@team-agent/installer 0.5.63 → 0.5.65

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (37) hide show
  1. package/Cargo.lock +1 -1
  2. package/Cargo.toml +1 -1
  3. package/crates/team-agent/src/cli/adapters.rs +5 -0
  4. package/crates/team-agent/src/cli/emit.rs +15 -0
  5. package/crates/team-agent/src/cli/mod.rs +9 -8
  6. package/crates/team-agent/src/cli/send/presentation.rs +1 -0
  7. package/crates/team-agent/src/cli/spec.rs +3 -0
  8. package/crates/team-agent/src/cli/tests/mod.rs +31 -0
  9. package/crates/team-agent/src/cli/tests/status_send.rs +13 -3
  10. package/crates/team-agent/src/cli/types.rs +7 -0
  11. package/crates/team-agent/src/coordinator/health.rs +17 -0
  12. package/crates/team-agent/src/coordinator/steps/abnormal.rs +55 -2
  13. package/crates/team-agent/src/db/migration.rs +2 -1
  14. package/crates/team-agent/src/db/schema.rs +16 -8
  15. package/crates/team-agent/src/leader/lease.rs +1 -2
  16. package/crates/team-agent/src/leader/start.rs +8 -8
  17. package/crates/team-agent/src/lifecycle/restart/agent.rs +51 -9
  18. package/crates/team-agent/src/lifecycle/restart/common.rs +3 -3
  19. package/crates/team-agent/src/lifecycle/restart/remove.rs +7 -139
  20. package/crates/team-agent/src/lifecycle/tests/agent_ops.rs +148 -5
  21. package/crates/team-agent/src/lifecycle/tests/lane_ops.rs +45 -13
  22. package/crates/team-agent/src/lifecycle/tests/startup_latency_contract.rs +10 -6
  23. package/crates/team-agent/src/lifecycle/tests.rs +11 -5
  24. package/crates/team-agent/src/mcp_server/helpers.rs +1 -0
  25. package/crates/team-agent/src/messaging/delivery.rs +29 -0
  26. package/crates/team-agent/src/messaging/helpers.rs +2 -0
  27. package/crates/team-agent/src/messaging/leader_receiver.rs +12 -0
  28. package/crates/team-agent/src/messaging/mod.rs +2 -0
  29. package/crates/team-agent/src/messaging/results.rs +5 -0
  30. package/crates/team-agent/src/messaging/send.rs +9 -0
  31. package/crates/team-agent/src/messaging/tests/runtime.rs +32 -0
  32. package/crates/team-agent/src/messaging/types.rs +2 -0
  33. package/crates/team-agent/src/messaging/wait.rs +366 -0
  34. package/crates/team-agent/src/messaging/watchers.rs +153 -6
  35. package/crates/team-agent/src/tmux_backend.rs +16 -12
  36. package/crates/team-agent/src/transport.rs +11 -11
  37. package/package.json +4 -4
@@ -0,0 +1,366 @@
1
+ use std::path::Path;
2
+
3
+ use serde_json::Value;
4
+
5
+ use super::MessagingError;
6
+
7
+ #[derive(Debug, Clone, PartialEq, Eq)]
8
+ pub struct WaitResult {
9
+ pub task_id: String,
10
+ pub result_id: String,
11
+ pub waited: bool,
12
+ }
13
+
14
+ impl WaitResult {
15
+ pub fn to_json(&self) -> Value {
16
+ serde_json::json!({
17
+ "ok": true,
18
+ "status": "completed",
19
+ "task_id": self.task_id,
20
+ "result_id": self.result_id,
21
+ "waited": self.waited,
22
+ })
23
+ }
24
+ }
25
+
26
+ #[cfg(not(unix))]
27
+ pub fn wait_for_result(_workspace: &Path, _task_id: &str) -> Result<WaitResult, MessagingError> {
28
