| 1 | //! Mailbox abstraction for sub-agent runtime coordination. |
| 2 | //! |
| 3 | //! Monotonic sequence numbers give every consumer a consistent ordering even |
| 4 | //! when multiple subscribers (e.g. UI card + parent agent) drain |
| 5 | //! independently; close-as-cancel lets a single signal both stop new mail and |
| 6 | //! propagate cancellation through nested children. |
| 7 | |
| 8 | use std::collections::VecDeque; |
| 9 | use std::sync::Arc; |
| 10 | use std::sync::atomic::{AtomicBool, AtomicU64, Ordering}; |
| 11 | #[cfg(test)] |
| 12 | use std::time::Duration; |
| 13 | |
| 14 | use serde::{Deserialize, Serialize}; |
| 15 | use tokio::sync::{mpsc, watch}; |
| 16 | use tokio_util::sync::CancellationToken; |
| 17 | |
| 18 | #[cfg(test)] |
| 19 | use crate::config::ApiProvider; |
| 20 | use crate::tools::todo::TodoListSnapshot; |
| 21 | use codewhale_models::Usage; |
| 22 | |
| 23 | use super::FleetRole; |
| 24 | |
| 25 | /// Stable, structured progress envelope shared across the sub-agent surface. |
| 26 | /// |
| 27 | /// Tracks the lifecycle of a single agent (identified by `agent_id`) end to |
| 28 | /// end: spawn, per-step progress, tool execution, completion / failure / |
| 29 | /// cancellation, and parent → child topology so consumers can render trees. |
| 30 | #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)] |
| 31 | #[serde(tag = "kind", rename_all = "snake_case")] |
| 32 | pub enum MailboxMessage { |
| 33 | /// Agent has been started (background task is running). |
| 34 | Started { |
| 35 | agent_id: String, |
| 36 | agent_type: String, |
| 37 | }, |
| 38 | /// Free-form human-readable progress (mirrors `Event::AgentProgress`). |
| 39 | Progress { agent_id: String, status: String }, |
| 40 | /// A tool call inside the agent has started. |
| 41 | ToolCallStarted { |
| 42 | agent_id: String, |
| 43 | tool_name: String, |
| 44 | step: u32, |
| 45 | }, |
| 46 | /// A tool call inside the agent has finished. |
| 47 | ToolCallCompleted { |
| 48 | agent_id: String, |
| 49 | tool_name: String, |
| 50 | step: u32, |
| 51 | ok: bool, |
| 52 | }, |
| 53 | /// A child agent was spawned by this agent. |
| 54 | ChildSpawned { parent_id: String, child_id: String }, |
| 55 | /// Agent completed successfully (carries the summary line shown in the |
| 56 | /// transcript; full result is still available through the transcript handle). |
| 57 | Completed { agent_id: String, summary: String }, |
| 58 | /// Agent failed with the carried error message. |
| 59 | Failed { agent_id: String, error: String }, |
| 60 | /// Agent was interrupted (e.g. API timeout) with a continuable |
| 61 | /// checkpoint; the worker is parked waiting for continuation input. |
| 62 | Interrupted { agent_id: String, reason: String }, |
| 63 | /// Cancellation propagated to this agent. |
| 64 | Cancelled { agent_id: String }, |
| 65 | /// This agent's **own** bounded To-do snapshot (#4810). |
| 66 | /// |
| 67 | /// Published by the agent that owns the ledger, from its private list, so a |
| 68 | /// consumer keyed on `agent_id` can never attribute a parent's or a |
| 69 | /// sibling's work to this agent. Emitted only when the snapshot actually |
