@zhivex-ai/core 0.7.0 → 0.8.0
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.
- package/README.md +5 -0
- package/dist/advanced-tool-registry.d.ts +112 -0
- package/dist/advanced-tool-registry.d.ts.map +1 -0
- package/dist/advanced-tool-registry.js +407 -0
- package/dist/advanced-tool-registry.js.map +1 -0
- package/dist/agent-evaluation.d.ts +176 -0
- package/dist/agent-evaluation.d.ts.map +1 -0
- package/dist/agent-evaluation.js +334 -0
- package/dist/agent-evaluation.js.map +1 -0
- package/dist/agent-store.d.ts.map +1 -1
- package/dist/agent-store.js +226 -8
- package/dist/agent-store.js.map +1 -1
- package/dist/agent-trace.d.ts +127 -0
- package/dist/agent-trace.d.ts.map +1 -0
- package/dist/agent-trace.js +331 -0
- package/dist/agent-trace.js.map +1 -0
- package/dist/agent.d.ts +12 -1
- package/dist/agent.d.ts.map +1 -1
- package/dist/agent.js +469 -36
- package/dist/agent.js.map +1 -1
- package/dist/api-stability.d.ts +9 -0
- package/dist/api-stability.d.ts.map +1 -0
- package/dist/api-stability.js +261 -0
- package/dist/api-stability.js.map +1 -0
- package/dist/artifact.d.ts +165 -0
- package/dist/artifact.d.ts.map +1 -0
- package/dist/artifact.js +994 -0
- package/dist/artifact.js.map +1 -0
- package/dist/errors.d.ts +2 -0
- package/dist/errors.d.ts.map +1 -1
- package/dist/errors.js +2 -0
- package/dist/errors.js.map +1 -1
- package/dist/generate-text.d.ts.map +1 -1
- package/dist/generate-text.js +6 -2
- package/dist/generate-text.js.map +1 -1
- package/dist/index.d.ts +27 -1
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +14 -1
- package/dist/index.js.map +1 -1
- package/dist/live-agent.d.ts.map +1 -1
- package/dist/live-agent.js +7 -1
- package/dist/live-agent.js.map +1 -1
- package/dist/provider-parity.d.ts +63 -0
- package/dist/provider-parity.d.ts.map +1 -0
- package/dist/provider-parity.js +175 -0
- package/dist/provider-parity.js.map +1 -0
- package/dist/runner.d.ts +111 -0
- package/dist/runner.d.ts.map +1 -0
- package/dist/runner.js +635 -0
- package/dist/runner.js.map +1 -0
- package/dist/safety-policy.d.ts +65 -0
- package/dist/safety-policy.d.ts.map +1 -0
- package/dist/safety-policy.js +308 -0
- package/dist/safety-policy.js.map +1 -0
- package/dist/types.d.ts +119 -3
- package/dist/types.d.ts.map +1 -1
- package/dist/workflow-artifacts.d.ts +28 -0
- package/dist/workflow-artifacts.d.ts.map +1 -0
- package/dist/workflow-artifacts.js +86 -0
- package/dist/workflow-artifacts.js.map +1 -0
- package/dist/workflow-evaluation-diff.d.ts +51 -0
- package/dist/workflow-evaluation-diff.d.ts.map +1 -0
- package/dist/workflow-evaluation-diff.js +141 -0
- package/dist/workflow-evaluation-diff.js.map +1 -0
- package/dist/workflow-evaluation.d.ts +94 -0
- package/dist/workflow-evaluation.d.ts.map +1 -0
- package/dist/workflow-evaluation.js +210 -0
- package/dist/workflow-evaluation.js.map +1 -0
- package/dist/workflow-state-service.d.ts +67 -0
- package/dist/workflow-state-service.d.ts.map +1 -0
- package/dist/workflow-state-service.js +498 -0
- package/dist/workflow-state-service.js.map +1 -0
- package/dist/workflow.d.ts +206 -0
- package/dist/workflow.d.ts.map +1 -0
- package/dist/workflow.js +727 -0
- package/dist/workflow.js.map +1 -0
- package/package.json +1 -1
|
@@ -0,0 +1,141 @@
|
|
|
1
|
+
const summarizeReport = (report) => ({
|
|
2
|
+
ok: report.ok,
|
|
3
|
+
total: report.total,
|
|
4
|
+
passed: report.passed,
|
|
5
|
+
failed: report.failed,
|
|
6
|
+
passRate: report.passRate
|
|
7
|
+
});
|
|
8
|
+
const failuresFor = (testCase) => new Set(testCase?.failures ?? []);
|
|
9
|
+
const difference = (left, right) => [...left].filter((value) => !right.has(value)).sort();
|
|
10
|
+
const reportCaseMap = (report) => new Map(report.cases.map((testCase) => [testCase.name, testCase]));
|
|
11
|
+
const caseChanged = (base, target, newFailures, resolvedFailures) => base.ok !== target.ok ||
|
|
12
|
+
base.status !== target.status ||
|
|
13
|
+
newFailures.length > 0 ||
|
|
14
|
+
resolvedFailures.length > 0;
|
|
15
|
+
export const compareWorkflowEvaluationReports = (base, target, options = {}) => {
|
|
16
|
+
const baseByName = reportCaseMap(base);
|
|
17
|
+
const targetByName = reportCaseMap(target);
|
|
18
|
+
const names = [...new Set([...baseByName.keys(), ...targetByName.keys()])].sort();
|
|
19
|
+
const cases = [];
|
|
20
|
+
const regressions = [];
|
|
21
|
+
const improvements = [];
|
|
22
|
+
if (target.passRate < base.passRate) {
|
|
23
|
+
regressions.push(`Pass rate decreased from ${base.passRate} to ${target.passRate}.`);
|
|
24
|
+
}
|
|
25
|
+
else if (target.passRate > base.passRate) {
|
|
26
|
+
improvements.push(`Pass rate increased from ${base.passRate} to ${target.passRate}.`);
|
|
27
|
+
}
|
|
28
|
+
if (target.ok !== base.ok) {
|
|
29
|
+
if (base.ok) {
|
|
30
|
+
regressions.push("Evaluation changed from passing to failing.");
|
|
31
|
+
}
|
|
32
|
+
else {
|
|
33
|
+
improvements.push("Evaluation changed from failing to passing.");
|
|
34
|
+
}
|
|
35
|
+
}
|
|
36
|
+
for (const name of names) {
|
|
37
|
+
const baseCase = baseByName.get(name);
|
|
38
|
+
const targetCase = targetByName.get(name);
|
|
39
|
+
const newFailures = difference(failuresFor(targetCase), failuresFor(baseCase));
|
|
40
|
+
const resolvedFailures = difference(failuresFor(baseCase), failuresFor(targetCase));
|
|
41
|
+
if (!baseCase && targetCase) {
|
|
42
|
+
cases.push({