+ Err(MessagingError::Validation(
29
+ "wait --task requires POSIX FIFO support".to_string(),
30
+ ))
31
+ }
32
+
33
+ #[cfg(unix)]
34
+ mod unix {
35
+ use std::ffi::CString;
36
+ use std::fs::{File, OpenOptions};
37
+ use std::io::{self, Read};
38
+ use std::os::fd::{AsRawFd, RawFd};
39
+ use std::os::unix::fs::OpenOptionsExt;
40
+ use std::path::{Path, PathBuf};
41
+ use std::sync::atomic::{AtomicI32, Ordering};
42
+
43
+ use rusqlite::{params, OptionalExtension};
44
+
45
+ use crate::event_log::EventLog;
46
+ use crate::message_store::MessageStore;
47
+
48
+ use super::{MessagingError, WaitResult};
49
+
50
+ static CAUGHT_SIGNAL: AtomicI32 = AtomicI32::new(0);
51
+
52
+ enum Registration {
53
+ Completed(String),
54
+ Pending,
55
+ }
56
+
57
+ struct FifoEndpoints {
58
+ path: PathBuf,
59
+ reader: File,
60
+ _keepalive_writer: File,
61
+ }
62
+
63
+ impl Drop for FifoEndpoints {
64
+ fn drop(&mut self) {
65
+ let _ = std::fs::remove_file(&self.path);
66
+ }
67
+ }
68
+
69
+ struct RegistrationCleanup {
70
+ db_path: PathBuf,
71
+ watcher_id: String,
72
+ active: bool,
73
+ }
74
+
75
+ impl RegistrationCleanup {
76
+ fn cleanup(&mut self) -> Result<(), MessagingError> {
77
+ if self.active {
78
+ let conn = crate::db::schema::open_db(&self.db_path)?;
79
+ conn.execute(
80
+ "delete from result_watchers where watcher_id = ?1",
81
+ params![self.watcher_id],
82
+ )?;
83
+ self.active = false;
84
+ }
85
+ Ok(())
86
+ }
87
+ }
88
+
89
+ impl Drop for RegistrationCleanup {
90
+ fn drop(&mut self) {
91
+ if self.active {
92
+ if let Ok(conn) = crate::db::schema::open_db(&self.db_path) {
93
+ let _ = conn.execute(
94
+ "delete from result_watchers where watcher_id = ?1",
95
+ params![self.watcher_id],
96
+ );
97
+ }
98
+ }
99
+ }
100
+ }
101
+
102
+ struct SignalGuard {
103
+ old_int: libc::sigaction,
104
+ old_term: libc::sigaction,
105
+ }
106
+
107
+ impl SignalGuard {
108
+ fn install() -> io::Result<Self> {
109
+ CAUGHT_SIGNAL.store(0, Ordering::SeqCst);
110
+ let mut action: libc::sigaction = unsafe { std::mem::zeroed() };
111
+ action.sa_sigaction = record_signal as *const () as libc::sighandler_t;
112
+ action.sa_flags = 0;
113
+ unsafe {
114
+ libc::sigemptyset(&mut action.sa_mask);
115
+ }
116
+ let mut old_int: libc::sigaction = unsafe { std::mem::zeroed() };
117
+ let mut old_term: libc::sigaction = unsafe { std::mem::zeroed() };
118
+ if unsafe { libc::sigaction(libc::SIGINT, &action, &mut old_int) } != 0 {
119
+ return Err(io::Error::last_os_error());
120
+ }
121
+ if unsafe { libc::sigaction(libc::SIGTERM, &action, &mut old_term) } != 0 {
122
+ let error = io::Error::last_os_error();
123
+ unsafe {
124
+ libc::sigaction(libc::SIGINT, &old_int, std::ptr::null_mut());
125
+ }
126
+ return Err(error);
127
+ }
128
+ Ok(Self { old_int, old_term })
129
+ }
130
+ }
131
+
132
+ impl Drop for SignalGuard {
133
+ fn drop(&mut self) {
134
+ unsafe {
135
+ libc::sigaction(libc::SIGINT, &self.old_int, std::ptr::null_mut());
136
+ libc::sigaction(libc::SIGTERM, &self.old_term, std::ptr::null_mut());
