symforge 11.3.1

SymForge — in-memory code intelligence MCP server for Claude Code, Codex, and Gemini CLI
Documentation
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
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
use std::fs;
use std::io::{self, Write};
use std::path::{Path, PathBuf};
use std::time::{SystemTime, UNIX_EPOCH};

use serde::{Deserialize, Serialize};
use serde_json::{Map, Value, json};

use crate::domain::{ControlStateDir, ProjectStateDir};
use crate::{hash, paths};

const KEY_HASH_FRAME_PREFIX: &[u8] = b"symforge-idempotency-key-v1\0";
const REQUEST_HASH_FRAME_PREFIX: &[u8] = b"symforge-idempotency-request-v1\0";
const REPLAY_RECORD_SCHEMA_VERSION: u8 = 1;
const RECORD_FILE_NAME: &str = "record.json";

#[derive(Debug, Clone, PartialEq, Eq)]
pub struct IdempotencyKey(String);

impl IdempotencyKey {
    pub fn new(raw: impl Into<String>) -> Result<Self, IdempotencyError> {
        let raw = raw.into();
        if raw.is_empty() {
            return Err(IdempotencyError::EmptyKey);
        }
        Ok(Self(raw))
    }

    fn key_hash(&self) -> String {
        let mut frame = Vec::with_capacity(KEY_HASH_FRAME_PREFIX.len() + self.0.len());
        frame.extend_from_slice(KEY_HASH_FRAME_PREFIX);
        frame.extend_from_slice(self.0.as_bytes());
        hash::digest_hex(&frame)
    }
}

#[derive(Debug, Clone, PartialEq, Eq, Hash, Serialize, Deserialize)]
#[serde(transparent)]
pub struct RequestHash(String);

impl RequestHash {
    pub fn for_tool_request(tool_name: &str, request: &Value) -> Result<Self, IdempotencyError> {
        if tool_name.is_empty() {
            return Err(IdempotencyError::EmptyToolName);
        }

        let canonical = canonical_json_bytes(request)?;
        let mut frame = Vec::with_capacity(
            REQUEST_HASH_FRAME_PREFIX.len() + tool_name.len() + 1 + canonical.len(),
        );
        frame.extend_from_slice(REQUEST_HASH_FRAME_PREFIX);
        frame.extend_from_slice(tool_name.as_bytes());
        frame.push(0);
        frame.extend_from_slice(&canonical);

        Ok(Self(hash::digest_hex(&frame)))
    }

    pub fn as_str(&self) -> &str {
        &self.0
    }
}

impl std::fmt::Display for RequestHash {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.write_str(&self.0)
    }
}

#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum ReplayStatus {
    Reserved,
    Completed,
    Failed,
}

#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
pub struct ReplayRecord {
    pub schema_version: u8,
    pub key_hash: String,
    pub request_hash: RequestHash,
    pub status: ReplayStatus,
    pub created_unix_millis: u64,
    pub updated_unix_millis: u64,
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub response_text: Option<String>,
    /// Source-bound operation receipt (Feature 020 Slice 4, the
    /// replay-authority fence): the disk bytes the completed operation left
    /// behind, read back at completion time. A stored response may be
    /// replayed ONLY while every target still holds these bytes; a record
    /// without a receipt never replays through the verified lanes. Absent on
    /// v1 records (serde default), which is exactly the fail-closed case.
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub post_image: Option<PostImageReceipt>,
}

/// One target the completed operation left on disk. `path` is the ABSOLUTE
/// path as actually written (edits can be rerouted into a worktree, so the
/// request-relative path is not always where the bytes landed). Absolute
/// paths make the record machine-local; the record's whole job is a
/// retry-window guard on this machine's project state, so that is the
/// correct scope. `content_digest: None` records the path as ABSENT (a
/// delete or rename-away), which is verified as absence, not skipped.
#[derive(Clone, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
pub struct PostImageTarget {
    pub path: String,
    pub content_digest: Option<String>,
}

/// The post-image the operation's completion observed: (path → digest) for
/// every file the operation wrote or removed.
#[derive(Clone, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
pub struct PostImageReceipt {
    pub targets: Vec<PostImageTarget>,
}

/// Digest a SINGLE target's post-image from bytes already in hand — the
/// caller's own just-written content — rather than reopening the file from
/// disk. T038 round-1 repair: `capture_post_image`'s disk re-read ran after
/// the write's permit was released (reindex, hooks, and formatting all run
/// between write and the original capture point), an unfenced window in
/// which a concurrent writer's bytes could be digested and bound to THIS
/// response's receipt. The single-target edit tools hold the bytes they
/// wrote for the rest of the call; using them here makes the receipt
/// describe exactly what THIS operation committed, with no reopen and no
/// window at all.
pub fn post_image_from_written_bytes(path: &Path, bytes: &[u8]) -> PostImageReceipt {
    PostImageReceipt {
        targets: vec![PostImageTarget {
            path: path.display().to_string(),
            content_digest: Some(crate::hash::digest_hex(bytes)),
        }],
    }
}

