@yanlinglabs/winter-conformance 0.0.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (44) hide show
  1. package/LICENSE +21 -0
  2. package/README.md +128 -0
  3. package/dist/bun-required.d.ts +54 -0
  4. package/dist/goldens.d.ts +11 -0
  5. package/dist/index-v0zvhx0z.js +1576 -0
  6. package/dist/index-z5gwvsfh.js +49 -0
  7. package/dist/index.d.ts +5 -0
  8. package/dist/index.js +49 -0
  9. package/dist/official/capture.d.ts +1 -0
  10. package/dist/official/fetch.d.ts +40 -0
  11. package/dist/official/index.d.ts +3 -0
  12. package/dist/official/index.js +27 -0
  13. package/dist/trace.d.ts +9 -0
  14. package/dist/trace.js +8 -0
  15. package/goldens/advertised-set-round.trace.json +81 -0
  16. package/goldens/background-task-round.trace.json +210 -0
  17. package/goldens/bash-background-round.trace.json +148 -0
  18. package/goldens/canusetool-approved-round.trace.json +119 -0
  19. package/goldens/compaction-auto-round.trace.json +219 -0
  20. package/goldens/compaction-manual-round.trace.json +232 -0
  21. package/goldens/denied-tool-round.trace.json +141 -0
  22. package/goldens/hook-denied-round.trace.json +169 -0
  23. package/goldens/hooked-tool-round.trace.json +147 -0
  24. package/goldens/interrupt.trace.json +129 -0
  25. package/goldens/mcp-tool-round.trace.json +131 -0
  26. package/goldens/messaging-facet-round.trace.json +278 -0
  27. package/goldens/mode-switch-mid-session.trace.json +200 -0
  28. package/goldens/multi-turn.trace.json +110 -0
  29. package/goldens/p6-anthropic-fake.trace.json +147 -0
  30. package/goldens/p6-gemini-fake.trace.json +132 -0
  31. package/goldens/p6-openai-chat-fake.trace.json +133 -0
  32. package/goldens/p6-openai-responses-fake.trace.json +145 -0
  33. package/goldens/p6-resolution-failure.trace.json +68 -0
  34. package/goldens/plain-query.trace.json +82 -0
  35. package/goldens/resume.trace.json +162 -0
  36. package/goldens/sendmessage-child-round.trace.json +158 -0
  37. package/goldens/skill-invocation-round.trace.json +123 -0
  38. package/goldens/structured-exhaustion-round.trace.json +182 -0
  39. package/goldens/structured-output-round.trace.json +106 -0
  40. package/goldens/subagent-permission-round.trace.json +174 -0
  41. package/goldens/subagent-spawn-round.trace.json +120 -0
  42. package/goldens/tool-round.trace.json +119 -0
  43. package/goldens/toolsearch-select-round.trace.json +185 -0
  44. package/package.json +50 -0
@@ -0,0 +1,182 @@
1
+ [
2
+ {
3
+ "sequence": 0,
4
+ "direction": "runtime-to-host",
5
+ "kind": "system/init",
6
+ "payload": {
7
+ "type": "system",
8
+ "subtype": "init",
9
+ "cwd": "/winter-fixture",