137
+ }
138
+ }
139
+ }
140
+
141
+ extern "C" fn record_signal(signal: libc::c_int) {
142
+ CAUGHT_SIGNAL.store(signal, Ordering::SeqCst);
143
+ }
144
+
145
+ pub fn wait_for_result(workspace: &Path, task_id: &str) -> Result<WaitResult, MessagingError> {
146
+ if task_id.trim().is_empty() {
147
+ return Err(MessagingError::Validation(
148
+ "wait --task requires a non-empty task id".to_string(),
149
+ ));
150
+ }
151
+ let _signals = SignalGuard::install()?;
152
+ let store = MessageStore::open(workspace)?;
153
+ let workspace = std::fs::canonicalize(workspace)?;
154
+ let watcher_id = format!("wait-{}", random_suffix()?);
155
+ let fifo_path = fifo_path(&workspace, task_id, &watcher_id)?;
156
+ let fifo = create_fifo(&fifo_path)?;
157
+ let conn = crate::db::schema::open_db(store.db_path())?;
158
+ let registration = register(&conn, &watcher_id, task_id, &fifo_path)?;
159
+ if let Registration::Completed(result_id) = registration {
160
+ return Ok(WaitResult {
161
+ task_id: task_id.to_string(),
162
+ result_id,
163
+ waited: false,
164
+ });
165
+ }
166
+
167
+ let mut cleanup = RegistrationCleanup {
168
+ db_path: store.db_path().to_path_buf(),
169
+ watcher_id: watcher_id.clone(),
170
+ active: true,
171
+ };
172
+ EventLog::new(&workspace).write(
173
+ "result_wake.registered",
174
+ serde_json::json!({
175
+ "task_id": task_id,
176
+ "watcher_id": watcher_id,
177
+ "recipient": fifo_path,
178
+ }),
179
+ )?;
180
+ let result_id = read_wake_line(fifo.reader.as_raw_fd())?;
181
+ cleanup.cleanup()?;
182
+ Ok(WaitResult {
183
+ task_id: task_id.to_string(),
184
+ result_id,
185
+ waited: true,
186
+ })
187
+ }
188
+
189
+ fn register(
190
+ conn: &rusqlite::Connection,
191
+ watcher_id: &str,
192
+ task_id: &str,
193
+ fifo_path: &Path,
194
+ ) -> Result<Registration, MessagingError> {
195
+ conn.execute_batch("BEGIN IMMEDIATE")?;
196
+ let operation = (|| -> Result<Registration, MessagingError> {
197
+ let completed = conn
198
+ .query_row(
199
+ "select result_id from results
200
+ where task_id = ?1 order by created_at, result_id limit 1",
201
+ params![task_id],
202
+ |row| row.get::<_, String>(0),
203
+ )
204
+ .optional()?;
205
+ if let Some(result_id) = completed {
206
+ return Ok(Registration::Completed(result_id));
207
+ }
208
+ conn.execute(
209
+ "insert into result_watchers(
210
+ watcher_id, owner_team_id, task_id, agent_id, message_id,
211
+ leader_id, recipient, status, created_at
212
+ ) values (?1, null, ?2, null, null, 'team-agent', ?3, 'pending', ?4)",
213
+ params![
214
+ watcher_id,
215
+ task_id,
216
+ fifo_path.to_string_lossy().as_ref(),
217
+ chrono::Utc::now().to_rfc3339(),
218
+ ],
219
+ )?;
220
+ Ok(Registration::Pending)
221
+ })();
222
+ match operation {
223
+ Ok(registration) => match conn.execute_batch("COMMIT") {
224
+ Ok(()) => Ok(registration),
225
+ Err(error) => {
226
+ let _ = conn.execute_batch("ROLLBACK");
227
+ Err(error.into())
228
+ }
229
+ },
230
+ Err(error) => {
231
+ let _ = conn.execute_batch("ROLLBACK");