/// Read back the current bytes of the written paths and digest them into a
/// receipt. A missing file records absence; any OTHER read error returns
/// `None` — capture failure means NO receipt, and a record without a receipt
/// never replays (fail closed rather than binding bytes nobody read).
///
/// For a single target whose bytes are already known, prefer
/// [`post_image_from_written_bytes`] — this disk re-read exists for the
/// batch executors, which return only written PATHS to `edit_tools.rs`, not
/// their per-file content.
pub fn capture_post_image(written: &[PathBuf]) -> Option<PostImageReceipt> {
    let mut targets = Vec::with_capacity(written.len());
    for path in written {
        let content_digest = match fs::read(path) {
            Ok(bytes) => Some(crate::hash::digest_hex(&bytes)),
            Err(error) if error.kind() == io::ErrorKind::NotFound => None,
            Err(_) => return None,
        };
        targets.push(PostImageTarget {
            path: path.display().to_string(),
            content_digest,
        });
    }
    Some(PostImageReceipt { targets })
}

/// True only when every receipt target matches the CURRENT disk state:
/// present targets byte-hash-equal, absent targets still absent. An empty
/// receipt verifies trivially — callers must capture every written path.
pub fn verify_post_image(receipt: &PostImageReceipt) -> bool {
    receipt
        .targets
        .iter()
        .all(|target| match fs::read(Path::new(&target.path)) {
            Ok(bytes) => target
                .content_digest
                .as_deref()
                .is_some_and(|digest| digest == crate::hash::digest_hex(&bytes)),
            Err(error) if error.kind() == io::ErrorKind::NotFound => {
                target.content_digest.is_none()
            }
            Err(_) => false,
        })
}

impl ReplayRecord {
    fn reserved(key_hash: String, request_hash: RequestHash) -> Self {
        let now = unix_millis();
        Self {
            schema_version: REPLAY_RECORD_SCHEMA_VERSION,
            key_hash,
            request_hash,
            status: ReplayStatus::Reserved,
            created_unix_millis: now,
            updated_unix_millis: now,
            response_text: None,
            post_image: None,
        }
    }

    fn with_status_and_response(
        mut self,
        status: ReplayStatus,
        response_text: Option<String>,
    ) -> Self {
        self.status = status;
        self.updated_unix_millis = unix_millis();
        self.response_text = response_text;
        self
    }
}

#[derive(Debug, Clone, PartialEq, Eq)]
pub enum ReplayDecision {
    FirstExecution(ReplayRecord),
    Replay(ReplayRecord),
}

#[derive(Debug, Clone)]
pub struct ActiveReplay {
    store: FileReplayStore,
    key: IdempotencyKey,
    request_hash: RequestHash,
}

impl ActiveReplay {
    pub fn complete(
        &self,
        response_text: impl Into<String>,
    ) -> Result<ReplayRecord, IdempotencyError> {
        self.store.update_status_with_response(
            &self.key,
            &self.request_hash,
            ReplayStatus::Completed,
            Some(response_text.into()),
        )
    }

    /// Complete with the source-bound receipt the verified replay lanes
    /// require. A `None` receipt stores a record that will never replay
    /// through those lanes (capture failed — fail closed).
    pub fn complete_with_post_image(
        &self,
        response_text: impl Into<String>,
        post_image: Option<PostImageReceipt>,
    ) -> Result<ReplayRecord, IdempotencyError> {
        let record = self.store.update_status_with_response(
            &self.key,
            &self.request_hash,
            ReplayStatus::Completed,
            Some(response_text.into()),
        )?;
        let mut record = record;
        record.post_image = post_image;
        self.store.write_record_atomic(&record)?;
        Ok(record)
    }

    pub fn fail(&self, response_text: impl Into<String>) -> Result<ReplayRecord, IdempotencyError> {
        self.store.update_status_with_response(
            &self.key,
            &self.request_hash,
            ReplayStatus::Failed,
            Some(response_text.into()),
        )
    }
}

#[derive(Debug, Clone)]
pub enum ReplayStart {
    FirstExecution(ActiveReplay),
    Replay(String),
}

