: now,
+            active_connections: 0,
+        }
     }
+
+    fn touch(&mut self, now: Instant) {
+        self.last_activity = now;
+    }
+
+    fn opened(&mut self, now: Instant) {
+        self.active_connections = self.active_connections.saturating_add(1);
+        self.last_activity = now;
+    }
+
+    fn closed(&mut self, now: Instant) {
+        self.active_connections = self.active_connections.saturating_sub(1);
+        self.last_activity = now;
+    }
+
+    fn should_exit_for_idle(&self, now: Instant, idle: Duration) -> bool {
+        self.active_connections == 0 && now.saturating_duration_since(self.last_activity) >= idle
+    }
+}
+
+fn lock_activity(
+    activity: &Arc<Mutex<IpcActivityState>>,
+) -> std::sync::MutexGuard<'_, IpcActivityState> {
+    activity.lock().unwrap_or_else(|poison| poison.into_inner())
+}
+
+fn record_activity(activity: &Arc<Mutex<IpcActivityState>>) {
+    lock_activity(activity).touch(Instant::now());
+}
+
+fn record_ipc_open(activity: &Arc<Mutex<IpcActivityState>>) {
+    lock_activity(activity).opened(Instant::now());
+}
+
+fn record_ipc_close(activity: &Arc<Mutex<IpcActivityState>>) {
+    lock_activity(activity).closed(Instant::now());
 }
 
 /// Initialise tracing for the daemon: write a daily-rotated file in
@@ -1067,4 +1117,37 @@ mod tests {
         assert_eq!(args.session_idle, Some(Duration::from_secs(11)));
         assert_eq!(args.daemon_idle, Some(Duration::from_secs(22)));
     }
+
+    #[test]
+    fn ipc_close_restarts_the_idle_window_atomically() {
+        let started = Instant::now();
+        let idle = Duration::from_secs(10);
+        let mut activity = IpcActivityState::new(started);
+
+        activity.opened(started);
+        let closed_at = started + Duration::from_secs(30);
+        assert!(
+            !activity.should_exit_for_idle(closed_at, idle),
+            "an open IPC connection must hold the daemon alive"
+        );
+
+        activity.closed(closed_at);
+        assert_eq!(activity.active_connections, 0);
+        assert_eq!(activity.last_activity, closed_at);
+        assert!(!activity.should_exit_for_idle(closed_at, idle));
+        assert!(!activity.should_exit_for_idle(closed_at + idle - Duration::from_millis(1), idle));
+        assert!(activity.should_exit_for_idle(closed_at + idle, idle));
+    }
+
+    #[test]
+    fn ipc_activity_keeps_the_daemon_alive_until_a_full_idle_interval_passes() {
+        let started = Instant::now();
+        let idle = Duration::from_secs(10);
+        let mut activity = IpcActivityState::new(started);
+        let request_at = started + Duration::from_secs(20);
+
+        activity.touch(request_at);
+        assert!(!activity.should_exit_for_idle(request_at + Duration::from_secs(9), idle));
+        assert!(activity.should_exit_for_idle(request_at + idle, idle));
+    }
 }

crates/bsk-cli/src/daemon/state.rs

@@ -7,6 +7,7 @@ use tokio::task::JoinHandle;
 
 use super::abort::AbortRegistry;
 use super::browsers::BrowserRegistry;
+use super::file_transfer::TransferRegistry;
 use super::inflight::ToolInflightRegistry;
 use super::ipc::IpcHandle;
 use super::queue::ToolQueueRegistry;
@@ -16,7 +17,7 @@ use super::start::DaemonConfig;
 use super::ws::WsHandle;
 