10
+ "model": "winter-test/echo",
11
+ "permissionMode": "default",
12
+ "tools": [
13
+ "Agent",
14
+ "AskUserQuestion",
15
+ "Bash",
16
+ "CronCreate",
17
+ "CronDelete",
18
+ "CronList",
19
+ "Edit",
20
+ "EnterPlanMode",
21
+ "EnterWorktree",
22
+ "ExitPlanMode",
23
+ "ExitWorktree",
24
+ "Glob",
25
+ "Grep",
26
+ "ListAgents",
27
+ "Monitor",
28
+ "NotebookEdit",
29
+ "PushNotification",
30
+ "Read",
31
+ "ReadNotifications",
32
+ "ReportFindings",
33
+ "ScheduleWakeup",
34
+ "SendMessage",
35
+ "Skill",
36
+ "StructuredOutput",
37
+ "TaskCreate",
38
+ "TaskGet",
39
+ "TaskList",
40
+ "TaskOutput",
41
+ "TaskStop",
42
+ "TaskUpdate",
43
+ "Workflow",
44
+ "Write"
45
+ ],
46
+ "apiKeySource": "none",
47
+ "slash_commands": [
48
+ "compact"
49
+ ],
50
+ "output_style": "default",
51
+ "skills": [],
52
+ "plugins": []
53
+ }
54
+ },
55
+ {
56
+ "sequence": 1,
57
+ "direction": "runtime-to-host",
58
+ "kind": "assistant",
59
+ "payload": {
60
+ "type": "assistant",
61
+ "message": {
62
+ "content": [
63
+ {
64
+ "type": "tool_use",
65
+ "id": "p5-structured-fail-1",
66
+ "name": "StructuredOutput",
67
+ "input": {
68
+ "answer": "forty-two"
69
+ }
70
+ }
71
+ ]
72
+ }
73
+ }
74
+ },
75
+ {
76
+ "sequence": 2,
77
+ "direction": "runtime-to-host",
78
+ "kind": "user",
79
+ "payload": {
80
+ "type": "user",
81
+ "message": {
82
+ "content": [
83
+ {
84
+ "type": "tool_result",
85
+ "tool_use_id": "p5-structured-fail-1",
86
+ "content": "Structured output validation failed: /answer: must be number. Call StructuredOutput again with a corrected value.",
87
+ "error": true
88
+ }
89
+ ]
90
+ }
91
+ }
92
+ },
93
+ {
94
+ "sequence": 3,
95
+ "direction": "runtime-to-host",
96
+ "kind": "assistant",
97
+ "payload": {
98
+ "type": "assistant",
99
+ "message": {
100
+ "content": [
101
+ {
102
+ "type": "tool_use",
103
+ "id": "p5-structured-fail-2",
104
+ "name": "StructuredOutput",
105
+ "input": {
106
+ "answer": "forty-two"
107
+ }
108
+ }
109
+ ]
110
+ }
111
+ }
112
+ },
113
+ {
114
+ "sequence": 4,
115
+ "direction": "runtime-to-host",
116
+ "kind": "user",
117
+ "payload": {
118
+ "type": "user",
119
+ "message": {
120
+ "content": [
121
+ {
122
+ "type": "tool_result",
123
+ "tool_use_id": "p5-structured-fail-2",
124
+ "content": "Structured output validation failed: /answer: must be number. Call StructuredOutput again with a corrected value.",
125
+ "error": true
126
+ }
127
+ ]
128
+ }
129
+ }
130
+ },
131
+ {
132
+ "sequence": 5,
133
+ "direction": "runtime-to-host",
134
+ "kind": "assistant",
135
+ "payload": {
136
+ "type": "assistant",
137
+ "message": {
138
+ "content": [
139
+ {
140
+ "type": "tool_use",