|
|
43
|
+
name,
|
|
44
|
+
status: "added",
|
|
45
|
+
targetOk: targetCase.ok,
|
|
46
|
+
targetStatus: targetCase.status,
|
|
47
|
+
newFailures,
|
|
48
|
+
resolvedFailures
|
|
49
|
+
});
|
|
50
|
+
if (!targetCase.ok || newFailures.length > 0) {
|
|
51
|
+
regressions.push(`Failing case "${name}" was added.`);
|
|
52
|
+
}
|
|
53
|
+
continue;
|
|
54
|
+
}
|
|
55
|
+
if (baseCase && !targetCase) {
|
|
56
|
+
cases.push({
|
|
57
|
+
name,
|
|
58
|
+
status: "removed",
|
|
59
|
+
baseOk: baseCase.ok,
|
|
60
|
+
baseStatus: baseCase.status,
|
|
61
|
+
newFailures,
|
|
62
|
+
resolvedFailures
|
|
63
|
+
});
|
|
64
|
+
if (!baseCase.ok || resolvedFailures.length > 0) {
|
|
65
|
+
improvements.push(`Failing case "${name}" was removed.`);
|
|
66
|
+
}
|
|
67
|
+
continue;
|
|
68
|
+
}
|
|
69
|
+
if (!baseCase || !targetCase) {
|
|
70
|
+
continue;
|
|
71
|
+
}
|
|
72
|
+
const status = caseChanged(baseCase, targetCase, newFailures, resolvedFailures)
|
|
73
|
+
? "changed"
|
|
74
|
+
: "unchanged";
|
|
75
|
+
cases.push({
|
|
76
|
+
name,
|
|
77
|
+
status,
|
|
78
|
+
baseOk: baseCase.ok,
|
|
79
|
+
targetOk: targetCase.ok,
|
|
80
|
+
baseStatus: baseCase.status,
|
|
81
|
+
targetStatus: targetCase.status,
|
|
82
|
+
newFailures,
|
|
83
|
+
resolvedFailures
|
|
84
|
+
});
|
|
85
|
+
if (baseCase.ok && !targetCase.ok) {
|
|
86
|
+
regressions.push(`Case "${name}" regressed.`);
|
|
87
|
+
}
|
|
88
|
+
else if (!baseCase.ok && targetCase.ok) {
|
|
89
|
+
improvements.push(`Case "${name}" recovered.`);
|
|
90
|
+
}
|
|
91
|
+
if (newFailures.length > 0) {
|
|
92
|
+
regressions.push(`Case "${name}" has ${newFailures.length} new failure(s).`);
|
|
93
|
+
}
|
|
94
|
+
if (resolvedFailures.length > 0) {
|
|
95
|
+
improvements.push(`Case "${name}" resolved ${resolvedFailures.length} failure(s).`);
|
|
96
|
+
}
|
|
97
|
+
if (baseCase.status !== targetCase.status) {
|
|
98
|
+
const message = `Case "${name}" status changed from "${baseCase.status}" to "${targetCase.status}".`;
|
|
99
|
+
if (targetCase.ok || !baseCase.ok) {
|
|
100
|
+
improvements.push(message);
|
|
101
|
+
}
|
|
102
|
+
else {
|
|
103
|
+
regressions.push(message);
|
|
104
|
+
}
|
|
105
|
+
}
|
|
106
|
+
}
|
|
107
|
+
return {
|
|
108
|
+
ok: regressions.length === 0,
|
|
109
|
+
base: summarizeReport(base),
|
|
110
|
+
target: summarizeReport(target),
|
|
111
|
+
delta: {
|
|
112
|
+
total: target.total - base.total,
|
|
113
|
+
passed: target.passed - base.passed,
|
|
114
|
+
failed: target.failed - base.failed,
|
|
115
|
+
passRate: target.passRate - base.passRate
|
|
116
|
+
},
|
|
117
|
+
regressions,
|
|
118
|
+
improvements,
|
|
119
|
+
cases,
|
|
120
|
+
metadata: options.metadata
|
|
121
|
+
};
|
|
122
|
+
};
|
|
123
|
+
export const createWorkflowEvaluationDiffReport = (diff) => ({
|
|
124
|
+
...diff,
|
|
125
|
+
cases: diff.cases.map((testCase) => ({
|
|
126
|
+
...testCase,
|
|
127
|
+
newFailures: [...testCase.newFailures],
|
|
128
|
+
resolvedFailures: [...testCase.resolvedFailures]
|
|
129
|
+
})),
|
|
130
|
+
regressions: [...diff.regressions],
|
|
131
|
+
improvements: [...diff.improvements],
|
|
132
|
+
summary: {
|
|
133
|
+
added: diff.cases.filter((testCase) => testCase.status === "added").length,
|
|
134
|
+
removed: diff.cases.filter((testCase) => testCase.status === "removed").length,
|
|
135
|
+
changed: diff.cases.filter((testCase) => testCase.status === "changed").length,
|
|
136
|
+
unchanged: diff.cases.filter((testCase) => testCase.status === "unchanged").length,
|
|
137
|
+
newFailures: diff.cases.reduce((total, testCase) => total + testCase.newFailures.length, 0),
|
|
138
|
+
resolvedFailures: diff.cases.reduce((total, testCase) => total + testCase.resolvedFailures.length, 0)
|
|
139
|
+
}
|
|
140
|
+
});
|
|
141
|
+
//# sourceMappingURL=workflow-evaluation-diff.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"workflow-evaluation-diff.js","sourceRoot":"","sources":["../src/workflow-evaluation-diff.ts"],"names":[],"mappings":"AAoDA,MAAM,eAAe,GAAG,CAAC,MAAgC,EAAiC,EAAE,CAAC,CAAC;IAC5F,EAAE,EAAE,MAAM,CAAC,EAAE;IACb,KAAK,EAAE,MAAM,CAAC,KAAK;IACnB,MAAM,EAAE,MAAM,CAAC,MAAM;IACrB,MAAM,EAAE,MAAM,CAAC,MAAM;IACrB,QAAQ,EAAE,MAAM,CAAC,QAAQ;CAC1B,CAAC,CAAC;AAEH,MAAM,WAAW,GAAG,CAAC,QAAkD,EAAe,EAAE,CACtF,IAAI,GAAG,CAAC,QAAQ,EAAE,QAAQ,IAAI,EAAE,CAAC,CAAC;AAEpC,MAAM,UAAU,GAAG,CAAC,IAAiB,EAAE,KAAkB,EAAY,EAAE,CACrE,CAAC,GAAG,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;AAExD,MAAM,aAAa,GAAG,CAAC,MAAgC,EAA6C,EAAE,CACpG,IAAI,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC;AAErE,MAAM,WAAW,GAAG,CAClB,IAAkC,EAClC,MAAoC,EACpC,WAAqB,EACrB,gBAA0B,EACjB,EAAE,CACX,IAAI,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE;IACrB,IAAI,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM;IAC7B,WAAW,CAAC,MAAM,GAAG,CAAC;IACtB,gBAAgB,CAAC,MAAM,GAAG,CAAC,CAAC;AAE9B,MAAM,CAAC,MAAM,gCAAgC,GAAG,CAC9C,IAA8B,EAC9B,MAAgC,EAChC,UAAoD,EAAE,EAC9B,EAAE;IAC1B,MAAM,UAAU,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC;IACvC,MAAM,YAAY,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC;IAC3C,MAAM,KAAK,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,CAAC,GAAG,UAAU,CAAC,IAAI,EAAE,EAAE,GAAG,YAAY,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAClF,MAAM,KAAK,GAAiC,EAAE,CAAC;IAC/C,MAAM,WAAW,GAAa,EAAE,CAAC;IACjC,MAAM,YAAY,GAAa,EAAE,CAAC;IAElC,IAAI,MAAM,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC;QACpC