 pub const DAEMON_VERSION: &str = env!("CARGO_PKG_VERSION");
-pub const PROTOCOL_VERSION: &str = "1.0";
+pub const PROTOCOL_VERSION: &str = "1.1";
 /// Lowest **protocol** version peers must speak (e.g. `"1.0"`).
 pub const MIN_COMPATIBLE_PROTOCOL: &str = "1.0";
 /// Legacy app-semver floor used only when `HandshakeResult.min_compatible_peer`
@@ -34,8 +35,8 @@ pub struct DaemonState {
     /// `stop_session` / browser disconnect.
     pub tool_queues: Arc<ToolQueueRegistry>,
     /// Per-rpc-id cancellation tokens for daemon-side long-runners
-    /// (M9.3 — currently only `tool.wait_ms`). The CLI's `cancel
-    /// { rpc_id }` consults this registry first.
+    /// (`tool.wait_ms` plus `session.*` lifecycle calls). The CLI's
+    /// `cancel { rpc_id }` consults this registry first.
     pub abort_registry: Arc<AbortRegistry>,
     /// Tracks `tool.*` RPCs that have been forwarded to an extension
     /// over WS but have not yet received a response. Indexed by the
@@ -51,6 +52,9 @@ pub struct DaemonState {
     /// `SessionRegistry` because the signal is a transient runtime
     /// control state.
     pub session_interrupts: Arc<SessionInterruptRegistry>,
+    /// Operation-scoped local file staging. The extension only sees paths
+    /// minted here; agent-facing RPCs use opaque transfer ids.
+    pub transfers: Arc<TransferRegistry>,
 }
 
 impl DaemonState {
@@ -64,6 +68,7 @@ impl DaemonState {
         ));
         let abort_registry = Arc::new(AbortRegistry::new());
         let session_interrupts = Arc::new(SessionInterruptRegistry::new());
+        let transfers = Arc::new(TransferRegistry::new().expect("initialise transfer staging"));
         Self {
             config,
             browsers,
@@ -72,6 +77,7 @@ impl DaemonState {
             abort_registry,
             tool_inflight,
             session_interrupts,
+            transfers,
         }
     }
 }

crates/bsk-cli/src/daemon/ws.rs

@@ -295,6 +295,7 @@ async fn drive_connection(
     for session in state.sessions.purge_browser(&browser_id) {
         state.tool_queues.remove(&session.id);
         state.session_interrupts.drop_session(&session.id);
+        state.transfers.release_session(&session.id.0);
         debug!(session = %session.id, "purged stale session before browser reconnect");
     }
     let (tx, mut rx) = mpsc::unbounded_channel::<Frame>();
@@ -407,6 +408,7 @@ async fn drive_connection(
         for s in state.sessions.purge_browser(&browser_id) {
             state.tool_queues.remove(&s.id);
             state.session_interrupts.drop_session(&s.id);
+            state.transfers.release_session(&s.id.0);
             debug!(session = %s.id, "purged session on browser disconnect");
         }
     } else {
@@ -503,6 +505,7 @@ fn handle_session_window_closed(
         &state.session_interrupts,
         &session_id,
     ) {
+        state.transfers.release_session(&session_id.0);
         info!(session = %session_id, "session removed: user closed Agent Window");
     } else {
         debug!(session = %session_id, "session.window_closed for unknown session id");

crates/bsk-cli/src/main.rs

@@ -99,6 +99,8 @@ fn dispatch(cli: Cli, format: Format) -> Result<(), CliError> {
         Command::Fill(args) => cli::interaction::dispatch_fill(args, format),
         Command::Press(args) => cli::interaction::dispatch_press(args, format),
         Command::Select(args) => cli::interaction::dispatch_select(args, format),
+        Command::Upload(args) => cli::upload::dispatch(args, format),
+        Command::Download(args) => cli::download::dispatch(args, format),
         Command::Evaluate(args) => cli::evaluate::dispatch(args, format),
         Command::WaitForNavigation(args) => cli::waits::dispatch_wait_for_navigation(args, format),
         Command::WaitMs(args) => cli::waits::dispatch_wait_ms(args, format),