#[derive(Debug, thiserror::Error)]
pub enum IdempotencyError {
    #[error("idempotency key cannot be empty")]
    EmptyKey,
    #[error("tool name cannot be empty for idempotency request hashing")]
    EmptyToolName,
    #[error(
        "idempotency conflict for key hash {key_hash}: existing request {existing}, incoming request {incoming}"
    )]
    Conflict {
        key_hash: String,
        existing: RequestHash,
        incoming: RequestHash,
    },
    #[error("idempotency reservation for key hash {key_hash} is incomplete at {path}")]
    IncompleteReservation { key_hash: String, path: PathBuf },
    #[error(
        "idempotency record at {path} is corrupt and was quarantined at {quarantine_path}: {reason}"
    )]
    CorruptRecordQuarantined {
        path: PathBuf,
        quarantine_path: PathBuf,
        reason: String,
    },
    #[error("idempotency record at {path} is corrupt and could not be quarantined: {reason}")]
    CorruptRecord {
        path: PathBuf,
        reason: String,
        quarantine_error: String,
    },
    #[error("idempotency I/O error: {0}")]
    Io(#[from] io::Error),
    #[error("idempotency JSON error: {0}")]
    Json(#[from] serde_json::Error),
}

/// Age past which a supersede marker is an orphan (its owner crashed between
/// two adjacent fs writes) and may be reclaimed. Generous against clock skew;
/// a healthy claim lives for microseconds.
const SUPERSEDE_MARKER_STALE: std::time::Duration = std::time::Duration::from_secs(60);

#[derive(Debug, Clone)]
pub struct FileReplayStore {
    records_dir: PathBuf,
    quarantine_dir: PathBuf,
}

impl FileReplayStore {
    pub fn open(project_state: &ProjectStateDir) -> Result<Self, IdempotencyError> {
        Self::open_in(paths::project_state_path(
            project_state,
            paths::IDEMPOTENCY_DIR_NAME,
        ))
    }

    pub fn open_control(control_state: &ControlStateDir) -> Result<Self, IdempotencyError> {
        Self::open_in(paths::control_state_path(
            control_state,
            paths::IDEMPOTENCY_DIR_NAME,
        ))
    }

    fn open_in(idempotency_dir: PathBuf) -> Result<Self, IdempotencyError> {
        fs::create_dir_all(&idempotency_dir)?;
        let records_dir = idempotency_dir.join("records");
        let quarantine_dir = idempotency_dir.join("quarantine");
        fs::create_dir_all(&records_dir)?;
        fs::create_dir_all(&quarantine_dir)?;
        Ok(Self {
            records_dir,
            quarantine_dir,
        })
    }

    pub fn check_or_reserve(
        &self,
        key: &IdempotencyKey,
        request_hash: &RequestHash,
    ) -> Result<ReplayDecision, IdempotencyError> {
        let key_hash = key.key_hash();
        let key_dir = self.key_dir_for_hash(&key_hash);

        match fs::create_dir(&key_dir) {
            Ok(()) => {
                let record = ReplayRecord::reserved(key_hash, request_hash.clone());
                self.write_record_atomic(&record)?;
                Ok(ReplayDecision::FirstExecution(record))
            }
            Err(error) if error.kind() == io::ErrorKind::AlreadyExists => {
                let record = self.load_existing(&key_hash)?;
                self.ensure_same_hash(&record, request_hash)?;
                Ok(ReplayDecision::Replay(record))
            }
            Err(error) if error.kind() == io::ErrorKind::NotFound => {
                fs::create_dir_all(&self.records_dir)?;
                self.check_or_reserve(key, request_hash)
            }
            Err(error) => Err(IdempotencyError::Io(error)),
        }
    }

    pub fn replay_if_present(
        &self,
        key: &IdempotencyKey,
        request_hash: &RequestHash,
    ) -> Result<Option<ReplayRecord>, IdempotencyError> {
        let key_hash = key.key_hash();
        let key_dir = self.key_dir_for_hash(&key_hash);
        if !key_dir.exists() {
            return Ok(None);
        }

        let record = self.load_existing(&key_hash)?;
        self.ensure_same_hash(&record, request_hash)?;
        Ok(Some(record))
    }

    pub fn update_status(
        &self,
        key: &IdempotencyKey,
        request_hash: &RequestHash,
        status: ReplayStatus,
    ) -> Result<ReplayRecord, IdempotencyError> {
        self.update_status_with_response(key, request_hash, status, None)
    }

    pub fn update_status_with_response(
        &self,
        key: &IdempotencyKey,
        request_hash: &RequestHash,
        status: ReplayStatus,
        response_text: Option<String>,
    ) -> Result<ReplayRecord, IdempotencyError> {
        let key_hash = key.key_hash();
        let record = self.load_existing(&key_hash)?;
        self.ensure_same_hash(&record, request_hash)?;
        let updated = record.with_status_and_response(status, response_text);
        self.write_record_atomic(&updated)?;
        Ok(updated)
    }

    pub fn record_path(&self, key: &IdempotencyKey) -> PathBuf {
        self.record_path_for_hash(&key.key_hash())
    }