,WAAW,CAAC,IAAI,CAAC,4BAA4B,IAAI,CAAC,QAAQ,OAAO,MAAM,CAAC,QAAQ,GAAG,CAAC,CAAC;IACvF,CAAC;SAAM,IAAI,MAAM,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC;QAC3C,YAAY,CAAC,IAAI,CAAC,4BAA4B,IAAI,CAAC,QAAQ,OAAO,MAAM,CAAC,QAAQ,GAAG,CAAC,CAAC;IACxF,CAAC;IAED,IAAI,MAAM,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,EAAE,CAAC;QAC1B,IAAI,IAAI,CAAC,EAAE,EAAE,CAAC;YACZ,WAAW,CAAC,IAAI,CAAC,6CAA6C,CAAC,CAAC;QAClE,CAAC;aAAM,CAAC;YACN,YAAY,CAAC,IAAI,CAAC,6CAA6C,CAAC,CAAC;QACnE,CAAC;IACH,CAAC;IAED,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,QAAQ,GAAG,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QACtC,MAAM,UAAU,GAAG,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QAC1C,MAAM,WAAW,GAAG,UAAU,CAAC,WAAW,CAAC,UAAU,CAAC,EAAE,WAAW,CAAC,QAAQ,CAAC,CAAC,CAAC;QAC/E,MAAM,gBAAgB,GAAG,UAAU,CAAC,WAAW,CAAC,QAAQ,CAAC,EAAE,WAAW,CAAC,UAAU,CAAC,CAAC,CAAC;QAEpF,IAAI,CAAC,QAAQ,IAAI,UAAU,EAAE,CAAC;YAC5B,KAAK,CAAC,IAAI,CAAC;gBACT,IAAI;gBACJ,MAAM,EAAE,OAAO;gBACf,QAAQ,EAAE,UAAU,CAAC,EAAE;gBACvB,YAAY,EAAE,UAAU,CAAC,MAAM;gBAC/B,WAAW;gBACX,gBAAgB;aACjB,CAAC,CAAC;YACH,IAAI,CAAC,UAAU,CAAC,EAAE,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAC7C,WAAW,CAAC,IAAI,CAAC,iBAAiB,IAAI,cAAc,CAAC,CAAC;YACxD,CAAC;YACD,SAAS;QACX,CAAC;QAED,IAAI,QAAQ,IAAI,CAAC,UAAU,EAAE,CAAC;YAC5B,KAAK,CAAC,IAAI,CAAC;gBACT,IAAI;gBACJ,MAAM,EAAE,SAAS;gBACjB,MAAM,EAAE,QAAQ,CAAC,EAAE;gBACnB,UAAU,EAAE,QAAQ,CAAC,MAAM;gBAC3B,WAAW;gBACX,gBAAgB;aACjB,CAAC,CAAC;YACH,IAAI,CAAC,QAAQ,CAAC,EAAE,IAAI,gBAAgB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAChD,YAAY,CAAC,IAAI,CAAC,iBAAiB,IAAI,gBAAgB,CAAC,CAAC;YAC3D,CAAC;YACD,SAAS;QACX,CAAC;QAED,IAAI,CAAC,QAAQ,IAAI,CAAC,UAAU,EAAE,CAAC;YAC7B,SAAS;QACX,CAAC;QAED,MAAM,MAAM,GAAqC,WAAW,CAAC,QAAQ,EAAE,UAAU,EAAE,WAAW,EAAE,gBAAgB,CAAC;YAC/G,CAAC,CAAC,SAAS;YACX,CAAC,CAAC,WAAW,CAAC;QAChB,KAAK,CAAC,IAAI,CAAC;YACT,IAAI;YACJ,MAAM;YACN,MAAM,EAAE,QAAQ,CAAC,EAAE;YACnB,QAAQ,EAAE,UAAU,CAAC,EAAE;YACvB,UAAU,EAAE,QAAQ,CAAC,MAAM;YAC3B,YAAY,EAAE,UAAU,CAAC,MAAM;YAC/B,WAAW;YACX,gBAAgB;SACjB,CAAC,CAAC;QAEH,IAAI,QAAQ,CAAC,EAAE,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;YAClC,WAAW,CAAC,IAAI,CAAC,SAAS,IAAI,cAAc,CAAC,CAAC;QAChD,CAAC;aAAM,IAAI,CAAC,QAAQ,CAAC,EAAE,IAAI,UAAU,CAAC,EAAE,EAAE,CAAC;YACzC,YAAY,CAAC,IAAI,CAAC,SAAS,IAAI,cAAc,CAAC,CAAC;QACjD,CAAC;QACD,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC3B,WAAW,CAAC,IAAI,CAAC,SAAS,IAAI,SAAS,WAAW,CAAC,MAAM,kBAAkB,CAAC,CAAC;QAC/E,CAAC;QACD,IAAI,gBAAgB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAChC,YAAY,CAAC,IAAI,CAAC,SAAS,IAAI,cAAc,gBAAgB,CAAC,MAAM,cAAc,CAAC,CAAC;QACtF,CAAC;QACD,IAAI,QAAQ,CAAC,MAAM,KAAK,UAAU,CAAC,MAAM,EAAE,CAAC;YAC1C,MAAM,OAAO,GAAG,SAAS,IAAI,0BAA0B,QAAQ,CAAC,MAAM,SAAS,UAAU,CAAC,MAAM,IAAI,CAAC;YACrG,IAAI,UAAU,CAAC,EAAE,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;gBAClC,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YAC7B,CAAC;iBAAM,CAAC;gBACN,WAAW,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YAC5B,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO;QACL,EAAE,EAAE,WAAW,CAAC,MAAM,KAAK,CAAC;QAC5B,IAAI,EAAE,eAAe,CAAC,IAAI,CAAC;QAC3B,MAAM,EAAE,eAAe,CAAC,MAAM,CAAC;QAC/B,KAAK,EAAE;YACL,KAAK,EAAE,MAAM,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK;YAChC,MAAM,EAAE,MAAM,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM;YACnC,MAAM,EAAE,MAAM,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM;YACnC,QAAQ,EAAE,MAAM,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ;SAC1C;QACD,WAAW;QACX,YAAY;QACZ,KAAK;QACL,QAAQ,EAAE,OAAO,CAAC,QAAQ;KAC3B,CAAC;AACJ,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,kCAAkC,GAAG,CAChD,IAA4B,EACE,EAAE,CAAC,CAAC;IAClC,GAAG,IAAI;IACP,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;QACnC,GAAG,QAAQ;QACX,WAAW,EAAE,CAAC,GAAG,QAAQ,CAAC,WAAW,CAAC;QACtC,gBAAgB,EAAE,CAAC,GAAG,QAAQ,CAAC,gBAAgB,CAAC;KACjD,CAAC,CAAC;IACH,WAAW,EAAE,CAAC,GAAG,IAAI,CAAC,WAAW,CAAC;IAClC,YAAY,EAAE,CAAC,GAAG,IAAI,CAAC,YAAY,CAAC;IACpC,OAAO,EAAE;QACP,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,MAAM;QAC1E,OAAO,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,MAAM;QAC9E,OAAO,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,MAAM;QAC9E,SAAS,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,MAAM,KAAK,WAAW,CAAC,CAAC,MAAM;QAClF,WAAW,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,QAAQ,EAAE,EAAE,CAAC,KAAK,GAAG,QAAQ,CAAC,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC;QAC3F,gBAAgB,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,QAAQ,EAAE,EAAE,CAAC,KAAK,GAAG,QAAQ,CAAC,gBAAgB,CAAC,MAAM,EAAE,CAAC,CAAC;KACtG;CACF,CAAC,CAAC"}
|
|
@@ -0,0 +1,94 @@
|
|
|
1
|
+
import type { JsonValue, LanguageModel } from "./types.js";
|
|
2
|
+
import { type WorkflowDefinition, type WorkflowReplayTimelineEvent, type WorkflowRunInput, type WorkflowRunOutput, type WorkflowStatus, type WorkflowStepStatus } from "./workflow.js";
|
|
3
|
+
export interface WorkflowEvaluationExpectations {
|
|
4
|
+
status?: WorkflowStatus;
|
|
5
|
+
outputs?: Record<string, JsonValue>;
|
|
6
|
+
outputContains?: Record<string, string> | string[];
|
|
7
|
+
stepStatuses?: Record<string, WorkflowStepStatus>;
|
|
8
|
+
stepOutputContains?: Record<string, string>;
|
|
9
|
+
errorContains?: string;
|
|
10
|
+
timelineContains?: WorkflowReplayTimelineEvent["type"][];
|
|
11
|
+
}
|
|
12
|
+
export interface WorkflowEvaluationCase {
|
|
13
|
+
name: string;
|
|
14
|
+
input: WorkflowRunInput;
|
|
15
|
+