141
+ "id": "p5-structured-fail-3",
142
+ "name": "StructuredOutput",
143
+ "input": {
144
+ "answer": "forty-two"
145
+ }
146
+ }
147
+ ]
148
+ }
149
+ }
150
+ },
151
+ {
152
+ "sequence": 6,
153
+ "direction": "runtime-to-host",
154
+ "kind": "user",
155
+ "payload": {
156
+ "type": "user",
157
+ "message": {
158
+ "content": [
159
+ {
160
+ "type": "tool_result",
161
+ "tool_use_id": "p5-structured-fail-3",
162
+ "content": "Structured output validation failed: /answer: must be number",
163
+ "error": true
164
+ }
165
+ ]
166
+ }
167
+ }
168
+ },
169
+ {
170
+ "sequence": 7,
171
+ "direction": "runtime-to-host",
172
+ "kind": "result",
173
+ "payload": {
174
+ "type": "result",
175
+ "subtype": "error_max_structured_output_retries",
176
+ "is_error": true,
177
+ "result": "Failed to provide valid structured output after 3 attempts",
178
+ "terminal_reason": "structured_output_retry_exhausted",
179
+ "permission_denials": []
180
+ }
181
+ }
182
+ ]
@@ -0,0 +1,106 @@
1
+ [
2
+ {
3
+ "sequence": 0,
4
+ "direction": "runtime-to-host",
5
+ "kind": "system/init",
6
+ "payload": {
7
+ "type": "system",
8
+ "subtype": "init",
9
+ "cwd": "/winter-fixture",
10
+ "model": "winter-test/echo",
11
+ "permissionMode": "default",
12
+ "tools": [
13
+ "Agent",
14
+ "AskUserQuestion",
15
+ "Bash",
16
+ "CronCreate",
17
+ "CronDelete",
18
+ "CronList",
19
+ "Edit",
20
+ "EnterPlanMode",
21
+ "EnterWorktree",
22
+ "ExitPlanMode",
23
+ "ExitWorktree",
24
+ "Glob",
25
+ "Grep",
26
+ "ListAgents",
27
+ "Monitor",
28
+ "NotebookEdit",
29
+ "PushNotification",
30
+ "Read",
31
+ "ReadNotifications",
32
+ "ReportFindings",
33
+ "ScheduleWakeup",
34
+ "SendMessage",
35
+ "Skill",
36
+ "StructuredOutput",
37
+ "TaskCreate",
38
+ "TaskGet",
39
+ "TaskList",
40
+ "TaskOutput",
41
+ "TaskStop",
42
+ "TaskUpdate",
43
+ "Workflow",
44
+ "Write"
45
+ ],
46
+ "apiKeySource": "none",
47
+ "slash_commands": [
48
+ "compact"
49
+ ],
50
+ "output_style": "default",
51
+ "skills": [],
52
+ "plugins": []
53
+ }
54
+ },
55
+ {
56
+ "sequence": 1,
57
+ "direction": "runtime-to-host",
58
+ "kind": "assistant",
59
+ "payload": {
60
+ "type": "assistant",
61
+ "message": {
62
+ "content": [
63
+ {
64
+ "type": "tool_use",
65
+ "id": "p5-structured-1",
66
+ "name": "StructuredOutput",
67
+ "input": {
68
+ "answer": 42
69
+ }
70
+ }
71
+ ]
72
+ }
73
+ }
74
+ },
75
+ {
76
+ "sequence": 2,
77
+ "direction": "runtime-to-host",
78
+ "kind": "user",
79
+ "payload": {
80
+ "type": "user",
81
+ "message": {
82
+ "content": [
83
+ {
84
+ "type": "tool_result",
85
+ "tool_use_id": "p5-structured-1",
86
+ "content": "Structured output accepted."