232
+ Err(error)
233
+ }
234
+ }
235
+ }
236
+
237
+ fn fifo_path(
238
+ workspace: &Path,
239
+ task_id: &str,
240
+ watcher_id: &str,
241
+ ) -> Result<PathBuf, MessagingError> {
242
+ let dir = workspace.join(".team/runtime/wake");
243
+ std::fs::create_dir_all(&dir)?;
244
+ let task = task_id
245
+ .chars()
246
+ .map(|ch| {
247
+ if ch.is_ascii_alphanumeric() || matches!(ch, '-' | '_') {
248
+ ch
249
+ } else {
250
+ '_'
251
+ }
252
+ })
253
+ .take(80)
254
+ .collect::<String>();
255
+ let task = if task.is_empty() { "task" } else { &task };
256
+ Ok(dir.join(format!("{task}-{watcher_id}.fifo")))
257
+ }
258
+
259
+ fn random_suffix() -> Result<String, MessagingError> {
260
+ let mut bytes = [0_u8; 16];
261
+ File::open("/dev/urandom")?.read_exact(&mut bytes)?;
262
+ Ok(bytes.iter().map(|byte| format!("{byte:02x}")).collect())
263
+ }
264
+
265
+ fn create_fifo(path: &Path) -> Result<FifoEndpoints, MessagingError> {
266
+ let c_path = CString::new(path.as_os_str().as_encoded_bytes())
267
+ .map_err(|_| MessagingError::Validation("FIFO path contains a NUL byte".to_string()))?;
268
+ if unsafe { libc::mkfifo(c_path.as_ptr(), 0o600) } != 0 {
269
+ return Err(io::Error::last_os_error().into());
270
+ }
271
+ let reader = match OpenOptions::new()
272
+ .read(true)
273
+ .custom_flags(libc::O_NONBLOCK)
274
+ .open(path)
275
+ {
276
+ Ok(reader) => reader,
277
+ Err(error) => {
278
+ let _ = std::fs::remove_file(path);
279
+ return Err(error.into());
280
+ }
281
+ };
282
+ let keepalive_writer = match OpenOptions::new()
283
+ .write(true)
284
+ .custom_flags(libc::O_NONBLOCK)
285
+ .open(path)
286
+ {
287
+ Ok(writer) => writer,
288
+ Err(error) => {
289
+ let _ = std::fs::remove_file(path);
290
+ return Err(error.into());
291
+ }
292
+ };
293
+ set_blocking(reader.as_raw_fd())?;
294
+ Ok(FifoEndpoints {
295
+ path: path.to_path_buf(),
296
+ reader,
297
+ _keepalive_writer: keepalive_writer,
298
+ })
299
+ }
300
+
301
+ fn set_blocking(fd: RawFd) -> io::Result<()> {
302
+ let flags = unsafe { libc::fcntl(fd, libc::F_GETFL) };
303
+ if flags < 0 {
304
+ return Err(io::Error::last_os_error());
305
+ }
306
+ if unsafe { libc::fcntl(fd, libc::F_SETFL, flags & !libc::O_NONBLOCK) } < 0 {
307
+ return Err(io::Error::last_os_error());
308
+ }
309
+ Ok(())
310
+ }
311
+
312
+ fn read_wake_line(fd: RawFd) -> Result<String, MessagingError> {
313
+ let mut line = Vec::new();
314
+ loop {
315
+ if let Some(signal) = caught_signal() {
316
+ return Err(interrupted(signal).into());
317
+ }
318
+ let mut bytes = [0_u8; 256];
319
+ let count =
320
+ unsafe { libc::read(fd, bytes.as_mut_ptr().cast::<libc::c_void>(), bytes.len()) };
321
+ if count > 0 {
322
+ line.extend_from_slice(&bytes[..count as usize]);
323
+ if let Some(newline) = line.iter().position(|byte| *byte == b'\n') {
324
+ line.truncate(newline);
325
+ let result_id = String::from_utf8(line).map_err(|error| {
326
+ MessagingError::Validation(format!("invalid FIFO wake line: {error}"))
327
+ })?;
328
