kanade-agent 0.43.40

Windows-side resident daemon for the kanade endpoint-management system. Subscribes to commands.* over NATS, runs scripts, publishes WMI inventory + heartbeats, watches for self-updates
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
//! `system.*` method handlers (SPEC §2.12.5).
//!
//! - `system.handshake` — protocol-version negotiation + session
//!   info (SPEC §2.12.6).
//! - `system.ping` — round-trip liveness check.
//! - `system.version` — running agent version + self-update
//!   target + (eventually) pinned client version.
//! - `system.log_tail` — last N lines of agent.log, for support
//!   handoff diagnostics.

use std::io;

use kanade_shared::ipc::error::{ErrorKind, RpcError};
use kanade_shared::ipc::handshake::{HandshakeParams, HandshakeResult, PROTOCOL_V1};
use kanade_shared::ipc::system::{
    LogTailParams, LogTailResult, PingParams, PingResult, VersionParams, VersionResult,
};
use tracing::warn;

use super::super::connection::ConnectionState;

/// Result type for KLP handlers — either a JSON-encoded response
/// payload or a [`RpcError`] the dispatcher will wrap into the
/// envelope's `error` slot. `anyhow::Error` is reserved for
/// internal failures the dispatcher turns into
/// [`ErrorKind::InternalError`] (-32603).
pub type HandlerResult<T> = std::result::Result<T, RpcError>;

/// Features this agent currently advertises in handshake. None
/// yet — push handlers (`push.notifications`, `push.jobs`,
/// `push.state`, `support.diagnostics`) land in their own PRs and
/// each one adds an entry here.
const SUPPORTED_FEATURES: &[&str] = &[];

/// Per-call cap on the number of log lines `system.log_tail`
/// returns, no matter what the caller asks for. Keeps the response
/// inside the 1 MiB framing cap (SPEC §2.12.2) with comfortable
/// headroom — at ~1 KiB per line (the agent's longest typical
/// `tracing` event), 1000 lines fits in ~1 MiB. Callers needing
/// more pull the full file via `support.upload_diagnostics`.
const LOG_TAIL_HARD_CAP: u32 = 1000;

/// Files at or below this size take the simple whole-file read
/// path. Past it, [`read_tail_lines`] seeks near the end and reads
/// forward so a chatty long-uptime endpoint with a hundreds-of-MB
/// log doesn't balloon agent RSS just to serve a few hundred tail
/// lines (issue #289). Daily-rotated logs are normally well under
/// this, so the common case never pays the extra complexity.
const TAIL_WHOLE_FILE_THRESHOLD: u64 = 4 * 1024 * 1024;

/// Per-line byte budget for the reverse-read seek heuristic. The
/// agent's `tracing` lines run comfortably under this; over-
/// estimating only costs a slightly larger initial read, never
/// correctness — the window grows and retries if it held too few
/// lines.
const TAIL_AVG_LINE_BYTES: u64 = 2 * 1024;

/// Hard cap on how far back [`read_tail_lines`] will seek while
/// hunting for enough lines. Bounds worst-case memory for
/// pathological logs (one giant line, or no newlines at all). On
/// hitting it we return whatever the window held rather than walk
/// the whole file.
const TAIL_MAX_WINDOW_BYTES: u64 = 32 * 1024 * 1024;

/// `system.handshake` — protocol negotiation + session info.
///
/// SPEC §2.12.6:
/// - Pick the highest mutually-supported version from
///   `params.protocol`. KLP v1 only knows `PROTOCOL_V1` today.
/// - Return the agent's session info derived from the OS
///   (`conn.session()`), never from payload.
/// - If no protocol overlap, return
///   [`ErrorKind::StaleProtocol`].
pub fn handle_handshake(
    conn: &mut ConnectionState,
    params: HandshakeParams,
) -> HandlerResult<HandshakeResult> {
    if params.protocol.is_empty() {
        return Err(RpcError::new(
            ErrorKind::InvalidParams,
            "handshake.protocol must contain at least one version",
        ));
    }

    // Pick highest mutually-supported. Today we only know v1, so
    // the question collapses to "does the client mention v1?".
    let agreed = params
        .protocol
        .iter()
        .copied()
        .filter(|&v| v == PROTOCOL_V1)
        .max();

    let Some(agreed) = agreed else {
        return Err(RpcError::new(
            ErrorKind::StaleProtocol,
            format!(
                "no overlap with client versions {:?} (agent supports {:?})",
                params.protocol,
                [PROTOCOL_V1],
            ),
        ));
    };

    conn.mark_handshake(agreed);