| 70 | /// changes; the payload is the canonical [`TodoListSnapshot`], not a second |
| 71 | /// ledger. |
| 72 | WorkState { |
| 73 | agent_id: String, |
| 74 | /// Absent in older persisted payloads, which predate per-agent Work |
| 75 | /// state; those deserialize to an empty (no work stated) snapshot. |
| 76 | #[serde(default)] |
| 77 | todo: TodoListSnapshot, |
| 78 | }, |
| 79 | /// Incremental token usage from a sub-agent's API call. |
| 80 | /// Published after each turn so the parent's cost counter updates live. |
| 81 | TokenUsage { |
| 82 | agent_id: String, |
| 83 | /// Stable identity of the provider response. Runtime accounting uses |
| 84 | /// this across direct durability, mailbox replay, and restart dedupe. |
| 85 | source_id: String, |
| 86 | /// Immutable provider/model/billing evidence captured before the |
| 87 | /// child request was sent. Boxed: the envelope dwarfs every other |
| 88 | /// variant, and mailboxes queue many messages. |
| 89 | route: Box<crate::cost_status::EffectiveRouteEnvelope>, |
| 90 | /// Provider usage payload, including cache-hit/cache-miss fields. |
| 91 | usage: Usage, |
| 92 | }, |
| 93 | } |
| 94 | |
| 95 | impl MailboxMessage { |
| 96 | /// `agent_id` of the message subject (for `ChildSpawned` this is the |
| 97 | /// child, since that's the new lifecycle being announced). |
| 98 | #[must_use] |
| 99 | pub fn agent_id(&self) -> &str { |
| 100 | match self { |
| 101 | Self::Started { agent_id, .. } |
| 102 | | Self::Progress { agent_id, .. } |
| 103 | | Self::ToolCallStarted { agent_id, .. } |
| 104 | | Self::ToolCallCompleted { agent_id, .. } |
| 105 | | Self::Completed { agent_id, .. } |
| 106 | | Self::Failed { agent_id, .. } |
| 107 | | Self::Interrupted { agent_id, .. } |
| 108 | | Self::Cancelled { agent_id } |
| 109 | | Self::WorkState { agent_id, .. } |
| 110 | | Self::TokenUsage { agent_id, .. } => agent_id, |
| 111 | Self::ChildSpawned { child_id, .. } => child_id, |
| 112 | } |
| 113 | } |
| 114 | |
| 115 | pub(crate) fn started(agent_id: impl Into<String>, agent_type: FleetRole) -> Self { |
| 116 | Self::Started { |
| 117 | agent_id: agent_id.into(), |
| 118 | agent_type: agent_type.as_str().to_string(), |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | pub(crate) fn progress(agent_id: impl Into<String>, status: impl Into<String>) -> Self { |
| 123 | Self::Progress { |
| 124 | agent_id: agent_id.into(), |
| 125 | status: status.into(), |
| 126 | } |
| 127 | } |
| 128 | |
| 129 | pub(crate) fn work_state(agent_id: impl Into<String>, todo: TodoListSnapshot) -> Self { |
| 130 | Self::WorkState { |
| 131 | agent_id: agent_id.into(), |
| 132 | todo, |
| 133 | } |
| 134 | } |
| 135 | |
| 136 | pub(crate) fn token_usage( |
| 137 | agent_id: impl Into<String>, |
| 138 | source_id: impl Into<String>, |
| 139 | route: crate::cost_status::EffectiveRouteEnvelope, |
| 140 | usage: Usage, |
| 141 | ) -> Self { |
| 142 | Self::TokenUsage { |
| 143 | agent_id: agent_id.into(), |
| 144 | source_id: source_id.into(), |
| 145 | route: Box::new(route), |
| 146 | usage, |
| 147 | } |
| 148 | } |
| 149 | } |
| 150 | |