expectations?: WorkflowEvaluationExpectations;
|
|
16
|
+
metadata?: Record<string, JsonValue>;
|
|
17
|
+
}
|
|
18
|
+
export interface WorkflowEvaluationCaseResult {
|
|
19
|
+
name: string;
|
|
20
|
+
ok: boolean;
|
|
21
|
+
output: WorkflowRunOutput;
|
|
22
|
+
failures: string[];
|
|
23
|
+
metadata?: Record<string, JsonValue>;
|
|
24
|
+
}
|
|
25
|
+
export interface WorkflowEvaluationResult {
|
|
26
|
+
ok: boolean;
|
|
27
|
+
cases: WorkflowEvaluationCaseResult[];
|
|
28
|
+
}
|
|
29
|
+
export interface WorkflowEvaluationJudgeResult {
|
|
30
|
+
score: number;
|
|
31
|
+
feedback?: string;
|
|
32
|
+
metadata?: Record<string, JsonValue>;
|
|
33
|
+
}
|
|
34
|
+
export type WorkflowEvaluationJudge = ((result: WorkflowEvaluationResult) => WorkflowEvaluationJudgeResult | Promise<WorkflowEvaluationJudgeResult>) | {
|
|
35
|
+
model: LanguageModel;
|
|
36
|
+
prompt?: string;
|
|
37
|
+
};
|
|
38
|
+
export interface RunWorkflowEvaluationOptions {
|
|
39
|
+
workflow: WorkflowDefinition | ((testCase: WorkflowEvaluationCase) => WorkflowDefinition | Promise<WorkflowDefinition>);
|
|
40
|
+
}
|
|
41
|
+
export interface WorkflowEvaluationFixture {
|
|
42
|
+
name?: string;
|
|
43
|
+
dataset: WorkflowEvaluationCase[];
|
|
44
|
+
expectedOk?: boolean;
|
|
45
|
+
metadata?: Record<string, JsonValue>;
|
|
46
|
+
createdAt?: number;
|
|
47
|
+
}
|
|
48
|
+
export interface WorkflowEvaluationReportCase {
|
|
49
|
+
name: string;
|
|
50
|
+
ok: boolean;
|
|
51
|
+
status: WorkflowStatus;
|
|
52
|
+
failures: string[];
|
|
53
|
+
outputPreview: string;
|
|
54
|
+
outputKeys: string[];
|
|
55
|
+
stepCount: number;
|
|
56
|
+
stepStatusCounts: Record<string, number>;
|
|
57
|
+
timelineEventCounts: Record<string, number>;
|
|
58
|
+
durationMs?: number;
|
|
59
|
+
metadata?: Record<string, JsonValue>;
|
|
60
|
+
}
|
|
61
|
+
export interface WorkflowEvaluationReport {
|
|
62
|
+
ok: boolean;
|
|
63
|
+
total: number;
|
|
64
|
+
passed: number;
|
|
65
|
+
failed: number;
|
|
66
|
+
passRate: number;
|
|
67
|
+
statusCounts: Record<string, number>;
|
|
68
|
+
stepCount: number;
|
|
69
|
+
stepStatusCounts: Record<string, number>;
|
|
70
|
+
timelineEventCounts: Record<string, number>;
|
|
71
|
+
failures: Array<{
|
|
72
|
+
name: string;
|
|
73
|
+
failures: string[];
|
|
74
|
+
}>;
|
|
75
|
+
cases: WorkflowEvaluationReportCase[];
|
|
76
|
+
judge?: WorkflowEvaluationJudgeResult;
|
|
77
|
+
metadata?: Record<string, JsonValue>;
|
|
78
|
+
}
|
|
79
|
+
export declare const runWorkflowEvaluation: (dataset: WorkflowEvaluationCase[], options: RunWorkflowEvaluationOptions) => Promise<WorkflowEvaluationResult>;
|
|
80
|
+
export declare const createWorkflowEvaluationFixture: (options: {
|
|
81
|
+
name?: string;
|
|
82
|
+
dataset: WorkflowEvaluationCase[];
|
|
83
|
+
expectedOk?: boolean;
|
|
84
|
+
metadata?: Record<string, JsonValue>;
|
|
85
|
+
createdAt?: number;
|
|
86
|
+
}) => WorkflowEvaluationFixture;
|
|
87
|
+
export declare const runWorkflowEvaluationFixture: (fixture: WorkflowEvaluationFixture, options: RunWorkflowEvaluationOptions) => Promise<WorkflowEvaluationResult>;
|
|
88
|
+
export declare const createWorkflowEvaluationReport: (result: WorkflowEvaluationResult, options?: {
|
|
89
|
+
judge?: WorkflowEvaluationJudgeResult;
|
|
90
|
+
metadata?: Record<string, JsonValue>;
|
|
91
|
+
outputPreviewLength?: number;
|
|
92
|
+
}) => WorkflowEvaluationReport;
|
|
93
|
+
export declare const judgeWorkflowEvaluation: (result: WorkflowEvaluationResult, judge: WorkflowEvaluationJudge) => Promise<WorkflowEvaluationJudgeResult>;
|
|
94
|
+
//# sourceMappingURL=workflow-evaluation.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"workflow-evaluation.d.ts","sourceRoot":"","sources":["../src/workflow-evaluation.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAC3D,OAAO,EAGL,KAAK,kBAAkB,EACvB,KAAK,2BAA2B,EAChC,KAAK,gBAAgB,EACrB,KAAK,iBAAiB,EACtB,KAAK,cAAc,EAEnB,KAAK,kBAAkB,EACxB,MAAM,eAAe,CAAC;AAEvB,MAAM,WAAW,8BAA8B;IAC7C,MAAM,CAAC,EAAE,cAAc,CAAC;IACxB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;IACpC,cAAc,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,MAAM,EAAE,CAAC;IACnD,YAAY,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAAC,CAAC;IAClD,kBAAkB,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC5C,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,gBAAgB,CAAC,EAAE,2BAA2B,CAAC,MAAM,CAAC,EAAE,CAAC;CAC1D;AAED,MAAM,WAAW,sBAAsB;IACrC,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,gBAAgB,CAAC;IACxB,YAAY,CAAC,EAAE,8BAA8B,CAAC;IAC9C,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;CACtC;AAED,MAAM,WAAW,4BAA4B;IAC3C,IAAI,EAAE,MAAM,CAAC;IACb,EAAE,EAAE,OAAO,CAAC;IACZ,MAAM,EAAE,iBAAiB,CAAC;IAC1B,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;CACtC;AAED,MAAM,WAAW,wBAAwB;IACvC,EAAE,EAAE,OAAO,CAAC;IACZ,KAAK,EAAE,4BAA4B,EAAE,CAAC;CACvC;AAED,MAAM,WAAW,6BAA6B;IAC5C,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;CACtC;AAED,MAAM,MAAM,uBAAuB,GAC/B,CAAC,CAAC,MAAM,EAAE,wBAAwB,KAAK,6BAA6B,GAAG,OAAO,CAAC,6BAA6B,CAAC,CAAC,GAC9G;IACE,KAAK,EAAE,aAAa,CAAC;IACrB,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB,CAAC;AAEN,MAAM,WAAW,4BAA4B;IAC3C,QAAQ,EACJ,kBAAkB,GAClB,CAAC,CAAC,QAAQ,EAAE,sBAAsB,KAAK,kBAAkB,GAAG,OAAO,CAAC,kBAAkB,CAAC,CAAC,CAAC;CAC9F;AAED,MAAM,WAAW,yBAAyB;IACxC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,sBAAsB,EAAE,CAAC;IAClC,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