    Ok(HandshakeResult {
        protocol: agreed,
        agent_version: conn.agent_version.clone(),
        features: SUPPORTED_FEATURES.iter().map(|&s| s.to_string()).collect(),
        session: conn.session(),
    })
}

/// `system.ping` — agent wall-clock at the moment it answered.
/// No params required; the envelope's `params: omitted` is
/// equivalent to `params: {}` via the
/// [`kanade_shared::ipc::envelope::decode_params`] helper.
pub fn handle_ping(_conn: &ConnectionState, _params: PingParams) -> HandlerResult<PingResult> {
    Ok(PingResult {
        agent_time: chrono::Utc::now(),
    })
}

/// `system.version` — running agent version + self-update target.
///
/// `target_agent_version` comes from the live `EffectiveConfig`
/// the supervisor (Sprint 6) publishes on the watch channel; when
/// the KV stack has set no rollout target (or set the empty
/// string by mistake), it falls back to the running version so
/// the SPA's "restart pending" banner stays hidden when nothing's
/// actually pending.
///
/// `target_client_version` is `None` until the backend publishes
/// a pinned Client App version through a future config field —
/// Sprint 8 ships the Client App without a forced-upgrade banner.
pub fn handle_version(
    conn: &ConnectionState,
    _params: VersionParams,
) -> HandlerResult<VersionResult> {
    // Defensive filter against `Some("")` — a backend that
    // confuses "cleared" with "empty string" would otherwise
    // surface `target_agent_version = ""` and trip a phantom
    // restart banner on the SPA.
    let target = conn
        .config_rx
        .borrow()
        .target_version
        .as_deref()
        .filter(|s| !s.is_empty())
        .map(str::to_owned)
        .unwrap_or_else(|| conn.agent_version.clone());
    Ok(VersionResult {
        agent_version: conn.agent_version.clone(),
        target_agent_version: target,
        target_client_version: None,
    })
}

/// `system.log_tail` — last N lines of `agent.log` for support
/// handoff.
///
/// The `log_path` on `ConnectionState` is the *template* path from
/// `cfg.log.path` (e.g. `agent.log`); the daily-rotation
/// `tracing_appender` actually writes to
/// `<dir>/<stem>.YYYY-MM-DD.<ext>`, so we resolve to the active
/// file via [`crate::logs::locate_active_file`] before reading
/// (same function the NATS `logs.fetch` path uses).
///
/// Clamping semantics:
/// - Caller's `lines` is capped at [`LOG_TAIL_HARD_CAP`]. When
///   the cap kicks in (caller asked for more than the cap),
///   `truncated` is set so the SPA can prompt to escalate to
///   `support.upload_diagnostics`. `truncated` is NOT set merely
///   because the file has more lines than the caller asked for
///   — the caller got what they asked for.
/// - File-not-found returns an empty `lines` (not an error) — the
///   agent's logger may not have written anything yet on a fresh
///   boot.
///
/// Reading is delegated to [`read_tail_lines`], which keeps the
/// simple whole-file path for small (daily-rotated) logs and falls
/// back to a bounded reverse-read for files past
/// [`TAIL_WHOLE_FILE_THRESHOLD`] so a hundreds-of-MB log can't spike
/// agent RSS (issue #289).
pub async fn handle_log_tail(
    conn: &ConnectionState,
    params: LogTailParams,
) -> HandlerResult<LogTailResult> {
    let requested = params.lines.min(LOG_TAIL_HARD_CAP);
    let truncated = params.lines > LOG_TAIL_HARD_CAP;

    let active_path = match crate::logs::locate_active_file(&conn.log_path).await {
        Ok(p) => p,
        Err(e) => {
            // `locate_active_file` only fails when the directory
            // itself is missing — treat the same as a missing
            // active file. Surface for diagnostics but don't
            // block the caller's UI.
            warn!(
                error = %e,
                template = %conn.log_path.display(),
                "system.log_tail: log directory missing, returning empty result",
            );
            return Ok(LogTailResult {
                lines: vec![],
                truncated,
            });
        }
    };

    let lines = match read_tail_lines(&active_path, requested as usize).await {
        Ok(lines) => lines,
        Err(e) if e.kind() == io::ErrorKind::NotFound => {
            warn!(
                path = %active_path.display(),
                "system.log_tail: log file not found, returning empty result",
            );
            return Ok(LogTailResult {
                lines: vec![],
                truncated,
            });
        }
        Err(e) => {
            return Err(RpcError::new(
                ErrorKind::InternalError,
                format!("read agent.log ({}): {e}", active_path.display()),
            ));
        }
    };