87
+ }
88
+ ]
89
+ }
90
+ }
91
+ },
92
+ {
93
+ "sequence": 3,
94
+ "direction": "runtime-to-host",
95
+ "kind": "result",
96
+ "payload": {
97
+ "type": "result",
98
+ "subtype": "success",
99
+ "is_error": false,
100
+ "structured_output": {
101
+ "answer": 42
102
+ },
103
+ "permission_denials": []
104
+ }
105
+ }
106
+ ]
@@ -0,0 +1,174 @@
1
+ [
2
+ {
3
+ "sequence": 0,
4
+ "direction": "runtime-to-host",
5
+ "kind": "system/init",
6
+ "payload": {
7
+ "type": "system",
8
+ "subtype": "init",
9
+ "cwd": "/winter-fixture",
10
+ "model": "winter-test/echo",
11
+ "permissionMode": "default",
12
+ "tools": [
13
+ "Agent",
14
+ "AskUserQuestion",
15
+ "Bash",
16
+ "CronCreate",
17
+ "CronDelete",
18
+ "CronList",
19
+ "Edit",
20
+ "EnterPlanMode",
21
+ "EnterWorktree",
22
+ "ExitPlanMode",
23
+ "ExitWorktree",
24
+ "Glob",
25
+ "Grep",
26
+ "ListAgents",
27
+ "Monitor",
28
+ "NotebookEdit",
29
+ "PushNotification",
30
+ "Read",
31
+ "ReadNotifications",
32
+ "ReportFindings",
33
+ "ScheduleWakeup",
34
+ "SendMessage",
35
+ "Skill",
36
+ "TaskCreate",
37
+ "TaskGet",
38
+ "TaskList",
39
+ "TaskOutput",
40
+ "TaskStop",
41
+ "TaskUpdate",
42
+ "Workflow",
43
+ "Write"
44
+ ],
45
+ "apiKeySource": "none",
46
+ "slash_commands": [
47
+ "compact"
48
+ ],
49
+ "output_style": "default",
50
+ "skills": [],
51
+ "plugins": []
52
+ }
53
+ },
54
+ {
55
+ "sequence": 1,
56
+ "direction": "runtime-to-host",
57
+ "kind": "assistant",
58
+ "payload": {
59
+ "type": "assistant",
60
+ "message": {
61
+ "content": [
62
+ {
63
+ "type": "tool_use",
64
+ "id": "agent-call-1",
65
+ "name": "Agent",
66
+ "input": {
67
+ "description": "permission probe",
68
+ "prompt": "child probe text"
69
+ }
70
+ }
71
+ ]
72
+ }
73
+ }
74
+ },
75
+ {
76
+ "sequence": 2,
77
+ "direction": "runtime-to-host",
78
+ "kind": "assistant",
79
+ "payload": {
80
+ "type": "assistant",
81
+ "message": {
82
+ "content": [
83
+ {
84
+ "type": "tool_use",
85
+ "id": "child-call-1",
86
+ "name": "ReadNotifications",
87
+ "input": {}
88
+ }
89
+ ]
90
+ },
91
+ "parent_tool_use_id": "agent-call-1"
92
+ }
93
+ },
94
+ {
95
+ "sequence": 3,
96
+ "direction": "runtime-to-host",
97
+ "kind": "user",
98
+ "payload": {
99
+ "type": "user",
100
+ "message": {
101
+ "content": [
102
+ {
103
+ "type": "tool_result",
104
+ "tool_use_id": "child-call-1",
105
+ "content": "{\"notifications\":[],\"remaining\":0}"
106
+ }
107
+ ]
108
+ },
109
+ "parent_tool_use_id": "agent-call-1"
110
+ }
111
+ },
112
+ {
113
+ "sequence": 4,
114
+ "direction": "runtime-to-host",
115
+ "kind": "assistant",
116
+ "payload": {
117
+ "type": "assistant",
118
+ "message": {
119
+ "content": [
120
+ {
121
+ "type": "text",
122
+ "text": "child finished after its own tool call"
123
+ }
124
+ ]
125
+ },
126
+ "parent_tool_use_id": "agent-call-1"
127
+ }
128
+ },
129
+ {
130
+ "sequence": 5,
131