| 151 | /// One delivery: a sequence number plus the message. The sequence is |
| 152 | /// monotonic across the entire mailbox (not per-agent) so a single ordering |
| 153 | /// is well-defined even when multiple sub-agents share one mailbox. |
| 154 | #[derive(Debug, Clone, PartialEq, Eq)] |
| 155 | pub struct MailboxEnvelope { |
| 156 | pub seq: u64, |
| 157 | pub message: MailboxMessage, |
| 158 | } |
| 159 | |
| 160 | /// Sender side of the mailbox. |
| 161 | /// |
| 162 | /// Cheaply cloneable (everything inside is `Arc`/atomic). Cloning a |
| 163 | /// `Mailbox` shares the same delivery channel, sequence counter, watch |
| 164 | /// notifier, and close/cancel state — so a child runtime that clones its |
| 165 | /// parent's `Mailbox` participates in the same stream. |
| 166 | #[derive(Clone)] |
| 167 | pub struct Mailbox { |
| 168 | inner: Arc<MailboxInner>, |
| 169 | } |
| 170 | |
| 171 | struct MailboxInner { |
| 172 | tx: mpsc::UnboundedSender<MailboxEnvelope>, |
| 173 | next_seq: AtomicU64, |
| 174 | seq_tx: watch::Sender<u64>, |
| 175 | closed: AtomicBool, |
| 176 | /// Linearizes publication against turn-end sealing. Without this gate a |
| 177 | /// producer could observe `closed = false`, lose the race to the turn |
| 178 | /// completion barrier, and enqueue usage after `TurnComplete`. |
| 179 | send_gate: std::sync::Mutex<()>, |
| 180 | #[cfg(test)] |
| 181 | cancel_token: CancellationToken, |
| 182 | } |
| 183 | |
| 184 | /// Receiver side of the mailbox. Not `Clone` — only the original creator |
| 185 | /// can drain. Use `Mailbox::subscribe()` for fanout (UI cards + parent both |
| 186 | /// observing the same stream). |
| 187 | pub struct MailboxReceiver { |
| 188 | rx: mpsc::UnboundedReceiver<MailboxEnvelope>, |
| 189 | pending: VecDeque<MailboxEnvelope>, |
| 190 | } |
| 191 | |
| 192 | impl Mailbox { |
| 193 | /// Create a new mailbox bound to the given cancellation token. Closing |
| 194 | /// the mailbox (or dropping the last sender) cancels this token. Runtimes |
| 195 | /// that derive from the same token observe that cancellation; detached |
| 196 | /// background `agent` sessions use their own runtime token. |
| 197 | #[must_use] |
| 198 | pub fn new(cancel_token: CancellationToken) -> (Self, MailboxReceiver) { |
| 199 | #[cfg(not(test))] |
| 200 | let _ = cancel_token; |
| 201 | let (tx, rx) = mpsc::unbounded_channel(); |
| 202 | let (seq_tx, _) = watch::channel(0); |
| 203 | let inner = MailboxInner { |
| 204 | tx, |
| 205 | next_seq: AtomicU64::new(0), |
| 206 | seq_tx, |
| 207 | closed: AtomicBool::new(false), |
| 208 | send_gate: std::sync::Mutex::new(()), |
| 209 | #[cfg(test)] |
| 210 | cancel_token, |
| 211 | }; |
| 212 | ( |
| 213 | Self { |
| 214 | inner: Arc::new(inner), |
| 215 | }, |
| 216 | MailboxReceiver { |
| 217 | rx, |
| 218 | pending: VecDeque::new(), |
| 219 | }, |
| 220 | ) |
| 221 | } |
| 222 | |
| 223 | /// Subscribe to seq-bump notifications. Each `recv()` returns when the |
| 224 | /// sequence counter advances, signaling new mail without copying it — |
| 225 | /// the consumer then calls `drain` (or `recv_one` on its own receiver). |