+ let result_id = result_id.trim().to_string();
329
+ if result_id.is_empty() {
330
+ return Err(MessagingError::Validation(
331
+ "FIFO wake line did not contain a result id".to_string(),
332
+ ));
333
+ }
334
+ return Ok(result_id);
335
+ }
336
+ continue;
337
+ }
338
+ if count == 0 {
339
+ continue;
340
+ }
341
+ let error = io::Error::last_os_error();
342
+ if error.kind() == io::ErrorKind::Interrupted {
343
+ if let Some(signal) = caught_signal() {
344
+ return Err(interrupted(signal).into());
345
+ }
346
+ continue;
347
+ }
348
+ return Err(error.into());
349
+ }
350
+ }
351
+
352
+ fn caught_signal() -> Option<i32> {
353
+ let signal = CAUGHT_SIGNAL.load(Ordering::SeqCst);
354
+ (signal != 0).then_some(signal)
355
+ }
356
+
357
+ fn interrupted(signal: i32) -> io::Error {
358
+ io::Error::new(
359
+ io::ErrorKind::Interrupted,
360
+ format!("wait interrupted by signal {signal}"),
361
+ )
362
+ }
363
+ }
364
+
365
+ #[cfg(unix)]
366
+ pub use unix::wait_for_result;
@@ -2,6 +2,13 @@
2
2
 
3
3
  use std::path::Path;
4
4
 
5
+ #[cfg(unix)]
6
+ use std::fs::OpenOptions;
7
+ #[cfg(unix)]
8
+ use std::io::Write as _;
9
+ #[cfg(unix)]
10
+ use std::os::unix::fs::OpenOptionsExt;
11
+
5
12
  use rusqlite::{params, OptionalExtension};
6
13
 
7
14
  use crate::event_log::EventLog;
@@ -119,6 +126,128 @@ fn watcher_matches(
119
126
  task_matches && agent_matches
120
127
  }
121
128
 
129
+ pub(crate) fn notify_fifo_result_watchers(
130
+ conn: &rusqlite::Connection,
131
+ event_log: &EventLog,
132
+ task_id: &str,
133
+ result_id: &str,
134
+ ) -> Result<(), MessagingError> {
135
+ let mut stmt = conn.prepare(
136
+ "select watcher_id, recipient from result_watchers
137
+ where task_id = ?1 and status = 'pending' and recipient is not null
138
+ order by created_at, watcher_id",
139
+ )?;
140
+ let rows = stmt
141
+ .query_map(params![task_id], |row| {
142
+ Ok((row.get::<_, String>(0)?, row.get::<_, String>(1)?))
143
+ })?
144
+ .collect::<Result<Vec<_>, _>>()?;
145
+ drop(stmt);
146
+ for (watcher_id, recipient) in rows {
147
+ let _ = notify_fifo_waiter(conn, event_log, &watcher_id, task_id, result_id, &recipient)?;
148
+ }
149
+ Ok(())
150
+ }
151
+
152
+ #[cfg(unix)]
153
+ fn notify_fifo_waiter(
154
+ conn: &rusqlite::Connection,
155
+ event_log: &EventLog,
156
+ watcher_id: &str,
157
+ task_id: &str,
158
+ result_id: &str,
159
+ recipient: &str,
160
+ ) -> Result<WatcherNotice, MessagingError> {
161
+ let write = (|| -> std::io::Result<()> {
162
+ let mut fifo = OpenOptions::new()
163
+ .write(true)
164
+ .custom_flags(libc::O_NONBLOCK)
165
+ .open(recipient)?;
166
+ fifo.write_all(format!("{result_id}\n").as_bytes())
167
+ })();
168
+ match write {
169
+ Ok(()) => {
170
+ event_log.write(
171
+ "result_wake.notified",
172
+ serde_json::json!({
173
+ "task_id": task_id,
174
+ "result_id": result_id,
175
+ "watcher_id": watcher_id,
176
+ "recipient": recipient,
177
+ }),
178
+ )?;
179
+ Ok(WatcherNotice {
180
+ watcher_id: watcher_id.to_string(),
181
+ result_id: Some(result_id.to_string()),