    Ok(LogTailResult { lines, truncated })
}

/// Read the last `requested` lines of `path`, oldest-first (the
/// order `str::lines` yields), without slurping the whole file when
/// it's large.
///
/// Small files (≤ [`TAIL_WHOLE_FILE_THRESHOLD`]) take the plain
/// whole-file read — the added seek/retry machinery isn't worth it
/// for the daily-rotated common case. Larger files seek to roughly
/// `len - requested × avg_line` and read forward, doubling the
/// window (capped at [`TAIL_MAX_WINDOW_BYTES`]) until it holds
/// enough lines or reaches the start of the file.
///
/// Propagates `io::Error` (the caller maps `NotFound` to an empty
/// result).
async fn read_tail_lines(path: &std::path::Path, requested: usize) -> io::Result<Vec<String>> {
    use tokio::io::{AsyncReadExt, AsyncSeekExt, SeekFrom};

    if requested == 0 {
        return Ok(vec![]);
    }

    let mut file = tokio::fs::File::open(path).await?;
    let len = file.metadata().await?.len();

    if len <= TAIL_WHOLE_FILE_THRESHOLD {
        let mut body = String::new();
        file.read_to_string(&mut body).await?;
        return Ok(tail_of(&body, requested));
    }

    // Large file: seek near the end and read forward, growing the
    // window until it holds enough lines (or we reach offset 0 /
    // the cap). Start from a line-count-based estimate.
    let mut window = (requested as u64)
        .saturating_mul(TAIL_AVG_LINE_BYTES)
        .clamp(TAIL_AVG_LINE_BYTES, TAIL_MAX_WINDOW_BYTES);

    loop {
        let start = len.saturating_sub(window);
        file.seek(SeekFrom::Start(start)).await?;
        let mut buf = Vec::with_capacity(window.min(len) as usize);
        // `take(window)` makes the window a *hard* byte bound. The
        // agent's own log is appended to continuously, so the file
        // can grow between `metadata()` above and this read; a plain
        // `read_to_end` would chase the moving EOF and could blow
        // past `window` (and `TAIL_MAX_WINDOW_BYTES` on the capped
        // iteration), defeating the whole point of the bounded read.
        (&mut file).take(window).read_to_end(&mut buf).await?;

        // When we seeked past offset 0 the first line is almost
        // certainly a mid-line fragment — drop everything up to and
        // including the first `\n`. `\n` (0x0A) never appears inside
        // a multibyte UTF-8 sequence, so the remainder sits on a
        // clean UTF-8 boundary even though the seek itself wasn't
        // char-aligned. No newline in the whole window means a
        // single line longer than `window`: drop it all and grow.
        let usable: &[u8] = if start > 0 {
            match buf.iter().position(|&b| b == b'\n') {
                Some(nl) => &buf[nl + 1..],
                None => &[],
            }
        } else {
            &buf
        };

        let text = String::from_utf8_lossy(usable);
        let lines = tail_of(&text, requested);

        // Done when we have enough, can't go back further, or hit
        // the cap (return whatever the bounded window held).
        if lines.len() >= requested || start == 0 || window >= TAIL_MAX_WINDOW_BYTES {
            return Ok(lines);
        }
        window = window.saturating_mul(2).min(TAIL_MAX_WINDOW_BYTES);
    }
}