    /// One-winner supersede claim (T038 round-1; heal narrowed in round-2):
    /// `create_new` is the same atomic first-claim primitive
    /// `check_or_reserve` uses, so of N concurrent contenders superseding
    /// the same unverified record, one retakes its reservation and the
    /// losers answer as reserved instead of double-executing the mutation.
    ///
    /// A marker orphaned by a crash between claim and release (two adjacent
    /// fs writes) heals by age — but healing NEVER claims in the same call:
    /// round-2 review showed delete-then-claim lets a second healer delete
    /// the first healer's FRESH marker by name and mint two winners. The
    /// healer removes the orphan, answers "not claimed" (one extra reserved
    /// response after a crash), and the NEXT retry claims the clean slot
    /// through the ordinary `create_new` path.
    ///
    /// Recorded residual, stated exactly: a contender whose staleness
    /// judgment predates another healer's removal can still delete a fresh
    /// marker created in between. That interleave needs a crash-orphan plus
    /// three parties racing inside the winner's two-fs-write claim window
    /// (microseconds); its degradation equals the pre-claim behavior, and
    /// no name-based marker scheme can close it without an
    /// identity-compare-and-delete primitive the filesystem does not offer.
    pub fn try_claim_supersede(&self, key_hash: &str) -> Result<bool, IdempotencyError> {
        let marker = self.supersede_marker_path(key_hash);
        match fs::OpenOptions::new()
            .write(true)
            .create_new(true)
            .open(&marker)
        {
            Ok(_) => Ok(true),
            Err(error) if error.kind() == io::ErrorKind::AlreadyExists => {
                let stale = fs::metadata(&marker)
                    .and_then(|meta| meta.modified())
                    .ok()
                    .and_then(|modified| modified.elapsed().ok())
                    .is_some_and(|age| age > SUPERSEDE_MARKER_STALE);
                if stale {
                    let _ = fs::remove_file(&marker);
                }
                Ok(false)
            }
            Err(error) => Err(IdempotencyError::Io(error)),
        }
    }

    /// Release a claim taken by [`Self::try_claim_supersede`]. Best-effort:
    /// an unremovable marker degrades to the age-healing path.
    pub fn release_supersede(&self, key_hash: &str) {
        let _ = fs::remove_file(self.supersede_marker_path(key_hash));
    }

    fn supersede_marker_path(&self, key_hash: &str) -> PathBuf {
        self.records_dir.join(format!("{key_hash}.superseding"))
    }

    fn ensure_same_hash(
        &self,
        record: &ReplayRecord,
        incoming: &RequestHash,
    ) -> Result<(), IdempotencyError> {
        if record.request_hash == *incoming {
            return Ok(());
        }
        Err(IdempotencyError::Conflict {
            key_hash: record.key_hash.clone(),
            existing: record.request_hash.clone(),
            incoming: incoming.clone(),
        })
    }

    fn load_existing(&self, key_hash: &str) -> Result<ReplayRecord, IdempotencyError> {
        let path = self.record_path_for_hash(key_hash);
        let bytes = match fs::read(&path) {
            Ok(bytes) => bytes,
            Err(error) if error.kind() == io::ErrorKind::NotFound => {
                return Err(IdempotencyError::IncompleteReservation {
                    key_hash: key_hash.to_string(),
                    path,
                });
            }
            Err(error) => return Err(IdempotencyError::Io(error)),
        };

        let record: ReplayRecord = match serde_json::from_slice(&bytes) {
            Ok(record) => record,
            Err(error) => return Err(self.quarantine_record(key_hash, &path, error.to_string())),
        };

        if record.schema_version != REPLAY_RECORD_SCHEMA_VERSION {
            return Err(self.quarantine_record(
                key_hash,
                &path,
                format!("unsupported schema version {}", record.schema_version),
            ));
        }
        if record.key_hash != key_hash {
            return Err(self.quarantine_record(
                key_hash,
                &path,
                format!(
                    "record key hash {} does not match path key hash {}",
                    record.key_hash, key_hash
                ),
            ));
        }