182
+ ok: true,
183
+ status: Some("notified".to_string()),
184
+ notified_message_id: None,
185
+ primary_watcher_id: None,
186
+ prior_state: None,
187
+ error: None,
188
+ })
189
+ }
190
+ Err(error)
191
+ if matches!(
192
+ error.raw_os_error(),
193
+ Some(code) if code == libc::ENXIO || code == libc::ENOENT
194
+ ) =>
195
+ {
196
+ let reason = if error.raw_os_error() == Some(libc::ENXIO) {
197
+ "ENXIO"
198
+ } else {
199
+ "ENOENT"
200
+ };
201
+ conn.execute(
202
+ "delete from result_watchers where watcher_id = ?1",
203
+ params![watcher_id],
204
+ )?;
205
+ event_log.write(
206
+ "result_wake.notify_failed",
207
+ serde_json::json!({
208
+ "task_id": task_id,
209
+ "result_id": result_id,
210
+ "watcher_id": watcher_id,
211
+ "recipient": recipient,
212
+ "reason": reason,
213
+ }),
214
+ )?;
215
+ Ok(WatcherNotice {
216
+ watcher_id: watcher_id.to_string(),
217
+ result_id: Some(result_id.to_string()),
218
+ ok: false,
219
+ status: Some("removed".to_string()),
220
+ notified_message_id: None,
221
+ primary_watcher_id: None,
222
+ prior_state: None,
223
+ error: Some(reason.to_string()),
224
+ })
225
+ }
226
+ Err(error) => Err(error.into()),
227
+ }
228
+ }
229
+
230
+ #[cfg(not(unix))]
231
+ fn notify_fifo_waiter(
232
+ _conn: &rusqlite::Connection,
233
+ _event_log: &EventLog,
234
+ watcher_id: &str,
235
+ _task_id: &str,
236
+ result_id: &str,
237
+ _recipient: &str,
238
+ ) -> Result<WatcherNotice, MessagingError> {
239
+ Ok(WatcherNotice {
240
+ watcher_id: watcher_id.to_string(),
241
+ result_id: Some(result_id.to_string()),
242
+ ok: false,
243
+ status: Some("unsupported".to_string()),
244
+ notified_message_id: None,
245
+ primary_watcher_id: None,
246
+ prior_state: None,
247
+ error: Some("POSIX FIFO support unavailable".to_string()),
248
+ })
249
+ }
250
+
122
251
  fn deliver_primary_watcher(
123
252
  workspace: &Path,
124
253
  conn: &rusqlite::Connection,
@@ -143,6 +272,20 @@ fn deliver_primary_watcher(
143
272
  "missing_result_id",
144
273
  );
145
274
  };
275
+ if let Some(recipient) = watcher
276
+ .get("recipient")
277
+ .and_then(|value| value.as_str())
278
+ .filter(|recipient| Path::new(recipient).is_absolute())
279
+ {
280
+ return notify_fifo_waiter(
281
+ conn,
282
+ event_log,
283
+ watcher_id,
284
+ result_task.unwrap_or_default(),
285
+ result_id,
286
+ recipient,
287
+ );
288
+ }
146
289
  if let Some(existing) = delivered_result_message(
147
290
  store,
148
291
  result_id,
@@ -196,7 +339,10 @@ fn deliver_primary_watcher(
196
339
  .get("leader_id")
197
340
  .and_then(|v| v.as_str())
198
341
  .unwrap_or("team-agent"),
199
- "leader",
342
+ watcher
343
+ .get("recipient")
344
+ .and_then(|value| value.as_str())
345
+ .unwrap_or("leader"),
200
346
  &content,
201
347
  None,
202
348
  false,
@@ -332,7 +478,7 @@ pub fn retry_result_deliveries(
332
478
  // delivery path with dedupe/attempt bounds); a watcher is never flipped to
333
479
  // `notified` without a delivery. Missing result rows are skipped (still retryable).