;IACrC,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,4BAA4B;IAC3C,IAAI,EAAE,MAAM,CAAC;IACb,EAAE,EAAE,OAAO,CAAC;IACZ,MAAM,EAAE,cAAc,CAAC;IACvB,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;IAClB,gBAAgB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACzC,mBAAmB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC5C,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;CACtC;AAED,MAAM,WAAW,wBAAwB;IACvC,EAAE,EAAE,OAAO,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACrC,SAAS,EAAE,MAAM,CAAC;IAClB,gBAAgB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACzC,mBAAmB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC5C,QAAQ,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,EAAE,CAAA;KAAE,CAAC,CAAC;IACtD,KAAK,EAAE,4BAA4B,EAAE,CAAC;IACtC,KAAK,CAAC,EAAE,6BAA6B,CAAC;IACtC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;CACtC;AA4GD,eAAO,MAAM,qBAAqB,GAChC,SAAS,sBAAsB,EAAE,EACjC,SAAS,4BAA4B,KACpC,OAAO,CAAC,wBAAwB,CAoBlC,CAAC;AAEF,eAAO,MAAM,+BAA+B,GAAI,SAAS;IACvD,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,sBAAsB,EAAE,CAAC;IAClC,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;IACrC,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,KAAG,yBAMF,CAAC;AAEH,eAAO,MAAM,4BAA4B,GACvC,SAAS,yBAAyB,EAClC,SAAS,4BAA4B,KACpC,OAAO,CAAC,wBAAwB,CAAoD,CAAC;AAWxF,eAAO,MAAM,8BAA8B,GACzC,QAAQ,wBAAwB,EAChC,UAAS;IACP,KAAK,CAAC,EAAE,6BAA6B,CAAC;IACtC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;IACrC,mBAAmB,CAAC,EAAE,MAAM,CAAC;CACzB,KACL,wBA6DF,CAAC;AAiBF,eAAO,MAAM,uBAAuB,GAClC,QAAQ,wBAAwB,EAChC,OAAO,uBAAuB,KAC7B,OAAO,CAAC,6BAA6B,CAgBvC,CAAC"}
|
|
@@ -0,0 +1,210 @@
|
|
|
1
|
+
import { generateText } from "./generate-text.js";
|
|
2
|
+
import { createTextMessage, serializeJsonValue } from "./messages.js";
|
|
3
|
+
import { replayWorkflowRun, runWorkflow } from "./workflow.js";
|
|
4
|
+
const preview = (text, length) => text.length > length ? `${text.slice(0, Math.max(0, length))}...` : text;
|
|
5
|
+
const flattenStepResults = (steps) => steps.flatMap((step) => [step, ...flattenStepResults(step.children ?? [])]);
|
|
6
|
+
const createStepMap = (steps) => {
|
|
7
|
+
const map = new Map();
|
|
8
|
+
for (const step of flattenStepResults(steps)) {
|
|
9
|
+
map.set(step.id, [...(map.get(step.id) ?? []), step]);
|
|
10
|
+
}
|
|
11
|
+
return map;
|
|
12
|
+
};
|
|
13
|
+
const getLatestStep = (stepMap, stepId) => stepMap.get(stepId)?.at(-1);
|
|
14
|
+
const valueEquals = (left, right) => JSON.stringify(left) === JSON.stringify(right);
|
|
15
|
+
const evaluateExpectations = (output, expectations) => {
|
|
16
|
+
const failures = [];
|
|
17
|
+
if (!expectations) {
|
|
18
|
+
return failures;
|
|
19
|
+
}
|
|
20
|
+
if (expectations.status && output.status !== expectations.status) {
|
|
21
|
+
failures.push(`Expected status "${expectations.status}", received "${output.status}".`);
|
|
22
|
+
}
|
|
23
|
+
if (expectations.outputs) {
|
|
24
|
+
for (const [key, expected] of Object.entries(expectations.outputs)) {
|
|
25
|
+
if (!(key in output.outputs)) {
|
|
26
|
+
failures.push(`Expected output key "${key}".`);
|
|
27
|
+
continue;
|
|
28
|
+
}
|
|
29
|
+
if (!valueEquals(output.outputs[key], expected)) {
|
|
30
|
+
failures.push(`Expected output "${key}" to equal ${JSON.stringify(expected)}.`);
|
|
31
|
+
}
|
|
32
|
+
}
|
|
33
|
+
}
|
|
34
|
+
if (Array.isArray(expectations.outputContains)) {
|
|
35
|
+
const serializedOutputs = JSON.stringify(output.outputs);
|
|
36
|
+
for (const expectedText of expectations.outputContains) {
|
|
37
|
+
if (!serializedOutputs.includes(expectedText)) {
|
|
38
|
+
failures.push(`Expected serialized outputs to contain "${expectedText}".`);
|
|
39
|
+
}
|
|
40
|
+
}
|
|
41
|
+
}
|
|
42
|
+
else if (expectations.outputContains) {
|
|
43
|
+
for (const [key, expectedText] of Object.entries(expectations.outputContains)) {
|
|
44
|
+
const value = output.outputs[key];
|
|
45
|
+
if (value === undefined || !String(value).includes(expectedText)) {
|
|
46
|
+
failures.push(`Expected output "${key}" to contain "${expectedText}".`);
|
|
47
|
+
}
|
|
48
|
+
}
|
|
49
|
+
}
|
|
50
|
+
const stepMap = createStepMap(output.steps);
|
|
51
|
+
if (expectations.stepStatuses) {
|
|
52
|
+
for (const [stepId, expectedStatus] of Object.entries(expectations.stepStatuses)) {
|
|
53
|
+
const step = getLatestStep(stepMap, stepId);
|
|
54
|
+
if (!step) {
|
|
55
|
+
failures.push(`Expected step "${stepId}".`);
|
|
56
|
+
continue;
|
|
57
|
+
}
|
|
58
|
+
if (step.status !== expectedStatus) {
|
|
59
|
+
failures.push(`Expected step "${stepId}" status "${expectedStatus}", received "${step.status}".`);
|
|
60
|
+
}
|
|
61
|
+
}
|
|
62
|
+
}
|
|
63
|
+
if (expectations.stepOutputContains) {
|
|
64
|
+
for (const [stepId, expectedText] of Object.entries(expectations.stepOutputContains)) {
|
|
65
|
+
const step = getLatestStep(stepMap, stepId);
|
|
66
|
+
if (!step?.outputText?.includes(expectedText)) {
|
|
67
|
+
failures.push(`Expected step "${stepId}" output to contain "${expectedText}".`);
|
|
68
|
+
}
|
|
69
|
+
}
|
|
70
|
+
}
|
|
71
|
+
if (expectations.errorContains) {
|
|
72
|
+
const errors = flattenStepResults(output.steps).flatMap((step) => step.error?.message ? [step.error.message] : []);
|
|
73
|
+
if (!errors.some((message) => message.includes(expectations.errorContains))) {
|
|
74
|
+
failures.push(`Expected workflow error to contain "${expectations.errorContains}".`);