| 226 | /// Multiple subscribers may exist; this is the fanout primitive. |
| 227 | #[cfg(test)] |
| 228 | #[must_use] |
| 229 | pub fn subscribe(&self) -> watch::Receiver<u64> { |
| 230 | self.inner.seq_tx.subscribe() |
| 231 | } |
| 232 | |
| 233 | /// Send a message; returns `Some(seq)` on success, `None` if the |
| 234 | /// mailbox is already closed (callers should treat this as "the |
| 235 | /// receiver is gone, stop publishing"). |
| 236 | pub fn send(&self, message: MailboxMessage) -> Option<u64> { |
| 237 | let _send_gate = self |
| 238 | .inner |
| 239 | .send_gate |
| 240 | .lock() |
| 241 | .unwrap_or_else(|error| error.into_inner()); |
| 242 | if self.inner.closed.load(Ordering::Acquire) { |
| 243 | return None; |
| 244 | } |
| 245 | let seq = self.inner.next_seq.fetch_add(1, Ordering::Relaxed) + 1; |
| 246 | let envelope = MailboxEnvelope { seq, message }; |
| 247 | if self.inner.tx.send(envelope).is_err() { |
| 248 | return None; |
| 249 | } |
| 250 | let _ = self.inner.seq_tx.send_replace(seq); |
| 251 | Some(seq) |
| 252 | } |
| 253 | |
| 254 | /// Stop publication for this turn without cancelling detached workers. |
| 255 | /// |
| 256 | /// The engine seals, drains, and awaits the mailbox before it emits |
| 257 | /// `TurnComplete`. The send gate makes this a hard ordering boundary: |
| 258 | /// once this returns, every accepted envelope is already in the receiver |
| 259 | /// and no later worker message can attach itself to the completed turn. |
| 260 | pub(crate) fn seal(&self) { |
| 261 | let _send_gate = self |
| 262 | .inner |
| 263 | .send_gate |
| 264 | .lock() |
| 265 | .unwrap_or_else(|error| error.into_inner()); |
| 266 | self.inner.closed.store(true, Ordering::Release); |
| 267 | } |
| 268 | |
| 269 | /// Whether the mailbox has been closed. |
| 270 | #[cfg(test)] |
| 271 | #[must_use] |
| 272 | pub fn is_closed(&self) -> bool { |
| 273 | self.inner.closed.load(Ordering::Acquire) |
| 274 | } |
| 275 | |
| 276 | /// Close the mailbox AND cancel the bound cancellation token. |
| 277 | /// |
| 278 | /// "Close-as-cancel": there's no useful state where the consumer is gone |
| 279 | /// but producers bound to this mailbox token should keep publishing. |
| 280 | /// Closing cancels the bound token; directly derived `child_runtime()` |
| 281 | /// children observe it, while detached `agent` sessions rely on their |
| 282 | /// own explicit cancellation. |
| 283 | #[cfg(test)] |
| 284 | pub fn close(&self) { |
| 285 | let was_closed = self.inner.closed.load(Ordering::Acquire); |
| 286 | self.seal(); |
| 287 | if !was_closed { |
| 288 | self.inner.cancel_token.cancel(); |
| 289 | } |
| 290 | } |
| 291 | } |
| 292 | |
| 293 | impl MailboxReceiver { |
| 294 | #[cfg(test)] |
| 295 | fn sync_pending(&mut self) { |
| 296 | while let Ok(env) = self.rx.try_recv() { |
| 297 | self.pending.push_back(env); |
| 298 | } |
| 299 | } |
| 300 | |
| 301 | /// Whether any envelopes are buffered (or arrived since last check). |
| 302 | #[cfg(test)] |
| 303 | pub fn has_pending(&mut self) -> bool { |
| 304 | self.sync_pending(); |
| 305 | !self.pending.is_empty() |
| 306 | } |
| 307 | |