334
480
  let mut stmt = conn.prepare(
335
- "select watcher_id, owner_team_id, task_id, agent_id, leader_id, status, created_at,
481
+ "select watcher_id, owner_team_id, task_id, agent_id, leader_id, recipient, status, created_at,
336
482
  result_id, notified_message_id
337
483
  from result_watchers
338
484
  where status in ('pending', 'notify_failed')
@@ -346,10 +492,11 @@ pub fn retry_result_deliveries(
346
492
  "task_id": row.get::<_, Option<String>>(2)?,
347
493
  "agent_id": row.get::<_, Option<String>>(3)?,
348
494
  "leader_id": row.get::<_, Option<String>>(4)?,
349
- "status": row.get::<_, Option<String>>(5)?,
350
- "created_at": row.get::<_, Option<String>>(6)?,
351
- "result_id": row.get::<_, Option<String>>(7)?,
352
- "notified_message_id": row.get::<_, Option<String>>(8)?,
495
+ "recipient": row.get::<_, Option<String>>(5)?,
496
+ "status": row.get::<_, Option<String>>(6)?,
497
+ "created_at": row.get::<_, Option<String>>(7)?,
498
+ "result_id": row.get::<_, Option<String>>(8)?,
499
+ "notified_message_id": row.get::<_, Option<String>>(9)?,
353
500
  }))
354
501
  })?
355
502
  .collect::<Result<Vec<_>, _>>()?;
@@ -1723,8 +1723,8 @@ pub fn worker_provider_exit_marker(provider_label: &str) -> String {
1723
1723
  /// as a CHILD of a long-lived shell so provider exit does NOT collapse the
1724
1724
  /// worker pane (which under upstream tmux 3.6a private-server bugs can
1725
1725
  /// cascade into whole-server death). When the provider exits, the worker
1726
- /// pane returns to an interactive shell with an explicit worker exit
1727
- /// marker, matching manual `tmux new-window` then `<provider>` behaviour.
1726
+ /// pane remains alive with an explicit worker exit marker, then runs an
1727
+ /// inert `sh` tail that does not read terminal input.
1728
1728
  pub fn worker_shell_wrapper_command(
1729
1729
  argv: &[String],
1730
1730
  cwd: &Path,
@@ -1758,23 +1758,21 @@ pub fn worker_shell_wrapper_command(
1758
1758
  worker_provider_exit_marker(provider_label)
1759
1759
  )));
1760
1760
  parts.push("\"$rc\";".to_string());
1761
- parts.push("exec".to_string());
1762
- parts.push("\"${SHELL:-/bin/zsh}\"".to_string());
1763
- parts.push("-l".to_string());
1761
+ parts.push(inert_pane_tail_command());
1764
1762
  parts.join(" ")
1765
1763
  }
1766
1764
 
1767
1765
  /// 0.4.x (CR C-2): leader shell wrapper — provider runs as a CHILD of a
1768
1766
  /// long-lived shell, not as the pane's primary process. When the provider
1769
- /// exits, the pane returns to an interactive shell with an explicit exit
1770
- /// marker, matching manual `tmux new-session` then `claude` behaviour.
1767
+ /// exits, the pane remains alive with an explicit exit marker, then runs an
1768
+ /// inert `sh` tail that does not read terminal input.
1771
1769
  ///
1772
1770
  /// Four required envelope sections (CR C-2):
1773
1771
  /// 1. cd <cwd> — same as `shell_command`
1774
1772
  /// 2. unset <KEY> ... — provider env_unset block
1775
1773
  /// 3. KEY=val ... <provider> — env exports + provider invocation
1776
1774
  /// (NO `exec` — runs as child)
1777
- /// 4. printf exit marker; exec shell -l
1775
+ /// 4. printf exit marker; exec inert sh tail
1778
1776
  ///
1779
1777
  /// `provider_label` is a human-readable provider name (e.g. "claude",
1780
1778
  /// "codex") embedded in the exit marker for diagnostics.
@@ -1809,7 +1807,7 @@ pub fn leader_shell_wrapper_command(
1809
1807
  }
1810
1808
  parts.extend(argv.iter().map(|arg| shell_quote(arg)));
1811
1809
  parts.push(";".to_string());
1812
- // 4. exit marker + fall back to interactive shell
1810
+ // 4. exit marker + inert shell tail
1813
1811
  parts.push("rc=$?;".to_string());
1814
1812
  parts.push("printf".to_string());
1815
1813
  // CR R6: marker text comes from single-source `leader_provider_exit_marker`.