+ "direction": "runtime-to-host",
132
+ "kind": "user",
133
+ "payload": {
134
+ "type": "user",
135
+ "message": {
136
+ "content": [
137
+ {
138
+ "type": "tool_result",
139
+ "tool_use_id": "agent-call-1",
140
+ "content": "{\"agentId\":\"<scrubbed>\",\"content\":[{\"type\":\"text\",\"text\":\"child finished after its own tool call\"}],\"totalToolUseCount\":1,\"totalDurationMs\":\"<scrubbed>\",\"resolvedModel\":\"winter-test/echo\",\"prompt\":\"child probe text\"}"
141
+ }
142
+ ]
143
+ }
144
+ }
145
+ },
146
+ {
147
+ "sequence": 6,
148
+ "direction": "runtime-to-host",
149
+ "kind": "assistant",
150
+ "payload": {
151
+ "type": "assistant",
152
+ "message": {
153
+ "content": [
154
+ {
155
+ "type": "text",
156
+ "text": "parent finished"
157
+ }
158
+ ]
159
+ }
160
+ }
161
+ },
162
+ {
163
+ "sequence": 7,
164
+ "direction": "runtime-to-host",
165
+ "kind": "result",
166
+ "payload": {
167
+ "type": "result",
168
+ "subtype": "success",
169
+ "is_error": false,
170
+ "result": "parent finished",
171
+ "permission_denials": []
172
+ }
173
+ }
174
+ ]
@@ -0,0 +1,120 @@
1
+ [
2
+ {
3
+ "sequence": 0,
4
+ "direction": "runtime-to-host",
5
+ "kind": "system/init",
6
+ "payload": {
7
+ "type": "system",
8
+ "subtype": "init",
9
+ "cwd": "/winter-fixture",
10
+ "model": "winter-test/echo",
11
+ "permissionMode": "default",
12
+ "tools": [
13
+ "Agent",
14
+ "AskUserQuestion",
15
+ "Bash",
16
+ "CronCreate",
17
+ "CronDelete",
18
+ "CronList",
19
+ "Edit",
20
+ "EnterPlanMode",
21
+ "EnterWorktree",
22
+ "ExitPlanMode",
23
+ "ExitWorktree",
24
+ "Glob",
25
+ "Grep",
26
+ "ListAgents",
27
+ "Monitor",
28
+ "NotebookEdit",
29
+ "PushNotification",
30
+ "Read",
31
+ "ReadNotifications",
32
+ "ReportFindings",
33
+ "ScheduleWakeup",
34
+ "SendMessage",
35
+ "Skill",
36
+ "TaskCreate",
37
+ "TaskGet",
38
+ "TaskList",
39
+ "TaskOutput",
40
+ "TaskStop",
41
+ "TaskUpdate",
42
+ "Workflow",
43
+ "Write"
44
+ ],
45
+ "apiKeySource": "none",
46
+ "slash_commands": [
47
+ "compact"
48
+ ],
49
+ "output_style": "default",
50
+ "skills": [],
51
+ "plugins": []
52
+ }
53
+ },
54
+ {
55
+ "sequence": 1,
56
+ "direction": "runtime-to-host",
57
+ "kind": "assistant",
58
+ "payload": {
59
+ "type": "assistant",
60
+ "message": {
61
+ "content": [
62
+ {
63
+ "type": "tool_use",
64
+ "id": "agent-call-1",
65
+ "name": "Agent",
66
+ "input": {
67
+ "description": "equivalence probe",
68
+ "prompt": "child probe text"
69
+ }
70
+ }
71
+ ]
72
+ }
73
+ }
74
+ },
75
+ {
76
+ "sequence": 2,
77
+ "direction": "runtime-to-host",
78
+ "kind": "user",
79
+ "payload": {
80
+ "type": "user",
81
+ "message": {
82
+ "content": [
83
+ {
84
+ "type": "tool_result",
85
+ "tool_use_id": "agent-call-1",
86
+ "content": "{\"agentId\":\"<scrubbed>\",\"content\":[{\"type\":\"text\",\"text\":\"child finished\"}],\"totalToolUseCount\":0,\"totalDurationMs\":\"<scrubbed>\",\"resolvedModel\":\"winter-test/echo\",\"prompt\":\"child probe text\"}"