        Ok(record)
    }

    fn write_record_atomic(&self, record: &ReplayRecord) -> Result<(), IdempotencyError> {
        let path = self.record_path_for_hash(&record.key_hash);
        let parent = path.parent().ok_or_else(|| {
            io::Error::new(
                io::ErrorKind::InvalidInput,
                format!("record path has no parent: {}", path.display()),
            )
        })?;
        fs::create_dir_all(parent)?;

        let bytes = serde_json::to_vec_pretty(record)?;
        let mut tmp = tempfile::NamedTempFile::new_in(parent)?;
        tmp.write_all(&bytes)?;
        tmp.flush()?;
        tmp.as_file().sync_all()?;
        tmp.persist(&path)
            .map_err(|error| IdempotencyError::Io(error.error))?;
        Ok(())
    }

    fn quarantine_record(&self, key_hash: &str, path: &Path, source: String) -> IdempotencyError {
        let quarantine_path = self.next_quarantine_path(key_hash);
        if let Err(error) = fs::create_dir_all(&self.quarantine_dir) {
            return IdempotencyError::CorruptRecord {
                path: path.to_path_buf(),
                reason: source,
                quarantine_error: error.to_string(),
            };
        }
        match fs::rename(path, &quarantine_path) {
            Ok(()) => IdempotencyError::CorruptRecordQuarantined {
                path: path.to_path_buf(),
                quarantine_path,
                reason: source,
            },
            Err(error) => IdempotencyError::CorruptRecord {
                path: path.to_path_buf(),
                reason: source,
                quarantine_error: error.to_string(),
            },
        }
    }

    fn next_quarantine_path(&self, key_hash: &str) -> PathBuf {
        let stamp = unix_millis();
        for attempt in 0..100 {
            let suffix = if attempt == 0 {
                String::new()
            } else {
                format!("-{attempt}")
            };
            let path = self
                .quarantine_dir
                .join(format!("{key_hash}-{stamp}{suffix}.json"));
            if !path.exists() {
                return path;
            }
        }
        self.quarantine_dir
            .join(format!("{key_hash}-{stamp}-overflow.json"))
    }

    fn key_dir_for_hash(&self, key_hash: &str) -> PathBuf {
        self.records_dir.join(key_hash)
    }

    fn record_path_for_hash(&self, key_hash: &str) -> PathBuf {
        self.key_dir_for_hash(key_hash).join(RECORD_FILE_NAME)
    }
}

pub fn begin_index_folder_replay(
    control_state: &ControlStateDir,
    canonical_request_root: &Path,
    raw_key: &str,
    reset_requested: bool,
    allow_protected_root: bool,
    activate: bool,
) -> Result<ReplayStart, IdempotencyError> {
    let key = IdempotencyKey::new(raw_key)?;
    let request_hash = index_folder_request_hash(
        canonical_request_root,
        reset_requested,
        allow_protected_root,
        activate,
    )?;

    let store = FileReplayStore::open_control(control_state)?;

    match store.check_or_reserve(&key, &request_hash)? {
        ReplayDecision::FirstExecution(_) => Ok(ReplayStart::FirstExecution(ActiveReplay {
            store,
            key,
            request_hash,
        })),
        ReplayDecision::Replay(record) => Ok(ReplayStart::Replay(replay_response(&record))),
    }
}

pub fn begin_tool_replay(
    project_state: &ProjectStateDir,
    tool_name: &str,
    raw_key: &str,
    request: &Value,
) -> Result<ReplayStart, IdempotencyError> {
    let key = IdempotencyKey::new(raw_key)?;
    let request_hash = RequestHash::for_tool_request(tool_name, request)?;
    let store = FileReplayStore::open(project_state)?;

    match store.check_or_reserve(&key, &request_hash)? {
        ReplayDecision::FirstExecution(_) => Ok(ReplayStart::FirstExecution(ActiveReplay {
            store,
            key,
            request_hash,
        })),
        ReplayDecision::Replay(record) => Ok(ReplayStart::Replay(replay_response(&record))),
    }
}

/// NON-RESERVING, read-only replay probe for a tool request.
///
/// Unlike [`begin_tool_replay`], this NEVER reserves a record and NEVER mutates
/// store state. It answers a single question: "does an identical
/// key+request already have a stored result?"
///
/// - `Ok(Some(response))` — an existing record for this key whose request hash
///   matches; returns the replay response text. The caller may short-circuit
///   and return it without writing any bytes.
/// - `Ok(None)` — no record exists for this key. The caller must fall through
///   to its normal execution path (which reserves via `begin_tool_replay`).
/// - `Err(Conflict)` — a record exists for this key but the incoming request
///   hash differs; the caller must surface the conflict, NOT replay.
///
/// This is the read-only sibling of `check_or_reserve`'s `FirstExecution`-vs
/// `Replay` decision: it observes the `Replay`/`Conflict` outcomes without ever
/// consuming a reservation, so a later `begin_tool_replay` on the miss path
/// still sees a clean slate.
pub fn probe_tool_replay(
    project_state: &ProjectStateDir,
    tool_name: &str,
    raw_key: &str,
    request: &Value,
) -> Result<Option<String>, IdempotencyError> {
    let key = IdempotencyKey::new(raw_key)?;
    let request_hash = RequestHash::for_tool_request(tool_name, request)?;
    let store = FileReplayStore::open(project_state)?;

    match store.replay_if_present(&key, &request_hash)? {
        Some(record) => Ok(Some(replay_response(&record))),
        None => Ok(None),
    }
}