/// Last `n` lines of `body`, trailing `\r\n` / `\n` stripped, in
/// file order. Shared by both read paths in [`read_tail_lines`].
///
/// `str::Lines` is a `DoubleEndedIterator`, so we walk backward and
/// `take(n)` instead of collecting every line: O(n) lines scanned
/// and allocated rather than O(buffer) — matters on the large-file
/// path where the window can hold far more lines than requested.
fn tail_of(body: &str, n: usize) -> Vec<String> {
    let mut lines: Vec<String> = body.lines().rev().take(n).map(str::to_string).collect();
    lines.reverse();
    lines
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::klp::auth::PeerCredentials;
    use kanade_shared::ipc::state::StateSnapshot;
    use kanade_shared::wire::EffectiveConfig;
    use std::path::PathBuf;
    use tempfile::NamedTempFile;
    use tokio::sync::{mpsc, watch};

    fn dummy_snapshot() -> StateSnapshot {
        StateSnapshot {
            pc_id: "PC1234".into(),
            online: true,
            vpn: "unknown".into(),
            checks: vec![],
            agent_version: "0.40.0".into(),
            target_version: "0.40.0".into(),
        }
    }

    fn fresh_conn_with(cfg: EffectiveConfig, log_path: PathBuf) -> ConnectionState {
        let (_cfg_tx, cfg_rx) = watch::channel(cfg);
        let (_state_tx, state_rx) = watch::channel(dummy_snapshot());
        let (push_tx, _push_rx) = mpsc::channel(8);
        ConnectionState::new(
            PeerCredentials {
                user: "DOMAIN\\alice".into(),
                session_id: 2,
            },
            "PC1234".into(),
            "0.40.0".into(),
            cfg_rx,
            state_rx,
            log_path,
            push_tx,
        )
    }

    fn fresh_conn() -> ConnectionState {
        fresh_conn_with(
            EffectiveConfig::builtin_defaults(),
            PathBuf::from("agent.log"),
        )
    }

    #[test]
    fn handshake_v1_only_client_succeeds() {
        let mut conn = fresh_conn();
        let result = handle_handshake(
            &mut conn,
            HandshakeParams {
                client: "kanade-client".into(),
                client_version: "0.1.0".into(),
                protocol: vec![PROTOCOL_V1],
                features: vec![],
            },
        )
        .expect("handshake should succeed");
        assert_eq!(result.protocol, PROTOCOL_V1);
        assert_eq!(result.agent_version, "0.40.0");
        assert_eq!(result.session.user, "DOMAIN\\alice");
        assert_eq!(result.session.pc_id, "PC1234");
        assert!(conn.handshake_complete());
    }

    #[test]
    fn handshake_picks_highest_mutual_version_from_multi_version_client() {
        // Future-proof: a client advertising [1, 2] talking to a
        // v1-only agent must downshift to 1, not error.
        let mut conn = fresh_conn();
        let result = handle_handshake(
            &mut conn,
            HandshakeParams {
                client: "kanade-client".into(),
                client_version: "0.2.0".into(),
                protocol: vec![1, 2, 3],
                features: vec![],
            },
        )
        .expect("handshake should succeed via downshift");
        assert_eq!(result.protocol, PROTOCOL_V1);
    }

    #[test]
    fn handshake_rejects_empty_protocol_list() {
        let mut conn = fresh_conn();
        let err = handle_handshake(
            &mut conn,
            HandshakeParams {
                client: "kanade-client".into(),
                client_version: "0.1.0".into(),
                protocol: vec![],
                features: vec![],
            },
        )
        .expect_err("empty protocol must fail");
        let data = err.data.as_ref().expect("data populated");
        assert_eq!(data.kind, ErrorKind::InvalidParams);
        assert!(!conn.handshake_complete(), "conn must stay pre-handshake");
    }

    #[test]
    fn handshake_rejects_when_no_version_overlap() {
        // Client speaks only v2 / v3; agent speaks only v1 → no
        // overlap → StaleProtocol per SPEC §2.12.6.
        let mut conn = fresh_conn();
        let err = handle_handshake(
            &mut conn,
            HandshakeParams {
                client: "kanade-client".into(),
                client_version: "9.9.9".into(),
                protocol: vec![2, 3],
                features: vec![],
            },
        )
        .expect_err("must fail with StaleProtocol");
        let data = err.data.as_ref().expect("data populated");
        assert_eq!(data.kind, ErrorKind::StaleProtocol);
        assert!(!conn.handshake_complete());
    }

    #[test]
    fn ping_returns_recent_agent_time() {
        let conn = fresh_conn();
        let before = chrono::Utc::now();
        let result = handle_ping(&conn, PingParams::default()).unwrap();
        let after = chrono::Utc::now();
        assert!(
            result.agent_time >= before && result.agent_time <= after,
            "agent_time {} should be in [{before}, {after}]",
            result.agent_time
        );
    }