|
|
75
|
+
}
|
|
76
|
+
}
|
|
77
|
+
if (expectations.timelineContains?.length) {
|
|
78
|
+
const replay = replayWorkflowRun(output.state);
|
|
79
|
+
const eventTypes = new Set(replay.timeline.map((event) => event.type));
|
|
80
|
+
for (const eventType of expectations.timelineContains) {
|
|
81
|
+
if (!eventTypes.has(eventType)) {
|
|
82
|
+
failures.push(`Expected replay timeline event "${eventType}".`);
|
|
83
|
+
}
|
|
84
|
+
}
|
|
85
|
+
}
|
|
86
|
+
return failures;
|
|
87
|
+
};
|
|
88
|
+
export const runWorkflowEvaluation = async (dataset, options) => {
|
|
89
|
+
const cases = [];
|
|
90
|
+
for (const testCase of dataset) {
|
|
91
|
+
const workflow = typeof options.workflow === "function" ? await options.workflow(testCase) : options.workflow;
|
|
92
|
+
const output = await runWorkflow(workflow, testCase.input);
|
|
93
|
+
const failures = evaluateExpectations(output, testCase.expectations);
|
|
94
|
+
cases.push({
|
|
95
|
+
name: testCase.name,
|
|
96
|
+
ok: failures.length === 0,
|
|
97
|
+
output,
|
|
98
|
+
failures,
|
|
99
|
+
metadata: testCase.metadata
|
|
100
|
+
});
|
|
101
|
+
}
|
|
102
|
+
return {
|
|
103
|
+
ok: cases.every((testCase) => testCase.ok),
|
|
104
|
+
cases
|
|
105
|
+
};
|
|
106
|
+
};
|
|
107
|
+
export const createWorkflowEvaluationFixture = (options) => ({
|
|
108
|
+
name: options.name,
|
|
109
|
+
dataset: serializeJsonValue(options.dataset),
|
|
110
|
+
expectedOk: options.expectedOk,
|
|
111
|
+
metadata: options.metadata,
|
|
112
|
+
createdAt: options.createdAt ?? Date.now()
|
|
113
|
+
});
|
|
114
|
+
export const runWorkflowEvaluationFixture = async (fixture, options) => runWorkflowEvaluation(fixture.dataset, options);
|
|
115
|
+
const workflowDuration = (output) => output.state.createdAt !== undefined && output.state.updatedAt !== undefined
|
|
116
|
+
? Math.max(0, output.state.updatedAt - output.state.createdAt)
|
|
117
|
+
: undefined;
|
|
118
|
+
const incrementCounts = (counts, key) => {
|
|
119
|
+
counts[key] = (counts[key] ?? 0) + 1;
|
|
120
|
+
};
|
|
121
|
+
export const createWorkflowEvaluationReport = (result, options = {}) => {
|
|
122
|
+
const outputPreviewLength = options.outputPreviewLength ?? 500;
|
|
123
|
+
const cases = result.cases.map((testCase) => {
|
|
124
|
+
const steps = flattenStepResults(testCase.output.steps);
|
|
125
|
+
const stepStatusCounts = {};
|
|
126
|
+
const timelineEventCounts = {};
|
|
127
|
+
const replay = replayWorkflowRun(testCase.output.state);
|
|
128
|
+
for (const step of steps) {
|
|
129
|
+
incrementCounts(stepStatusCounts, step.status);
|
|
130
|
+
}
|
|
131
|
+
for (const event of replay.timeline) {
|
|
132
|
+
incrementCounts(timelineEventCounts, event.type);
|
|
133
|
+
}
|
|
134
|
+
return {
|
|
135
|
+
name: testCase.name,
|
|
136
|
+
ok: testCase.ok,
|
|
137
|
+
status: testCase.output.status,
|
|
138
|
+
failures: [...testCase.failures],
|
|
139
|
+
outputPreview: preview(JSON.stringify(testCase.output.outputs), outputPreviewLength),
|
|
140
|
+
outputKeys: Object.keys(testCase.output.outputs),
|
|
141
|
+
stepCount: steps.length,
|
|
142
|
+
stepStatusCounts,
|
|
143
|
+
timelineEventCounts,
|
|
144
|
+
durationMs: workflowDuration(testCase.output),
|
|
145
|
+
metadata: testCase.metadata
|
|
146
|
+
};
|
|
147
|
+
});
|
|
148
|
+
const statusCounts = {};
|
|
149
|
+
const stepStatusCounts = {};
|
|
150
|
+
const timelineEventCounts = {};
|
|
151
|
+
for (const testCase of cases) {
|
|
152
|
+
incrementCounts(statusCounts, testCase.status);
|
|
153
|
+
for (const [status, count] of Object.entries(testCase.stepStatusCounts)) {
|
|
154
|
+
stepStatusCounts[status] = (stepStatusCounts[status] ?? 0) + count;
|
|
155
|
+
}
|
|
156
|
+
for (const [eventType, count] of Object.entries(testCase.timelineEventCounts)) {
|
|
157
|
+
timelineEventCounts[eventType] = (timelineEventCounts[eventType] ?? 0) + count;
|
|
158
|
+
}
|
|
159
|
+
}
|
|
160
|
+
const passed = cases.filter((testCase) => testCase.ok).length;
|
|
161
|
+
const failed = cases.length - passed;
|
|
162
|
+
return {
|
|
163
|
+
ok: result.ok,
|
|
164
|
+
total: cases.length,
|
|
165
|
+
passed,
|
|
166
|
+
failed,
|
|
167
|
+
passRate: cases.length ? passed / cases.length : 1,
|
|
168
|
+
statusCounts,
|
|
169
|
+
stepCount: cases.reduce((total, testCase) => total + testCase.stepCount, 0),
|
|
170
|
+
stepStatusCounts,
|
|
171
|
+
timelineEventCounts,
|
|
172
|
+
failures: cases.filter((testCase) => testCase.failures.length).map((testCase) => ({
|
|
173
|
+
name: testCase.name,
|
|
174
|
+
failures: testCase.failures
|
|
175
|
+
})),
|
|
176
|
+
cases,
|
|
177
|
+
judge: options.judge,
|
|
178
|
+
metadata: options.metadata
|
|
179
|
+
};
|
|
180
|
+
};
|
|
181
|
+
const parseJudgeResponse = (text) => {
|
|
182
|
+
try {
|
|
183
|
+
const parsed = JSON.parse(text);
|
|
184
|
+
return {
|
|
185
|
+
score: typeof parsed.score === "number" ? Math.max(0, Math.min(1, parsed.score)) : 0,
|
|
186
|
+
feedback: parsed.feedback
|
|
187
|
+
};
|
|
188
|
+
}
|
|
189
|
+
catch {
|
|
190
|
+
return {
|
|
191
|
+
score: 0,
|
|
192
|
+
feedback: text
|
|
193
|
+
};
|
|
194
|
+
}
|
|
195
|
+
};
|
|
196
|
+
export const judgeWorkflowEvaluation = async (result, judge) => {
|
|
197
|
+
if (typeof judge === "function") {
|
|
198
|
+
return judge(result);
|
|
199
|
+
}
|
|
200
|
+
const prompt = judge.prompt ??
