mustflow 2.11.0 → 2.16.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -33,6 +33,18 @@
33
33
  "status": { "enum": ["passed", "partial", "failed", "blocked"] },
34
34
  "completion_verdict": { "$ref": "#/$defs/completionVerdict" },
35
35
  "evidence_model": { "$ref": "#/$defs/evidenceModel" },
36
+ "failure_fingerprint": {
37
+ "anyOf": [
38
+ { "$ref": "#/$defs/failureFingerprint" },
39
+ { "type": "null" }
40
+ ]
41
+ },
42
+ "repeated_failure_summary": {
43
+ "anyOf": [
44
+ { "$ref": "#/$defs/repeatedFailureSummary" },
45
+ { "type": "null" }
46
+ ]
47
+ },
36
48
  "summary": { "$ref": "#/$defs/summary" },
37
49
  "repro_evidence": { "$ref": "#/$defs/reproEvidence" },
38
50
  "external_checks": {
@@ -45,6 +57,66 @@
45
57
  }
46
58
  },
47
59
  "$defs": {
60
+ "failureFingerprint": {
61
+ "type": "object",
62
+ "additionalProperties": false,
63
+ "required": [
64
+ "schema_version",
65
+ "fingerprint",
66
+ "verification_plan_id",
67
+ "failed_intents_hash",
68
+ "exit_code_classes_hash",
69
+ "timeout_flags_hash",
70
+ "error_kinds_hash",
71
+ "diagnostic_hash",
72
+ "risk_codes_hash",
73
+ "affected_surfaces_hash",
74
+ "command_fingerprints_hash"
75
+ ],
76
+ "properties": {
77
+ "schema_version": { "const": "1" },
78
+ "fingerprint": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
79
+ "verification_plan_id": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
80
+ "failed_intents_hash": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
81
+ "exit_code_classes_hash": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
82
+ "timeout_flags_hash": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
83
+ "error_kinds_hash": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
84
+ "diagnostic_hash": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
85
+ "risk_codes_hash": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
86
+ "affected_surfaces_hash": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
87
+ "command_fingerprints_hash": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" }
88
+ }
89
+ },
90
+ "repeatedFailureSummary": {
91
+ "type": "object",
92
+ "additionalProperties": false,
93
+ "required": [
94
+ "schema_version",
95
+ "fingerprint",
96
+ "verification_plan_id",
97
+ "status",
98
+ "failed_intents_hash",
99
+ "risk_codes_hash",
100
+ "affected_surfaces_hash",
101
+ "first_seen_at",
102
+ "last_seen_at",
103
+ "seen_count",
104
+ "requires_new_evidence"
105
+ ],
106
+ "properties": {
107
+ "schema_version": { "const": "1" },
108
+ "fingerprint": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
109
+ "verification_plan_id": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
110
+ "status": { "type": "string" },
111
+ "failed_intents_hash": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
112
+ "risk_codes_hash": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
113
+ "affected_surfaces_hash": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" },
114
+ "first_seen_at": { "type": "string" },
115
+ "last_seen_at": { "type": "string" },
116
+ "seen_count": { "type": "integer", "minimum": 1 },
117
+ "requires_new_evidence": { "type": "boolean" }
118
+ }
119
+ },
48
120
  "completionVerdict": {
49
121
  "type": "object",
50
122
  "additionalProperties": false,
@@ -60,6 +132,7 @@
60
132
  "source",
61
133
  "verification_plan_id",
62
134
  "changed_file_count",
135
+ "criteria",
63
136
  "matched_intents",
64
137
  "ran_intents",
65
138
  "passed_intents",
@@ -71,6 +144,14 @@
71
144
  "scope_diff_risk_count",
72
145
  "repeated_failure_count",
73
146
  "validation_ratchet_risk_count",
147
+ "repro_evidence_risk_count",
148
+ "external_evidence_risk_count",
149
+ "write_drift_risk_count",
150
+ "receipt_binding_risk_count",
151
+ "stale_receipt_count",
152
+ "plan_mismatch_count",
153
+ "risks",
154
+ "receipt_binding",
74
155
  "latest_run_status"
75
156
  ],
76
157
  "properties": {
@@ -80,6 +161,7 @@
80
161
  "pattern": "^sha256:[0-9a-f]{64}$"
81
162
  },
82
163
  "changed_file_count": { "type": ["integer", "null"] },
164
+ "criteria": { "$ref": "#/$defs/completionVerdictCriteria" },
83
165