/// Test-only interleave hook: fires between the supersede claim and the
/// double-checked re-read in [`begin_tool_replay_verified`], so an oracle can
/// deterministically stand in for a concurrent contender that already
/// superseded (the round-2 Critical's exact window). Same pattern as
/// `edit.rs`'s `write_interleave`.
#[cfg(test)]
pub(crate) mod supersede_interleave {
    use std::cell::Cell;
    thread_local! {
        static HOOK: Cell<Option<fn()>> = const { Cell::new(None) };
    }
    pub fn install(hook: fn()) {
        HOOK.with(|cell| cell.set(Some(hook)));
    }
    pub fn clear() {
        HOOK.with(|cell| cell.set(None));
    }
    pub(super) fn fire() {
        if let Some(hook) = HOOK.with(|cell| cell.get()) {
            hook();
        }
    }
}

/// [`begin_tool_replay`] with the replay-authority fence (Feature 020
/// Slice 4): a stored result is replayed ONLY when its source-bound
/// post-image receipt verifies against the CURRENT bytes at the recorded
/// written paths. A completed/failed record whose receipt is missing or no
/// longer true is SUPERSEDED — the reservation is retaken and the caller
/// executes fresh, so the record ends up holding the current truth instead
/// of a claim the disk no longer supports. In-flight reservations keep their
/// existing replay-unavailable answer; superseding a live reservation would
/// race the owner.
pub fn begin_tool_replay_verified(
    project_state: &ProjectStateDir,
    tool_name: &str,
    raw_key: &str,
    request: &Value,
) -> Result<ReplayStart, IdempotencyError> {
    let key = IdempotencyKey::new(raw_key)?;
    let request_hash = RequestHash::for_tool_request(tool_name, request)?;
    let store = FileReplayStore::open(project_state)?;

    match store.check_or_reserve(&key, &request_hash)? {
        ReplayDecision::FirstExecution(_) => Ok(ReplayStart::FirstExecution(ActiveReplay {
            store,
            key,
            request_hash,
        })),
        ReplayDecision::Replay(record) => {
            let verified = record.post_image.as_ref().is_some_and(verify_post_image);
            if verified || record.status == ReplayStatus::Reserved {
                return Ok(ReplayStart::Replay(replay_response(&record)));
            }
            // T038 round-1 repair: superseding must have ONE winner. Without
            // the claim, two concurrent identical-key retries could both
            // retake the reservation and both execute the mutation.
            if !store.try_claim_supersede(&key.key_hash())? {
                return Ok(ReplayStart::Replay(replay_response(
                    &ReplayRecord::reserved(key.key_hash(), request_hash.clone()),
                )));
            }
            #[cfg(test)]
            supersede_interleave::fire();
            // T038 round-2 repair (Critical): the read and verify above ran
            // UNFENCED — a contender may have claimed, superseded, and
            // released between our read and our claim, making the pre-claim
            // decision stale. Re-read and RE-DECIDE under the marker
            // (double-checked locking): only a record that is STILL the
            // unverified completed/failed one may be superseded. A record
            // now Reserved answers as reserved; a record whose fresh receipt
            // now verifies replays the fresh truth. Recorded residual: the
            // crash-orphan heal path can, in a multi-party interleave, strip
            // a live marker — the double check bounds even that to two
            // re-reads landing inside one contender's re-read-to-write
            // window (microseconds), versus the pre-fix exposure of the
            // whole verify window.
            let decision = match store.replay_if_present(&key, &request_hash) {
                Ok(Some(current)) => {
                    let current_verified =
                        current.post_image.as_ref().is_some_and(verify_post_image);
                    if current_verified || current.status == ReplayStatus::Reserved {
                        Some(replay_response(&current))
                    } else {
                        None
                    }
                }
                // Record gone under the claim (external tampering): the
                // conservative answer is the transient reserved response —
                // never a blind re-execution off a vanished decision base.
                Ok(None) => Some(replay_response(&ReplayRecord::reserved(
                    key.key_hash(),
                    request_hash.clone(),
                ))),
                Err(error) => {
                    store.release_supersede(&key.key_hash());
                    return Err(error);
                }
            };
            if let Some(response) = decision {
                store.release_supersede(&key.key_hash());
                return Ok(ReplayStart::Replay(response));
            }
            let superseding = ReplayRecord::reserved(key.key_hash(), request_hash.clone());
            let written = store.write_record_atomic(&superseding);
            store.release_supersede(&key.key_hash());
            written?;
            Ok(ReplayStart::FirstExecution(ActiveReplay {
                store,
                key,
                request_hash,
            }))
        }
    }
}