|
|
201
|
+
"Score this workflow evaluation from 0 to 1. Return JSON with fields score and feedback.";
|
|
202
|
+
const response = await generateText({
|
|
203
|
+
model: judge.model,
|
|
204
|
+
messages: [
|
|
205
|
+
createTextMessage("user", `${prompt}\n\n${JSON.stringify(serializeJsonValue(result), null, 2)}`)
|
|
206
|
+
]
|
|
207
|
+
});
|
|
208
|
+
return parseJudgeResponse(response.text);
|
|
209
|
+
};
|
|
210
|
+
//# sourceMappingURL=workflow-evaluation.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"workflow-evaluation.js","sourceRoot":"","sources":["../src/workflow-evaluation.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,iBAAiB,EAAE,kBAAkB,EAAE,MAAM,eAAe,CAAC;AAEtE,OAAO,EACL,iBAAiB,EACjB,WAAW,EAQZ,MAAM,eAAe,CAAC;AAyFvB,MAAM,OAAO,GAAG,CAAC,IAAY,EAAE,MAAc,EAAU,EAAE,CACvD,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC;AAE3E,MAAM,kBAAkB,GAAG,CAAC,KAA2B,EAAwB,EAAE,CAC/E,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,GAAG,kBAAkB,CAAC,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;AAE9E,MAAM,aAAa,GAAG,CAAC,KAA2B,EAAqC,EAAE;IACvF,MAAM,GAAG,GAAG,IAAI,GAAG,EAAgC,CAAC;IACpD,KAAK,MAAM,IAAI,IAAI,kBAAkB,CAAC,KAAK,CAAC,EAAE,CAAC;QAC7C,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;IACxD,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC,CAAC;AAEF,MAAM,aAAa,GAAG,CACpB,OAA0C,EAC1C,MAAc,EACkB,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;AAEjE,MAAM,WAAW,GAAG,CAAC,IAAe,EAAE,KAAgB,EAAW,EAAE,CACjE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;AAEjD,MAAM,oBAAoB,GAAG,CAC3B,MAAyB,EACzB,YAAwD,EAC9C,EAAE;IACZ,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,CAAC,YAAY,EAAE,CAAC;QAClB,OAAO,QAAQ,CAAC;IAClB,CAAC;IAED,IAAI,YAAY,CAAC,MAAM,IAAI,MAAM,CAAC,MAAM,KAAK,YAAY,CAAC,MAAM,EAAE,CAAC;QACjE,QAAQ,CAAC,IAAI,CAAC,oBAAoB,YAAY,CAAC,MAAM,gBAAgB,MAAM,CAAC,MAAM,IAAI,CAAC,CAAC;IAC1F,CAAC;IAED,IAAI,YAAY,CAAC,OAAO,EAAE,CAAC;QACzB,KAAK,MAAM,CAAC,GAAG,EAAE,QAAQ,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,YAAY,CAAC,OAAO,CAAC,EAAE,CAAC;YACnE,IAAI,CAAC,CAAC,GAAG,IAAI,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;gBAC7B,QAAQ,CAAC,IAAI,CAAC,wBAAwB,GAAG,IAAI,CAAC,CAAC;gBAC/C,SAAS;YACX,CAAC;YACD,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,GAAG,CAAE,EAAE,QAAQ,CAAC,EAAE,CAAC;gBACjD,QAAQ,CAAC,IAAI,CAAC,oBAAoB,GAAG,cAAc,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;YAClF,CAAC;QACH,CAAC;IACH,CAAC;IAED,IAAI,KAAK,CAAC,OAAO,CAAC,YAAY,CAAC,cAAc,CAAC,EAAE,CAAC;QAC/C,MAAM,iBAAiB,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACzD,KAAK,MAAM,YAAY,IAAI,YAAY,CAAC,cAAc,EAAE,CAAC;YACvD,IAAI,CAAC,iBAAiB,CAAC,QAAQ,CAAC,YAAY,CAAC,EAAE,CAAC;gBAC9C,QAAQ,CAAC,IAAI,CAAC,2CAA2C,YAAY,IAAI,CAAC,CAAC;YAC7E,CAAC;QACH,CAAC;IACH,CAAC;SAAM,IAAI,YAAY,CAAC,cAAc,EAAE,CAAC;QACvC,KAAK,MAAM,CAAC,GAAG,EAAE,YAAY,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,YAAY,CAAC,cAAc,CAAC,EAAE,CAAC;YAC9E,MAAM,KAAK,GAAG,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;YAClC,IAAI,KAAK,KAAK,SAAS,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,QAAQ,CAAC,YAAY,CAAC,EAAE,CAAC;gBACjE,QAAQ,CAAC,IAAI,CAAC,oBAAoB,GAAG,iBAAiB,YAAY,IAAI,CAAC,CAAC;YAC1E,CAAC;QACH,CAAC;IACH,CAAC;IAED,MAAM,OAAO,GAAG,aAAa,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAC5C,IAAI,YAAY,CAAC,YAAY,EAAE,CAAC;QAC9B,KAAK,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,YAAY,CAAC,YAAY,CAAC,EAAE,CAAC;YACjF,MAAM,IAAI,GAAG,aAAa,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YAC5C,IAAI,CAAC,IAAI,EAAE,CAAC;gBACV,QAAQ,CAAC,IAAI,CAAC,kBAAkB,MAAM,IAAI,CAAC,CAAC;gBAC5C,SAAS;YACX,CAAC;YACD,IAAI,IAAI,CAAC,MAAM,KAAK,cAAc,EAAE,CAAC;gBACnC,QAAQ,CAAC,IAAI,CAAC,kBAAkB,MAAM,aAAa,cAAc,gBAAgB,IAAI,CAAC,MAAM,IAAI,CAAC,CAAC;YACpG,CAAC;QACH,CAAC;IACH,CAAC;IAED,IAAI,YAAY,CAAC,kBAAkB,EAAE,CAAC;QACpC,KAAK,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,YAAY,CAAC,kBAAkB,CAAC,EAAE,CAAC;YACrF,MAAM,IAAI,GAAG,aAAa,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YAC5C,IAAI,CAAC,IAAI,EAAE,UAAU,EAAE,QAAQ,CAAC,YAAY,CAAC,EAAE,CAAC;gBAC9C,QAAQ,CAAC,IAAI,CAAC,kBAAkB,MAAM,wBAAwB,YAAY,IAAI,CAAC,CAAC;YAClF,CAAC;QACH,CAAC;IACH,CAAC;IAED,IAAI,YAAY,CAAC,aAAa,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAG,kBAAkB,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QACnH,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAC,aAAc,CAAC,CAAC,EAAE,CAAC;YAC7E,QAAQ,CAAC,IAAI,CAAC,uCAAuC,YAAY,CAAC,aAAa,IAAI,CAAC,CAAC;QACvF,CAAC;IACH,CAAC;IAED,IAAI,YAAY,CAAC,gBAAgB,EAAE,MAAM,EAAE,CAAC;QAC1C,MAAM,MAAM,GAAG,iBAAiB,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAC/C,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;QACvE,KAAK,MAAM,SAAS,IAAI,YAAY,CAAC,gBAAgB,EAAE,CAAC;YACtD,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,SAAS,CAAC,EAAE,CAAC;gBAC/B,QAAQ,CAAC,IAAI,CAAC,mCAAmC,SAAS,IAAI,CAAC,CAAC;YAClE,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,qBAAqB,GAAG,KAAK,EACxC,OAAiC,EACjC,OAAqC,EACF,EAAE;IACrC,MAAM,KAAK,GAAmC,EAAE,CAAC;IAEjD,KAAK,MAAM,QAAQ,IAAI,OAAO,EAAE,CAAC;QAC/B,MAAM,QAAQ,GAAG,OAAO,OAAO,CAAC,QAAQ,KAAK,UAAU,CAAC,CAAC,CAAC,MAAM,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC;QAC9G,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC,QAAQ,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAC;QAC3D,MAAM,QAAQ,GAAG,oBAAoB,CAAC,MAAM,EAAE,QAAQ,CAAC,YAAY,CAAC,CAAC;QACrE,KAAK,CAAC,IAAI,CAAC;YACT,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,EAAE,EAAE,QAAQ,CAAC,MAAM,KAAK,CAAC;YACzB,MAAM;YACN,QAAQ;YACR,QAAQ,EAAE,QAAQ,CAAC,QAAQ;SAC5B,CAAC,CAAC;IACL,CAAC;IAED,OAAO;QACL,EAAE,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC;QAC1C,KAAK;KACN,CAAC;AACJ,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,+BAA+B,GAAG,CAAC,OAM/C,EAA6B,EAAE,CAAC,CAAC;IAChC,IAAI,EAAE,OAAO,CAAC,IAAI;IAClB,OAAO,EAAE,kBAAkB,CAAC,OAAO,CAAC,OAAO,CAAwC;IACnF,UAAU,EAAE,OAAO,CAAC,UAAU;IAC9B,QAAQ,EAAE,OAAO,CAAC,QAAQ;IAC1B,SAAS,EAAE,OAAO,CAAC,SAAS,IAAI,IAAI,CAAC,GAAG,EAAE;CAC3C,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,4BAA4B,GAAG,KAAK,EAC/C,OAAkC,EAClC,OAAqC,EACF,EAAE,CAAC,qBAAqB,CAAC,OAAO,