  "matched_intents": { "type": "integer" },
84
166
  "ran_intents": { "type": "integer" },
85
167
  "passed_intents": { "type": "integer" },
@@ -91,6 +173,14 @@
91
173
  "scope_diff_risk_count": { "type": "integer" },
92
174
  "repeated_failure_count": { "type": "integer" },
93
175
  "validation_ratchet_risk_count": { "type": "integer" },
176
+ "repro_evidence_risk_count": { "type": "integer" },
177
+ "external_evidence_risk_count": { "type": "integer" },
178
+ "write_drift_risk_count": { "type": "integer" },
179
+ "receipt_binding_risk_count": { "type": "integer" },
180
+ "stale_receipt_count": { "type": "integer" },
181
+ "plan_mismatch_count": { "type": "integer" },
182
+ "risks": { "$ref": "#/$defs/completionVerdictRisks" },
183
+ "receipt_binding": { "$ref": "#/$defs/completionVerdictReceiptBinding" },
94
184
  "latest_run_status": { "type": ["string", "null"] }
95
185
  }
96
186
  },
@@ -108,6 +198,47 @@
108
198
  }
109
199
  }
110
200
  },
201
+ "completionVerdictCriteria": {
202
+ "type": "object",
203
+ "additionalProperties": false,
204
+ "required": ["total", "covered", "partially_covered", "uncovered", "blocked", "contradicted"],
205
+ "properties": {
206
+ "total": { "type": "integer" },
207
+ "covered": { "type": "integer" },
208
+ "partially_covered": { "type": "integer" },
209
+ "uncovered": { "type": "integer" },
210
+ "blocked": { "type": "integer" },
211
+ "contradicted": { "type": "integer" }
212
+ }
213
+ },
214
+ "completionVerdictRisks": {
215
+ "type": "object",
216
+ "additionalProperties": false,
217
+ "required": [
218
+ "source_anchor",
219
+ "scope_diff",
220
+ "repeated_failure",
221
+ "validation_ratchet",
222
+ "repro_evidence",
223
+ "external_evidence",
224
+ "write_drift",
225
+ "receipt_binding",
226
+ "stale_receipt",
227
+ "plan_mismatch"
228
+ ],
229
+ "properties": {
230
+ "source_anchor": { "type": "integer" },
231
+ "scope_diff": { "type": "integer" },
232
+ "repeated_failure": { "type": "integer" },
233
+ "validation_ratchet": { "type": "integer" },
234
+ "repro_evidence": { "type": "integer" },
235
+ "external_evidence": { "type": "integer" },
236
+ "write_drift": { "type": "integer" },
237
+ "receipt_binding": { "type": "integer" },
238
+ "stale_receipt": { "type": "integer" },
239
+ "plan_mismatch": { "type": "integer" }
240
+ }
241
+ },
111
242
  "summary": {
112
243
  "type": "object",
113
244
  "additionalProperties": false,
@@ -146,6 +277,30 @@
146
277
  }
147
278
  }
148
279
  },
280
+ "completionVerdictReceiptBinding": {
281
+ "type": "object",
282
+ "additionalProperties": false,
283
+ "required": [
284
+ "plan_bound_count",
285
+ "plan_unbound_count",
286
+ "fingerprint_bound_count",
287
+ "fingerprint_unbound_count",
288
+ "current_state_bound_count",
289
+ "current_state_unavailable_count",
290
+ "stale_count",
291
+ "plan_mismatch_count"
292
+ ],
293
+ "properties": {
294
+ "plan_bound_count": { "type": "integer" },
295
+ "plan_unbound_count": { "type": "integer" },
296
+ "fingerprint_bound_count": { "type": "integer" },
297
+ "fingerprint_unbound_count": { "type": "integer" },
298
+ "current_state_bound_count": { "type": "integer" },
299
+ "current_state_unavailable_count": { "type": "integer" },
300
+ "stale_count": { "type": "integer" },
301
+ "plan_mismatch_count": { "type": "integer" }
302
+ }
303
+ },
149
304
  "evidenceModel": {
150
305
  "type": "object",
151
306
  "additionalProperties": false,
@@ -342,9 +497,9 @@
342
497
  "reported_symptom",
343
498
  "expected_behavior",
344
499
  "observed_behavior",
345
- "original_reproduction",
346
- "evidence_before_fix",
347
- "evidence_after_fix",
500
+ "reproduction_route",
501
+ "before_fix",
502
+ "after_fix",
348
503
  "regression_guard"
349
504
  ],
350
505
  "properties": {
@@ -353,18 +508,80 @@
353
508
  "reported_symptom": { "type": ["string", "null"] },
354
509
  "expected_behavior": { "type": ["string", "null"] },
355
510
  "observed_behavior": { "type": ["string", "null"] },
356
- "original_reproduction": { "$ref": "#/$defs/reproEvidenceItem" },
357
- "evidence_before_fix": { "$ref": "#/$defs/reproEvidenceItem" },
358
- "evidence_after_fix": { "$ref": "#/$defs/reproEvidenceItem" },