/// [`probe_tool_replay`] with the replay-authority fence: non-reserving, so
/// an unverified record simply answers `None` and the caller falls through
/// to its normal (reserving) execution path, which supersedes it there.
pub fn probe_tool_replay_verified(
    project_state: &ProjectStateDir,
    tool_name: &str,
    raw_key: &str,
    request: &Value,
) -> Result<Option<String>, IdempotencyError> {
    let key = IdempotencyKey::new(raw_key)?;
    let request_hash = RequestHash::for_tool_request(tool_name, request)?;
    let store = FileReplayStore::open(project_state)?;

    match store.replay_if_present(&key, &request_hash)? {
        Some(record) => {
            let verified = record.post_image.as_ref().is_some_and(verify_post_image);
            Ok(verified.then(|| replay_response(&record)))
        }
        None => Ok(None),
    }
}

pub fn index_folder_request_hash(
    canonical_root: &Path,
    reset_requested: bool,
    allow_protected_root: bool,
    activate: bool,
) -> Result<RequestHash, IdempotencyError> {
    RequestHash::for_tool_request(
        "index_folder",
        &json!({
            // The project key hashes the canonical native path bytes. Never
            // use a lossy UTF-8 rendering here: two distinct Unix roots must
            // not share one replay identity.
            "project_id": canonical_project_key(canonical_root),
            "reset": reset_requested,
            "allow_protected_root": allow_protected_root,
            // The `add` spelling changes observable side effects (per-session
            // activation), so it MUST distinguish the canonical request:
            // replaying an `add:true` record for a default call would skip
            // activation, and vice versa.
            "activate": activate,
        }),
    )
}

fn canonical_project_key(canonical_root: &Path) -> String {
    crate::discovery::project_id_for_canonical_root(canonical_root).0
}

pub fn replay_response(record: &ReplayRecord) -> String {
    match (record.status, record.response_text.as_ref()) {
        (ReplayStatus::Completed | ReplayStatus::Failed, Some(response_text)) => {
            response_text.clone()
        }
        (ReplayStatus::Reserved, _) => format!(
            "Idempotency replay unavailable: request for key hash {} is still reserved.",
            record.key_hash
        ),
        (status, None) => format!(
            "Idempotency replay unavailable: record for key hash {} has status {:?} but no stored response.",
            record.key_hash, status
        ),
    }
}

pub fn format_tool_error(error: &IdempotencyError) -> String {
    match error {
        IdempotencyError::Conflict { .. } => format!("Idempotency conflict: {error}"),
        _ => format!("Idempotency error: {error}"),
    }
}

pub fn format_live_postcondition_unavailable(
    historical_receipt: &str,
    error: impl std::fmt::Display,
) -> String {
    format!(
        "applied=false outcome=live_postcondition_unavailable\n\
         historical_receipt_begin\n{historical_receipt}\n\
         historical_receipt_end\n\
         live_postcondition_error={error}"
    )
}

fn canonical_json_bytes(value: &Value) -> Result<Vec<u8>, serde_json::Error> {
    serde_json::to_vec(&canonicalize_value(value))
}

fn canonicalize_value(value: &Value) -> Value {
    match value {
        Value::Array(items) => Value::Array(items.iter().map(canonicalize_value).collect()),
        Value::Object(map) => {
            let mut keys: Vec<&String> = map.keys().collect();
            keys.sort_unstable();
            let mut canonical = Map::new();
            for key in keys {
                if let Some(value) = map.get(key) {
                    canonical.insert(key.clone(), canonicalize_value(value));
                }
            }
            Value::Object(canonical)
        }
        other => other.clone(),
    }
}

fn unix_millis() -> u64 {
    SystemTime::now()
        .duration_since(UNIX_EPOCH)
        .unwrap_or_default()
        .as_millis()
        .min(u64::MAX as u128) as u64
}

#[cfg(test)]
mod tests {
    use super::*;

    /// T038 round-1 (replay supersede atomicity): the supersede claim has
    /// exactly one winner, releases cleanly, and heals a crash-orphaned
    /// marker by age. A deterministic RED for the underlying race is not
    /// constructible; this pins the primitive the race fix rests on.
    #[test]
    fn supersede_claim_is_one_winner_releases_and_heals_orphans() {
        let dir = tempfile::TempDir::new().expect("tempdir");
        let store = FileReplayStore::open_in(dir.path().join("idempotency")).expect("store");

        assert!(store.try_claim_supersede("k1").expect("first claim"));
        assert!(
            !store.try_claim_supersede("k1").expect("second claim"),
            "a live claim must have exactly one winner"
        );
        store.release_supersede("k1");
        assert!(
            store.try_claim_supersede("k1").expect("post-release claim"),
            "a released claim is reclaimable"
        );