87
+ }
88
+ ]
89
+ }
90
+ }
91
+ },
92
+ {
93
+ "sequence": 3,
94
+ "direction": "runtime-to-host",
95
+ "kind": "assistant",
96
+ "payload": {
97
+ "type": "assistant",
98
+ "message": {
99
+ "content": [
100
+ {
101
+ "type": "text",
102
+ "text": "parent finished"
103
+ }
104
+ ]
105
+ }
106
+ }
107
+ },
108
+ {
109
+ "sequence": 4,
110
+ "direction": "runtime-to-host",
111
+ "kind": "result",
112
+ "payload": {
113
+ "type": "result",
114
+ "subtype": "success",
115
+ "is_error": false,
116
+ "result": "parent finished",
117
+ "permission_denials": []
118
+ }
119
+ }
120
+ ]
@@ -0,0 +1,119 @@
1
+ [
2
+ {
3
+ "sequence": 0,
4
+ "direction": "runtime-to-host",
5
+ "kind": "system/init",
6
+ "payload": {
7
+ "type": "system",
8
+ "subtype": "init",
9
+ "cwd": "/winter-fixture",
10
+ "model": "winter-test/echo",
11
+ "permissionMode": "default",
12
+ "tools": [
13
+ "Agent",
14
+ "AskUserQuestion",
15
+ "Bash",
16
+ "CronCreate",
17
+ "CronDelete",
18
+ "CronList",
19
+ "Edit",
20
+ "EnterPlanMode",
21
+ "EnterWorktree",
22
+ "ExitPlanMode",
23
+ "ExitWorktree",
24
+ "Glob",
25
+ "Grep",
26
+ "ListAgents",
27
+ "Monitor",
28
+ "NotebookEdit",
29
+ "PushNotification",
30
+ "Read",
31
+ "ReadNotifications",
32
+ "ReportFindings",
33
+ "ScheduleWakeup",
34
+ "SendMessage",
35
+ "Skill",
36
+ "TaskCreate",
37
+ "TaskGet",
38
+ "TaskList",
39
+ "TaskOutput",
40
+ "TaskStop",
41
+ "TaskUpdate",
42
+ "Workflow",
43
+ "Write"
44
+ ],
45
+ "apiKeySource": "none",
46
+ "slash_commands": [
47
+ "compact"
48
+ ],
49
+ "output_style": "default",
50
+ "skills": [],
51
+ "plugins": []
52
+ }
53
+ },
54
+ {
55
+ "sequence": 1,
56
+ "direction": "runtime-to-host",
57
+ "kind": "assistant",
58
+ "payload": {
59
+ "type": "assistant",
60
+ "message": {
61
+ "content": [
62
+ {
63
+ "type": "tool_use",
64
+ "id": "test-call-1",
65
+ "name": "test_tool",
66
+ "input": {
67
+ "probe": true
68
+ }
69
+ }
70
+ ]
71
+ }
72
+ }
73
+ },
74
+ {
75
+ "sequence": 2,
76
+ "direction": "runtime-to-host",
77
+ "kind": "user",
78
+ "payload": {
79
+ "type": "user",
80
+ "message": {
81
+ "content": [
82
+ {
83
+ "type": "tool_result",
84
+ "tool_use_id": "test-call-1",
85
+ "content": "test_tool:{\"probe\":true}"
86
+ }
87
+ ]
88
+ }
89
+ }
90
+ },
91
+ {
92
+ "sequence": 3,
93
+ "direction": "runtime-to-host",
94
+ "kind": "assistant",
95
+ "payload": {
96
+ "type": "assistant",
97
+ "message": {
98
+ "content": [
99
+ {
100
+ "type": "text",
101
+ "text": "tool round done"
102
+ }
103
+ ]
104
+ }
105
+ }
106
+ },
107
+ {
108
+ "sequence": 4,
109
+ "direction": "runtime-to-host",
110
+ "kind": "result",
111
+ "payload": {
112
+ "type": "result",
113
+ "subtype": "success",
114
+ "is_error": false,
115
+ "result": "tool round done",
116
+ "permission_denials": []
117
+ }
118
+ }
119
+ ]