    #[test]
    fn handshake_advertises_no_features_yet() {
        // SUPPORTED_FEATURES stays empty until each push handler
        // actually lands — advertising features whose methods
        // aren't routed would mislead clients about what we can do.
        let mut conn = fresh_conn();
        let result = handle_handshake(
            &mut conn,
            HandshakeParams {
                client: "kanade-client".into(),
                client_version: "0.1.0".into(),
                protocol: vec![PROTOCOL_V1],
                features: vec!["push.notifications".into()],
            },
        )
        .unwrap();
        assert!(
            result.features.is_empty(),
            "no features advertised yet; got {:?}",
            result.features,
        );
    }

    // ---- system.version ----

    #[test]
    fn version_falls_back_to_running_when_no_target_set() {
        // EffectiveConfig::builtin_defaults() has target_version =
        // None. The handler should report target_agent_version =
        // agent_version so the SPA doesn't show a phantom
        // "restart pending" banner.
        let conn = fresh_conn();
        let result = handle_version(&conn, VersionParams::default()).unwrap();
        assert_eq!(result.agent_version, "0.40.0");
        assert_eq!(result.target_agent_version, "0.40.0");
        assert!(result.target_client_version.is_none());
    }

    #[test]
    fn version_returns_distinct_target_when_supervisor_set_one() {
        // Sprint 6 supervisor has published a rollout target —
        // the handler must surface it so the SPA shows the
        // "restart pending" banner.
        let mut cfg = EffectiveConfig::builtin_defaults();
        cfg.target_version = Some("0.42.0".into());
        let conn = fresh_conn_with(cfg, PathBuf::from("agent.log"));
        let result = handle_version(&conn, VersionParams::default()).unwrap();
        assert_eq!(result.agent_version, "0.40.0");
        assert_eq!(result.target_agent_version, "0.42.0");
    }

    // ---- system.log_tail ----

    #[tokio::test]
    async fn log_tail_returns_empty_when_file_missing() {
        // Fresh-boot scenario: parent dir exists (the agent set
        // up its log directory at startup) but no log file has
        // been written yet. Don't error — return an empty result
        // so the SPA support flow works the first time it's
        // invoked.
        let tmpdir = tempfile::tempdir().unwrap();
        let conn = fresh_conn_with(
            EffectiveConfig::builtin_defaults(),
            tmpdir.path().join("agent.log"),
        );
        let result = handle_log_tail(&conn, LogTailParams::default())
            .await
            .expect("missing file is not an error");
        assert!(result.lines.is_empty());
        assert!(!result.truncated);
    }

    #[tokio::test]
    async fn log_tail_picks_rotated_file_matching_template() {
        // The HIGH bug from agy's review: tracing_appender writes
        // to `<stem>.YYYY-MM-DD.<ext>`, not the bare template
        // path. The handler must resolve to the active file via
        // locate_active_file before reading.
        let tmpdir = tempfile::tempdir().unwrap();
        let template = tmpdir.path().join("agent.log");
        // The actual file the appender would have created today.
        let active = tmpdir.path().join("agent.2026-05-24.log");
        std::fs::write(&active, "first\nsecond\nthird\n").unwrap();
        let conn = fresh_conn_with(EffectiveConfig::builtin_defaults(), template);
        let result = handle_log_tail(&conn, LogTailParams { lines: 10 })
            .await
            .expect("locate_active_file should find the rotated file");
        assert_eq!(result.lines, vec!["first", "second", "third"]);
    }

    #[tokio::test]
    async fn log_tail_returns_all_lines_when_file_smaller_than_request() {
        let f = NamedTempFile::new().unwrap();
        std::fs::write(f.path(), "alpha\nbeta\ngamma\n").unwrap();
        let conn = fresh_conn_with(EffectiveConfig::builtin_defaults(), f.path().to_path_buf());
        let result = handle_log_tail(
            &conn,
            LogTailParams {
                lines: 100, // way more than the file has
            },
        )
        .await
        .unwrap();
        assert_eq!(result.lines, vec!["alpha", "beta", "gamma"]);
        assert!(!result.truncated);
    }