| 308 | /// Drain all currently available envelopes, in delivery order. |
| 309 | #[cfg(test)] |
| 310 | pub fn drain(&mut self) -> Vec<MailboxEnvelope> { |
| 311 | self.sync_pending(); |
| 312 | self.pending.drain(..).collect() |
| 313 | } |
| 314 | |
| 315 | /// Await the next envelope, with backpressure-aware blocking. Returns |
| 316 | /// `None` when every sender has been dropped and the buffer is drained. |
| 317 | pub async fn recv(&mut self) -> Option<MailboxEnvelope> { |
| 318 | if let Some(env) = self.pending.pop_front() { |
| 319 | return Some(env); |
| 320 | } |
| 321 | self.rx.recv().await |
| 322 | } |
| 323 | |
| 324 | /// Drain all envelopes accepted before a mailbox was sealed. |
| 325 | pub(crate) fn drain_available(&mut self) -> Vec<MailboxEnvelope> { |
| 326 | while let Ok(envelope) = self.rx.try_recv() { |
| 327 | self.pending.push_back(envelope); |
| 328 | } |
| 329 | self.pending.drain(..).collect() |
| 330 | } |
| 331 | |
| 332 | /// Awaits the next envelope with a timeout. Useful in tests. |
| 333 | #[cfg(test)] |
| 334 | pub async fn recv_timeout(&mut self, timeout: Duration) -> Option<MailboxEnvelope> { |
| 335 | tokio::time::timeout(timeout, self.recv()) |
| 336 | .await |
| 337 | .ok() |
| 338 | .flatten() |
| 339 | } |
| 340 | } |
| 341 | |
| 342 | #[cfg(test)] |
| 343 | mod tests { |
| 344 | use super::*; |
| 345 | use tokio::time::Duration; |
| 346 | |
| 347 | fn open() -> (Mailbox, MailboxReceiver, CancellationToken) { |
| 348 | let token = CancellationToken::new(); |
| 349 | let (mb, rx) = Mailbox::new(token.clone()); |
| 350 | (mb, rx, token) |
| 351 | } |
| 352 | |
| 353 | fn test_route( |
| 354 | provider: ApiProvider, |
| 355 | model: &str, |
| 356 | ) -> crate::cost_status::EffectiveRouteEnvelope { |
| 357 | crate::cost_status::EffectiveRouteEnvelope::capture( |
| 358 | None, |
| 359 | provider, |
| 360 | provider.as_str(), |
| 361 | model, |
| 362 | Some(provider.default_base_url()), |
| 363 | chrono::Utc::now(), |
| 364 | ) |
| 365 | } |
| 366 | |
| 367 | #[tokio::test] |
| 368 | async fn mailbox_assigns_monotonic_sequence_numbers() { |
| 369 | let (mb, _rx, _tok) = open(); |
| 370 | let s1 = mb |
| 371 | .send(MailboxMessage::progress("a", "one")) |
| 372 | .expect("seq 1"); |
| 373 | let s2 = mb |
| 374 | .send(MailboxMessage::progress("a", "two")) |
| 375 | .expect("seq 2"); |
| 376 | let s3 = mb |
| 377 | .send(MailboxMessage::progress("b", "three")) |
| 378 | .expect("seq 3"); |
| 379 | assert_eq!(s1, 1); |
| 380 | assert_eq!(s2, 2); |
| 381 | assert_eq!(s3, 3); |
| 382 | assert!(s2 > s1 && s3 > s2); |
| 383 | } |
| 384 | |
| 385 | #[tokio::test] |
| 386 | async fn mailbox_drains_in_delivery_order() { |
| 387 | let (mb, mut rx, _tok) = open(); |
| 388 | mb.send(MailboxMessage::progress("a", "first")); |
| 389 | mb.send(MailboxMessage::progress("a", "second")); |
| 390 | mb.send(MailboxMessage::Completed { |
| 391 | agent_id: "a".into(), |
| 392 | summary: "done".into(), |
| 393 | }); |
| 394 | let drained = rx.drain(); |
| 395 | assert_eq!(drained.len(), 3); |
| 396 | assert_eq!(drained[0].seq, 1); |
| 397 | assert_eq!(drained[1].seq, 2); |
| 398 | assert_eq!(drained[2].seq, 3); |