359
- "regression_guard": { "$ref": "#/$defs/reproEvidenceItem" }
511
+ "reproduction_route": { "$ref": "#/$defs/reproRouteIdentity" },
512
+ "before_fix": { "$ref": "#/$defs/reproBeforeFixEvidence" },
513
+ "after_fix": { "$ref": "#/$defs/reproAfterFixEvidence" },
514
+ "regression_guard": { "$ref": "#/$defs/reproRegressionGuardEvidence" }
515
+ }
516
+ },
517
+ "reproRouteIdentity": {
518
+ "type": "object",
519
+ "additionalProperties": false,
520
+ "required": ["route_id", "route_kind", "route_digest", "failure_oracle_hash", "steps"],
521
+ "properties": {
522
+ "route_id": { "type": ["string", "null"] },
523
+ "route_kind": { "enum": ["test", "cli", "browser", "api", "manual", "unknown", null] },
524
+ "route_digest": { "type": ["string", "null"] },
525
+ "failure_oracle_hash": { "type": ["string", "null"] },
526
+ "steps": {
527
+ "type": "array",
528
+ "items": { "$ref": "#/$defs/reproRouteStep" }
529
+ }
530
+ }
531
+ },
532
+ "reproRouteStep": {
533
+ "type": "object",
534
+ "additionalProperties": false,
535
+ "required": ["ordinal", "action", "target", "input_digest", "observation_digest", "summary"],
536
+ "properties": {
537
+ "ordinal": { "type": "integer", "minimum": 1 },
538
+ "action": { "type": ["string", "null"] },
539
+ "target": { "type": ["string", "null"] },
540
+ "input_digest": { "type": ["string", "null"] },
541
+ "observation_digest": { "type": ["string", "null"] },
542
+ "summary": { "type": ["string", "null"] }
360
543
  }
361
544
  },
362
- "reproEvidenceItem": {
545
+ "reproBeforeFixEvidence": {
363
546
  "type": "object",
364
547
  "additionalProperties": false,
365
- "required": ["status", "summary", "reason"],
548
+ "required": ["status", "outcome", "receipt_path", "receipt_sha256", "verification_plan_id", "summary", "reason"],
366
549
  "properties": {
367
- "status": { "enum": ["present", "unavailable", "missing"] },
550
+ "status": { "enum": ["reproduced", "unavailable", "missing"] },
551
+ "outcome": { "enum": ["failed_as_expected", "failed_differently", "passed_unexpectedly", null] },
552
+ "receipt_path": { "type": ["string", "null"] },
553
+ "receipt_sha256": { "type": ["string", "null"] },
554
+ "verification_plan_id": { "type": ["string", "null"] },
555
+ "summary": { "type": ["string", "null"] },
556
+ "reason": { "type": ["string", "null"] }
557
+ }
558
+ },
559
+ "reproAfterFixEvidence": {
560
+ "type": "object",
561
+ "additionalProperties": false,
562
+ "required": ["status", "outcome", "same_route_as", "receipt_path", "receipt_sha256", "verification_plan_id", "summary", "reason"],
563
+ "properties": {
564
+ "status": { "enum": ["passed", "failed", "unavailable", "missing"] },
565
+ "outcome": { "enum": ["passed_expected_behavior", "failed_same_route", "failed_differently", null] },
566
+ "same_route_as": { "type": ["string", "null"] },
567
+ "receipt_path": { "type": ["string", "null"] },
568
+ "receipt_sha256": { "type": ["string", "null"] },
569
+ "verification_plan_id": { "type": ["string", "null"] },
570
+ "summary": { "type": ["string", "null"] },
571
+ "reason": { "type": ["string", "null"] }
572
+ }
573
+ },
574
+ "reproRegressionGuardEvidence": {
575
+ "type": "object",
576
+ "additionalProperties": false,
577
+ "required": ["status", "intent", "test_path", "receipt_path", "receipt_sha256", "verification_plan_id", "summary", "reason"],
578
+ "properties": {
579
+ "status": { "enum": ["passed", "failed", "unavailable", "missing"] },
580
+ "intent": { "type": ["string", "null"] },
581
+ "test_path": { "type": ["string", "null"] },
582
+ "receipt_path": { "type": ["string", "null"] },
583
+ "receipt_sha256": { "type": ["string", "null"] },
584
+ "verification_plan_id": { "type": ["string", "null"] },
368
585
  "summary": { "type": ["string", "null"] },
369
586
  "reason": { "type": ["string", "null"] }
370
587
  }
@@ -1,6 +1,6 @@
1
1
  id = "default"
2
2
  name = "default"
3
- version = "2.11.0"
3
+ version = "2.16.0"
4
4
  description = "Minimal workflow for LLM agents to read, edit, and verify their work in a repository."
5
5
  common_root = "common"
6
6
  locales_root = "locales"