        // Crash-orphan healing: backdate the marker past the staleness bound.
        // Healing must NOT claim in the same call (round-2: delete-then-claim
        // re-opens the two-winner race) — it removes the orphan and answers
        // "not claimed"; the NEXT claim owns the clean slot.
        let marker = store.supersede_marker_path("k1");
        std::fs::File::options()
            .write(true)
            .open(&marker)
            .expect("open marker")
            .set_times(
                std::fs::FileTimes::new()
                    .set_modified(std::time::SystemTime::now() - (SUPERSEDE_MARKER_STALE * 2)),
            )
            .expect("backdate marker");
        assert!(
            !store.try_claim_supersede("k1").expect("stale heal"),
            "healing an orphaned marker must not claim in the same call"
        );
        assert!(
            !marker.exists(),
            "the orphaned marker must be removed by the heal"
        );
        assert!(
            store.try_claim_supersede("k1").expect("post-heal claim"),
            "the claim after the heal owns the clean slot"
        );
    }

    /// T038 round-2 (Critical repair, deterministic interleave): the
    /// pre-claim read and verify run UNFENCED, so a contender can claim,
    /// supersede, and release between our read and our claim. The hook
    /// stands in for that contender by rewriting the record to Reserved
    /// right after we win the claim; `begin_tool_replay_verified` must
    /// RE-DECIDE under the marker and answer as reserved — never
    /// double-execute off the stale pre-claim decision.
    #[test]
    fn a_record_superseded_between_read_and_claim_is_not_double_executed() {
        use std::sync::OnceLock;
        static STATE: OnceLock<(ProjectStateDir, String)> = OnceLock::new();

        let dir = tempfile::TempDir::new().expect("tempdir");
        let project_state = ProjectStateDir::new(
            dunce::canonicalize(dir.path())
                .expect("canonical tempdir")
                .join("state"),
        );
        std::fs::create_dir_all(project_state.as_path()).expect("state dir");
        let request = json!({ "path": "src/x.rs" });

        // Seed an UNVERIFIED completed record (v1 shape: no post-image).
        let ReplayStart::FirstExecution(active) =
            begin_tool_replay_verified(&project_state, "t", "race-key", &request)
                .expect("first begin")
        else {
            panic!("fresh key must be a first execution");
        };
        active
            .complete_with_post_image("first result".to_string(), None)
            .expect("complete without receipt");

        let key = IdempotencyKey::new("race-key").expect("key");
        STATE
            .set((project_state.clone(), key.key_hash()))
            .expect("state slot");

        // The stand-in contender: after our claim, the record becomes
        // Reserved (as a real winner's superseding write would make it).
        fn contender() {
            let (project_state, key_hash) = STATE.get().expect("state");
            let store = FileReplayStore::open(project_state).expect("store");
            let request_hash =
                RequestHash::for_tool_request("t", &json!({ "path": "src/x.rs" })).expect("hash");
            store
                .write_record_atomic(&ReplayRecord::reserved(key_hash.clone(), request_hash))
                .expect("contender write");
        }
        supersede_interleave::install(contender);
        let outcome = begin_tool_replay_verified(&project_state, "t", "race-key", &request);
        supersede_interleave::clear();

        match outcome.expect("begin under interleave") {
            ReplayStart::Replay(response) => assert!(
                response.contains("still reserved"),
                "the re-decision under the claim must answer as reserved: {response}"
            ),
            ReplayStart::FirstExecution(_) => panic!(
                "double execution: the stale pre-claim decision was acted on \
                 although the record was superseded before the claim"
            ),
        }
    }

    #[test]
    fn index_folder_identity_uses_native_project_identity() {
        let literal = Path::new("/work/a\\b");
        let nested = Path::new("/work/a/b");

        if cfg!(windows) {
            assert_eq!(
                canonical_project_key(literal),
                canonical_project_key(nested),
                "Windows separator compatibility must remain intact"
            );
        } else {
            assert_ne!(
                canonical_project_key(literal),
                canonical_project_key(nested),
                "distinct Unix roots must produce distinct idempotency request identities"
            );
        }
    }

    #[cfg(unix)]
    #[test]
    fn index_folder_identity_preserves_non_utf8_native_bytes() {
        use std::os::unix::ffi::OsStringExt;

        let native = PathBuf::from(std::ffi::OsString::from_vec(vec![b'a', 0xff, b'b']));
        let lossy_collision = PathBuf::from("a\u{fffd}b");
        assert_eq!(native.to_string_lossy(), lossy_collision.to_string_lossy());
        assert_ne!(
            canonical_project_key(&native),
            canonical_project_key(&lossy_collision)
        );
    }
}