| 399 | assert!(matches!( |
| 400 | drained[0].message, |
| 401 | MailboxMessage::Progress { .. } |
| 402 | )); |
| 403 | assert!(matches!( |
| 404 | drained[2].message, |
| 405 | MailboxMessage::Completed { .. } |
| 406 | )); |
| 407 | assert!(!rx.has_pending()); |
| 408 | } |
| 409 | |
| 410 | #[tokio::test] |
| 411 | async fn subscribers_receive_seq_bumps_for_backpressure() { |
| 412 | let (mb, _rx, _tok) = open(); |
| 413 | let mut sub_a = mb.subscribe(); |
| 414 | let mut sub_b = mb.subscribe(); |
| 415 | // Initial state: both at 0. |
| 416 | assert_eq!(*sub_a.borrow(), 0); |
| 417 | assert_eq!(*sub_b.borrow(), 0); |
| 418 | |
| 419 | mb.send(MailboxMessage::progress("x", "tick")); |
| 420 | sub_a.changed().await.expect("subscriber a sees bump"); |
| 421 | sub_b.changed().await.expect("subscriber b sees bump"); |
| 422 | assert_eq!(*sub_a.borrow(), 1); |
| 423 | assert_eq!(*sub_b.borrow(), 1); |
| 424 | |
| 425 | // A second send updates both subscribers' watch values too — even |
| 426 | // though they share a single watch channel, fanout is N-to-many. |
| 427 | mb.send(MailboxMessage::progress("x", "tick2")); |
| 428 | sub_a.changed().await.expect("a sees second bump"); |
| 429 | assert_eq!(*sub_a.borrow(), 2); |
| 430 | } |
| 431 | |
| 432 | #[tokio::test] |
| 433 | async fn close_cancels_bound_token_and_blocks_further_sends() { |
| 434 | let (mb, _rx, token) = open(); |
| 435 | assert!(!token.is_cancelled()); |
| 436 | mb.send(MailboxMessage::progress("a", "before close")); |
| 437 | mb.close(); |
| 438 | assert!(token.is_cancelled(), "close-as-cancel: token must fire"); |
| 439 | assert!(mb.is_closed()); |
| 440 | // Further sends are no-ops, returning None instead of poisoning seq. |
| 441 | assert!( |
| 442 | mb.send(MailboxMessage::progress("a", "after close")) |
| 443 | .is_none() |
| 444 | ); |
| 445 | } |
| 446 | |
| 447 | #[test] |
| 448 | fn turn_end_seal_forms_a_flush_barrier_without_cancelling_worker() { |
| 449 | let (mb, mut rx, token) = open(); |
| 450 | assert_eq!( |
| 451 | mb.send(MailboxMessage::progress("a", "accepted before barrier")), |
| 452 | Some(1) |
| 453 | ); |
| 454 | mb.seal(); |
| 455 | |
| 456 | assert!(!token.is_cancelled(), "detached worker is not cancelled"); |
| 457 | assert!( |
| 458 | mb.send(MailboxMessage::progress("a", "too late")).is_none(), |
| 459 | "no event may be accepted after the completion barrier" |
| 460 | ); |
| 461 | let drained = rx.drain_available(); |
| 462 | assert_eq!(drained.len(), 1); |
| 463 | assert_eq!(drained[0].seq, 1); |
| 464 | } |
| 465 | |
| 466 | #[tokio::test] |
| 467 | async fn close_propagates_to_child_tokens_across_max_spawn_depth() { |
| 468 | // Mirror the runtime: root → child → grandchild (default depth 3). |
| 469 | let root = CancellationToken::new(); |
| 470 | let child = root.child_token(); |
| 471 | let grandchild = child.child_token(); |
| 472 | let (mb, _rx) = Mailbox::new(root.clone()); |
| 473 | |
| 474 | assert!(!child.is_cancelled()); |
| 475 | assert!(!grandchild.is_cancelled()); |
| 476 | mb.close(); |
| 477 | assert!(child.is_cancelled(), "child inherits root close"); |
| 478 | assert!( |
| 479 | grandchild.is_cancelled(), |