@@ -1818,12 +1816,18 @@ pub fn leader_shell_wrapper_command(
1818
1816
  leader_provider_exit_marker(provider_label)
1819
1817
  )));
1820
1818
  parts.push("\"$rc\";".to_string());
1821
- parts.push("exec".to_string());
1822
- parts.push("\"${SHELL:-/bin/zsh}\"".to_string());
1823
- parts.push("-l".to_string());
1819
+ parts.push(inert_pane_tail_command());
1824
1820
  parts.join(" ")
1825
1821
  }
1826
1822
 
1823
+ fn inert_pane_tail_command() -> String {
1824
+ // `sh` is deliberate: provider-exit health checks use the shell basename
1825
+ // to reach the exit-marker branch. `sh -c` does not read commands from
1826
+ // stdin; disabling echo also prevents input from appearing accepted.
1827
+ let script = r#"trap '' INT QUIT; stty -echo 2>/dev/null; printf "%s\n" "[team-agent] Provider exited; this pane no longer accepts input. Restart from another pane with the appropriate team-agent start command."; while :; do sleep 3600 & wait "$!"; done"#;
1828
+ format!("exec /bin/sh -c {}", shell_quote(script))
1829
+ }
1830
+
1827
1831
  fn shell_quote(raw: &str) -> String {
1828
1832
  if raw.is_empty() {
1829
1833
  return "''".to_string();
@@ -613,11 +613,11 @@ pub trait Transport: Send + Sync {
613
613
  /// variant that reuses the leader-wrapper mechanism so worker provider
614
614
  /// exit does not collapse the pane into `[exited]` (which under
615
615
  /// upstream tmux 3.6a private-server bugs can cascade into server
616
- /// death). When the provider exits, the worker pane returns to an
617
- /// interactive shell with an explicit worker exit marker matching
618
- /// manual `tmux new-window` then `<provider>` behaviour. Default falls
619
- /// back to plain `spawn_first_with_env_unset` for backends that have
620
- /// no shell layer (test-only `OfflineTransport`).
616
+ /// death). When the provider exits, the worker pane keeps an explicit
617
+ /// worker exit marker and runs the inert `/bin/sh -c` tail: it ignores
618
+ /// INT/QUIT and does not read pane stdin. Default falls back to plain
619
+ /// `spawn_first_with_env_unset` for backends that have no shell layer
620
+ /// (test-only `OfflineTransport`).
621
621
  fn spawn_first_with_worker_shell_wrapper(
622
622
  &self,
623
623
  session: &SessionName,
@@ -650,12 +650,12 @@ pub trait Transport: Send + Sync {
650
650
  /// 0.4.x (CR C-2): leader-specific spawn variant. Instead of `exec <cmd>`
651
651
  /// (which makes the provider the pane's primary process and turns the
652
652
  /// pane into `[exited]` when the provider exits), build a shell line
653
- /// that runs the provider as a CHILD of a long-lived shell:
654
- /// `cd ... && unset ... && KEY=val ... <cmd>; rc=$?; printf '\n[team-agent] <provider> exited with %s\n' "$rc"; exec "${SHELL:-/bin/zsh}" -l`.
655
- /// When the provider exits, the pane returns to an interactive shell with
656
- /// an explicit exit marker matching manual `tmux new-session` then
657
- /// `claude` behaviour. Default falls back to plain `spawn_first_with_env_unset`
658
- /// for backends that have no shell layer (test-only `OfflineTransport`).
653
+ /// that runs the provider as a CHILD of a long-lived shell. When the
654
+ /// provider exits, the wrapper records the exit code and marker, then
655
+ /// execs an inert `/bin/sh -c` tail that ignores INT/QUIT and does not
656
+ /// read pane stdin. Default falls back to plain
657
+ /// `spawn_first_with_env_unset` for backends that have no shell layer
658
+ /// (test-only `OfflineTransport`).
659
659
  fn spawn_first_with_leader_shell_wrapper(
660
660
  &self,
661
661
  session: &SessionName,