    #[tokio::test]
    async fn log_tail_returns_only_last_n_when_file_larger_than_request() {
        let f = NamedTempFile::new().unwrap();
        let body = (1..=20)
            .map(|i| format!("line-{i:02}"))
            .collect::<Vec<_>>()
            .join("\n");
        std::fs::write(f.path(), body).unwrap();
        let conn = fresh_conn_with(EffectiveConfig::builtin_defaults(), f.path().to_path_buf());
        let result = handle_log_tail(&conn, LogTailParams { lines: 5 })
            .await
            .unwrap();
        assert_eq!(
            result.lines,
            vec!["line-16", "line-17", "line-18", "line-19", "line-20"]
        );
        // Caller asked for 5, we returned 5 — well within cap, so
        // not truncated (the extra unread lines don't count as
        // "truncated by the agent" — that label is reserved for
        // the cap-applied case so SPA can prompt for full log).
        assert!(!result.truncated);
    }

    #[tokio::test]
    async fn log_tail_clamps_to_hard_cap_and_flags_truncated() {
        // Build a file with > HARD_CAP lines so the clamp visibly
        // bites and the truncated flag flips.
        let f = NamedTempFile::new().unwrap();
        let body = (1..=(LOG_TAIL_HARD_CAP + 50))
            .map(|i| format!("L{i}"))
            .collect::<Vec<_>>()
            .join("\n");
        std::fs::write(f.path(), body).unwrap();
        let conn = fresh_conn_with(EffectiveConfig::builtin_defaults(), f.path().to_path_buf());
        // Caller asks for HARD_CAP + 100 → clamped down.
        let result = handle_log_tail(
            &conn,
            LogTailParams {
                lines: LOG_TAIL_HARD_CAP + 100,
            },
        )
        .await
        .unwrap();
        assert_eq!(result.lines.len(), LOG_TAIL_HARD_CAP as usize);
        assert!(result.truncated, "asking past the cap must set truncated");
        // Confirm it's the tail (newest), not the head.
        assert_eq!(
            result.lines.last().unwrap(),
            &format!("L{}", LOG_TAIL_HARD_CAP + 50)
        );
    }

    #[tokio::test]
    async fn log_tail_default_is_200_lines() {
        // SPEC default in LogTailParams is 200. Just verify the
        // default doesn't trigger the truncated flag (200 << cap).
        let f = NamedTempFile::new().unwrap();
        let body = (1..=500)
            .map(|i| format!("L{i}"))
            .collect::<Vec<_>>()
            .join("\n");
        std::fs::write(f.path(), body).unwrap();
        let conn = fresh_conn_with(EffectiveConfig::builtin_defaults(), f.path().to_path_buf());
        let result = handle_log_tail(&conn, LogTailParams::default())
            .await
            .unwrap();
        assert_eq!(result.lines.len(), 200);
        assert!(!result.truncated);
    }

    // ---- system.log_tail: bounded reverse-read (issue #289) ----

    /// Build a file comfortably past `TAIL_WHOLE_FILE_THRESHOLD` so
    /// the large-file seek path is exercised, with line content that
    /// uniquely identifies each line so we can prove the tail is
    /// correct (and that the mid-line fragment at the seek boundary
    /// was dropped). Each line is padded to ~80 bytes; `count` lines
    /// of that easily clears the 4 MiB threshold by count ≥ 60k.
    fn write_big_log(path: &std::path::Path, count: usize) {
        let pad = "x".repeat(64);
        let body = (1..=count)
            .map(|i| format!("line-{i:08}-{pad}"))
            .collect::<Vec<_>>()
            .join("\n");
        assert!(
            body.len() as u64 > TAIL_WHOLE_FILE_THRESHOLD,
            "fixture must exceed the whole-file threshold to hit the seek path; \
             got {} bytes",
            body.len(),
        );
        std::fs::write(path, body).unwrap();
    }