| 480 | "grandchild inherits too — covers default max_spawn_depth = 3" |
| 481 | ); |
| 482 | } |
| 483 | |
| 484 | #[tokio::test] |
| 485 | async fn recv_returns_envelope_then_none_after_close_and_drop() { |
| 486 | let (mb, mut rx, _tok) = open(); |
| 487 | mb.send(MailboxMessage::progress("a", "queued")); |
| 488 | let env = rx.recv().await.expect("buffered envelope"); |
| 489 | assert_eq!(env.seq, 1); |
| 490 | |
| 491 | // After closing AND dropping the sender, recv must yield None. |
| 492 | mb.close(); |
| 493 | drop(mb); |
| 494 | let next = rx.recv_timeout(Duration::from_millis(100)).await; |
| 495 | assert!(next.is_none(), "drained + dropped → recv yields None"); |
| 496 | } |
| 497 | |
| 498 | #[tokio::test] |
| 499 | async fn cloned_mailbox_shares_sequence_and_close_state() { |
| 500 | let (mb, mut rx, token) = open(); |
| 501 | let mb_clone = mb.clone(); |
| 502 | let s1 = mb |
| 503 | .send(MailboxMessage::progress("a", "from original")) |
| 504 | .unwrap(); |
| 505 | let s2 = mb_clone |
| 506 | .send(MailboxMessage::progress("a", "from clone")) |
| 507 | .unwrap(); |
| 508 | assert_eq!(s1, 1); |
| 509 | assert_eq!(s2, 2, "clones share the seq counter"); |
| 510 | |
| 511 | let drained = rx.drain(); |
| 512 | assert_eq!(drained.len(), 2); |
| 513 | |
| 514 | // Closing through one clone closes them all (the AtomicBool is shared). |
| 515 | mb_clone.close(); |
| 516 | assert!(mb.is_closed()); |
| 517 | assert!(token.is_cancelled()); |
| 518 | } |
| 519 | |
| 520 | #[test] |
| 521 | fn work_state_payload_round_trips_and_tolerates_a_missing_snapshot() { |
| 522 | use crate::tools::todo::{TodoItem, TodoStatus}; |
| 523 | |
| 524 | let message = MailboxMessage::work_state( |
| 525 | "agent_child", |
| 526 | TodoListSnapshot { |
| 527 | items: vec![TodoItem { |
| 528 | id: 2, |
| 529 | content: "write the projection".to_string(), |
| 530 | status: TodoStatus::InProgress, |
| 531 | }], |
| 532 | completion_pct: 50, |
| 533 | in_progress_id: Some(2), |
| 534 | }, |
| 535 | ); |
| 536 | let encoded = serde_json::to_string(&message).expect("encode"); |
| 537 | let decoded: MailboxMessage = serde_json::from_str(&encoded).expect("decode"); |
| 538 | assert_eq!(decoded, message); |
| 539 | |
| 540 | // An older payload that predates per-agent Work state decodes to an |
| 541 | // empty snapshot rather than failing the whole stream. |
| 542 | let legacy: MailboxMessage = |
| 543 | serde_json::from_str(r#"{"kind":"work_state","agent_id":"agent_old"}"#) |
| 544 | .expect("legacy"); |
| 545 | assert_eq!(legacy.agent_id(), "agent_old"); |
| 546 | match legacy { |
| 547 | MailboxMessage::WorkState { todo, .. } => assert!(todo.is_empty()), |
| 548 | other => panic!("expected work state, got {other:?}"), |
| 549 | } |
| 550 | |
| 551 | // Every pre-existing variant still decodes unchanged. |
| 552 | let started: MailboxMessage = |
| 553 | serde_json::from_str(r#"{"kind":"started","agent_id":"a","agent_type":"worker"}"#) |
| 554 | .expect("started"); |
| 555 | assert_eq!(started.agent_id(), "a"); |
| 556 | } |
| 557 | |
| 558 | #[tokio::test] |
| 559 | async fn agent_id_is_extractable_from_every_variant() { |
| 560 | let cases: Vec<(MailboxMessage, &str)> = vec![ |
| 561 | (MailboxMessage::started("a1", FleetRole::Worker), "a1"), |
| 562 | (MailboxMessage::progress("a2", "x"), "a2"), |
| 563 | ( |
| 564 | MailboxMessage::ToolCallStarted { |
| 565 | agent_id: "a3".into(), |
| 566 | tool_name: "read_file".into(), |
| 567 | step: 1, |
| 568 | }, |
| 569 | "a3", |
| 570 | ), |
| 571 | ( |
| 572 | MailboxMessage::ToolCallCompleted { |
| 573 | agent_id: "a4".into(), |
| 574 | tool_name: "read_file".into(), |
| 575 | step: 1, |
| 576 | ok: true, |
| 577 | }, |
| 578 | "a4", |
| 579 | ), |
| 580 | ( |
| 581 | MailboxMessage::ChildSpawned { |
| 582 | parent_id: "parent".into(), |
| 583 | child_id: "a5".into(), |
| 584 | }, |
| 585 | "a5", |
| 586 | ), |
| 587 | ( |
| 588 | MailboxMessage::Completed { |
| 589 | agent_id: "a6".into(), |
| 590 | summary: "done".into(), |
| 591 | }, |
| 592 | "a6", |
| 593 | ), |
| 594 | ( |
| 595 | MailboxMessage::Failed { |
| 596 | agent_id: "a7".into(), |
| 597 | error: "boom".into(), |
| 598 | }, |
| 599 | "a7", |
| 600 | ), |
| 601 | ( |
| 602 | MailboxMessage::Cancelled { |
| 603 | agent_id: "a8".into(), |
| 604 | }, |
| 605 | "a8", |
| 606 | ), |
| 607 | ( |
| 608 | MailboxMessage::Interrupted { |
| 609 | agent_id: "a10".into(), |
| 610 | reason: "API call timed out".into(), |
| 611 | }, |
| 612 | "a10", |
| 613 | ), |
| 614 | ( |
| 615 | MailboxMessage::work_state("a11", TodoListSnapshot::default()), |
| 616 | "a11", |
| 617 | ), |
| 618 | ( |
| 619 | MailboxMessage::TokenUsage { |
| 620 | agent_id: "a9".into(), |
| 621 | source_id: "response-a9".into(), |
| 622 | route: Box::new(test_route(ApiProvider::Deepseek, "deepseek-v4-flash")), |
| 623 | usage: Usage { |
| 624 | input_tokens: 100, |
| 625 | output_tokens: 50, |
| 626 | ..Default::default() |
| 627 | }, |
| 628 | }, |
| 629 | "a9", |
| 630 | ), |
| 631 | ]; |
| 632 | for (msg, expected) in cases { |
| 633 | assert_eq!(msg.agent_id(), expected, "extract failed for {msg:?}"); |
| 634 | } |
| 635 | } |
| 636 | |
| 637 | #[test] |
| 638 | fn token_usage_serde_round_trip_preserves_immutable_route_evidence() { |
| 639 | let route = crate::cost_status::EffectiveRouteEnvelope { |
| 640 | openrouter_vendor: None, |
| 641 | provider: ApiProvider::Moonshot, |
| 642 | provider_identity: "kimi-membership".to_string(), |
| 643 | model: "k3".to_string(), |
| 644 | billing_surface: Some(crate::pricing::MOONSHOT_KIMI_CODE_BILLING_SURFACE.to_string()), |
| 645 | endpoint_fingerprint: Some("a".repeat(64)), |
| 646 | provider_live_pricing: None, |
| 647 | billing_mode: crate::cost_status::RouteBillingMode::Subscription, |
| 648 | dispatched_at: chrono::DateTime::<chrono::Utc>::from_timestamp(1_234, 0) |
| 649 | .expect("timestamp"), |
| 650 | }; |
| 651 | let message = |
| 652 | MailboxMessage::token_usage("agent-k3", "response-k3", route, Usage::default()); |
| 653 | let json = serde_json::to_string(&message).expect("serialize token usage"); |
| 654 | let restored: MailboxMessage = |
| 655 | serde_json::from_str(&json).expect("deserialize token usage"); |
| 656 | assert_eq!(restored, message); |
| 657 | } |
| 658 | } |
| 659 |