    #[tokio::test]
    async fn log_tail_large_file_returns_correct_tail_via_seek() {
        // The whole point of #289: a multi-MB log must still yield
        // the exact last N lines, and the partial leading line at
        // the seek boundary must be dropped (otherwise the first
        // returned line would be a fragment, not "line-00099901-…").
        let f = NamedTempFile::new().unwrap();
        let count = 100_000usize;
        write_big_log(f.path(), count);
        let conn = fresh_conn_with(EffectiveConfig::builtin_defaults(), f.path().to_path_buf());

        let result = handle_log_tail(&conn, LogTailParams { lines: 100 })
            .await
            .unwrap();

        assert_eq!(result.lines.len(), 100);
        let pad = "x".repeat(64);
        // First and last returned lines are intact (no fragment) and
        // are exactly the last 100 lines of the file.
        assert_eq!(result.lines[0], format!("line-{:08}-{pad}", count - 99));
        assert_eq!(
            result.lines.last().unwrap(),
            &format!("line-{count:08}-{pad}")
        );
        assert!(!result.truncated);
    }

    #[tokio::test]
    async fn log_tail_large_file_grows_window_when_initial_estimate_too_small() {
        // Force the grow loop: with ~3 KiB lines, the initial
        // `requested × TAIL_AVG_LINE_BYTES` (= 2 KiB/line) window
        // holds fewer lines than requested, so the first read falls
        // short and the window must double before we have enough.
        // The result must still be the exact tail.
        let f = NamedTempFile::new().unwrap();
        let line_pad = "z".repeat(3 * 1024); // ~3 KiB per line, > 2 KiB estimate
        let count = 2_000usize; // ~6 MiB total, past the threshold
        let body = (1..=count)
            .map(|i| format!("line-{i:06}-{line_pad}"))
            .collect::<Vec<_>>()
            .join("\n");
        assert!(body.len() as u64 > TAIL_WHOLE_FILE_THRESHOLD);
        std::fs::write(f.path(), &body).unwrap();
        let conn = fresh_conn_with(EffectiveConfig::builtin_defaults(), f.path().to_path_buf());

        let result = handle_log_tail(&conn, LogTailParams { lines: 50 })
            .await
            .unwrap();

        assert_eq!(result.lines.len(), 50);
        assert_eq!(
            result.lines[0],
            format!("line-{:06}-{line_pad}", count - 49)
        );
        assert_eq!(
            result.lines.last().unwrap(),
            &format!("line-{count:06}-{line_pad}")
        );
    }

    #[tokio::test]
    async fn log_tail_large_file_request_exceeds_line_count() {
        // Large file but fewer lines than requested: must return all
        // of them (exiting the grow loop at start == 0), not loop
        // forever or drop the genuine first line of the file.
        let f = NamedTempFile::new().unwrap();
        // Few but very long lines to clear the threshold without a
        // huge line count.
        let huge = "y".repeat(2 * 1024 * 1024); // 2 MiB per line
        let body = format!("first-{huge}\nsecond-{huge}\nthird-{huge}");
        assert!(body.len() as u64 > TAIL_WHOLE_FILE_THRESHOLD);
        std::fs::write(f.path(), &body).unwrap();
        let conn = fresh_conn_with(EffectiveConfig::builtin_defaults(), f.path().to_path_buf());

        let result = handle_log_tail(&conn, LogTailParams { lines: 100 })
            .await
            .unwrap();

        assert_eq!(result.lines.len(), 3);
        assert!(result.lines[0].starts_with("first-"));
        assert!(result.lines[2].starts_with("third-"));
    }

    #[tokio::test]
    async fn log_tail_large_file_window_with_no_newline_grows_until_it_finds_one() {
        // Exercises the `None => &[]` branch in read_tail_lines:
        // the file's last line is longer than the initial window, so
        // the first seek (start > 0) lands inside it and finds zero
        // newlines. The window must keep doubling — discarding each
        // newline-less read — until it reaches the `head\n` boundary
        // and returns the (single) requested last line intact.
        let f = NamedTempFile::new().unwrap();
        let huge = "B".repeat(5 * 1024 * 1024); // 5 MiB single last line
        let body = format!("head\n{huge}");
        assert!(body.len() as u64 > TAIL_WHOLE_FILE_THRESHOLD);
        // The last line must dwarf the initial 1-line window estimate
        // so the first read genuinely contains no `\n`.
        assert!(huge.len() as u64 > TAIL_AVG_LINE_BYTES);
        std::fs::write(f.path(), &body).unwrap();
        let conn = fresh_conn_with(EffectiveConfig::builtin_defaults(), f.path().to_path_buf());

        let result = handle_log_tail(&conn, LogTailParams { lines: 1 })
            .await
            .unwrap();

        assert_eq!(result.lines, vec![huge]);
    }
}