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;AAExF,MAAM,gBAAgB,GAAG,CAAC,MAAyB,EAAsB,EAAE,CACzE,MAAM,CAAC,KAAK,CAAC,SAAS,KAAK,SAAS,IAAI,MAAM,CAAC,KAAK,CAAC,SAAS,KAAK,SAAS;IAC1E,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,KAAK,CAAC,SAAS,GAAG,MAAM,CAAC,KAAK,CAAC,SAAS,CAAC;IAC9D,CAAC,CAAC,SAAS,CAAC;AAEhB,MAAM,eAAe,GAAG,CAAC,MAA8B,EAAE,GAAW,EAAE,EAAE;IACtE,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;AACvC,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,8BAA8B,GAAG,CAC5C,MAAgC,EAChC,UAII,EAAE,EACoB,EAAE;IAC5B,MAAM,mBAAmB,GAAG,OAAO,CAAC,mBAAmB,IAAI,GAAG,CAAC;IAC/D,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAgC,EAAE;QACxE,MAAM,KAAK,GAAG,kBAAkB,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACxD,MAAM,gBAAgB,GAA2B,EAAE,CAAC;QACpD,MAAM,mBAAmB,GAA2B,EAAE,CAAC;QACvD,MAAM,MAAM,GAAG,iBAAiB,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACxD,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,eAAe,CAAC,gBAAgB,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC;QACjD,CAAC;QACD,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACpC,eAAe,CAAC,mBAAmB,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;QACnD,CAAC;QAED,OAAO;YACL,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,EAAE,EAAE,QAAQ,CAAC,EAAE;YACf,MAAM,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM;YAC9B,QAAQ,EAAE,CAAC,GAAG,QAAQ,CAAC,QAAQ,CAAC;YAChC,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,mBAAmB,CAAC;YACpF,UAAU,EAAE,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC;YAChD,SAAS,EAAE,KAAK,CAAC,MAAM;YACvB,gBAAgB;YAChB,mBAAmB;YACnB,UAAU,EAAE,gBAAgB,CAAC,QAAQ,CAAC,MAAM,CAAC;YAC7C,QAAQ,EAAE,QAAQ,CAAC,QAAQ;SAC5B,CAAC;IACJ,CAAC,CAAC,CAAC;IACH,MAAM,YAAY,GAA2B,EAAE,CAAC;IAChD,MAAM,gBAAgB,GAA2B,EAAE,CAAC;IACpD,MAAM,mBAAmB,GAA2B,EAAE,CAAC;IACvD,KAAK,MAAM,QAAQ,IAAI,KAAK,EAAE,CAAC;QAC7B,eAAe,CAAC,YAAY,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;QAC/C,KAAK,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAC,EAAE,CAAC;YACxE,gBAAgB,CAAC,MAAM,CAAC,GAAG,CAAC,gBAAgB,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,GAAG,KAAK,CAAC;QACrE,CAAC;QACD,KAAK,MAAM,CAAC,SAAS,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAC,EAAE,CAAC;YAC9E,mBAAmB,CAAC,SAAS,CAAC,GAAG,CAAC,mBAAmB,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,GAAG,KAAK,CAAC;QACjF,CAAC;IACH,CAAC;IACD,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,MAAM,CAAC;IAC9D,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,GAAG,MAAM,CAAC;IAErC,OAAO;QACL,EAAE,EAAE,MAAM,CAAC,EAAE;QACb,KAAK,EAAE,KAAK,CAAC,MAAM;QACnB,MAAM;QACN,MAAM;QACN,QAAQ,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QAClD,YAAY;QACZ,SAAS,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,QAAQ,EAAE,EAAE,CAAC,KAAK,GAAG,QAAQ,CAAC,SAAS,EAAE,CAAC,CAAC;QAC3E,gBAAgB;QAChB,mBAAmB;QACnB,QAAQ,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;YAChF,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,QAAQ,EAAE,QAAQ,CAAC,QAAQ;SAC5B,CAAC,CAAC;QACH,KAAK;QACL,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,QAAQ,EAAE,OAAO,CAAC,QAAQ;KAC3B,CAAC;AACJ,CAAC,CAAC;AAEF,MAAM,kBAAkB,GAAG,CAAC,IAAY,EAAiC,EAAE;IACzE,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAA2C,CAAC;QAC1E,OAAO;YACL,KAAK,EAAE,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACpF,QAAQ,EAAE,MAAM,CAAC,QAAQ;SAC1B,CAAC;IACJ,CAAC;IAAC,MAAM,CAAC;QACP,OAAO;YACL,KAAK,EAAE,CAAC;YACR,QAAQ,EAAE,IAAI;SACf,CAAC;IACJ,CAAC;AACH,CAAC,CAAC;AAEF,MAAM,CAAC,MAAM,uBAAuB,GAAG,KAAK,EAC1C,MAAgC,EAChC,KAA8B,EACU,EAAE;IAC1C,IAAI,OAAO,KAAK,KAAK,UAAU,EAAE,CAAC;QAChC,OAAO,KAAK,CAAC,MAAM,CAAC,CAAC;IACvB,CAAC;IAED,MAAM,MAAM,GACV,KAAK,CAAC,MAAM;QACZ,yFAAyF,CAAC;IAC5F,MAAM,QAAQ,GAAG,MAAM,YAAY,CAAC;QAClC,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,QAAQ,EAAE;YACR,iBAAiB,CAAC,MAAM,EAAE,GAAG,MAAM,OAAO,IAAI,CAAC,SAAS,CAAC,kBAAkB,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC;SACjG;KACF,CAAC,CAAC;IAEH,OAAO,kBAAkB,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;AAC3C,CAAC,CAAC"}
|
|
@@ -0,0 +1,67 @@
|
|
|
1
|
+
import type { PostgresClientLike, SqliteDatabaseLike } from "./types.js";
|
|
2
|
+
import { type PersistedWorkflowRunState, type WorkflowStatus } from "./workflow.js";
|
|
3
|
+
export declare const WORKFLOW_STATE_RECORD_SCHEMA_VERSION: 1;
|
|
4
|
+
export interface WorkflowStateLookup {
|
|
5
|
+
appName: string;
|
|
6
|
+
userId: string;
|
|
7
|
+
sessionId: string;
|
|
8
|
+
workflowKey: string;
|
|
9
|
+
}
|
|
10
|
+
export interface WorkflowStateListInput {
|
|
11
|
+
appName: string;
|
|
12
|
+
userId: string;
|
|
13
|
+
sessionId?: string;
|
|
14
|
+
workflowKey?: string;
|
|
15
|
+
status?: WorkflowStatus;
|
|
16
|
+
}
|
|
17
|
+
export interface WorkflowStateSaveInput extends WorkflowStateLookup {
|
|
18
|
+
state: PersistedWorkflowRunState;
|
|
19
|
+
expectedRevision?: number;
|
|
20
|
+
}
|
|
21
|
+
export interface WorkflowStateRecord extends WorkflowStateLookup {
|
|
22
|
+
schemaVersion: typeof WORKFLOW_STATE_RECORD_SCHEMA_VERSION;
|
|
23
|
+
revision: number;
|
|
24
|
+
state: PersistedWorkflowRunState;
|
|
25
|
+
status: WorkflowStatus;
|
|
26
|
+
runId: string;
|
|
27
|
+
createdAt: number;
|
|
28
|
+
updatedAt: number;
|
|
29
|
+
}
|
|
30
|
+
export interface WorkflowStateService {
|
|
31
|
+
saveWorkflowState(input: WorkflowStateSaveInput): Promise<WorkflowStateRecord> | WorkflowStateRecord;
|
|
32
|
+
loadWorkflowState(input: WorkflowStateLookup): Promise<WorkflowStateRecord | undefined> | WorkflowStateRecord | undefined;
|
|
33
|
+
listWorkflowStates(input: WorkflowStateListInput): Promise<WorkflowStateRecord[]> | WorkflowStateRecord[];
|
|
34
|
+
deleteWorkflowState(input: WorkflowStateLookup): Promise<void> | void;
|
|
35
|
+
}
|
|
36
|
+
export interface FileWorkflowStateServiceOptions {
|
|
37
|
+
directory: string;
|
|
38
|
+
}
|
|
39
|
+
export interface FileWorkflowStateStorePruneOptions extends FileWorkflowStateServiceOptions {
|
|
40
|
+
olderThanMs?: number;
|
|
41
|
+
keepLast?: number;
|
|
42
|
+
now?: number;
|
|
43
|
+
dryRun?: boolean;
|
|
44
|
+
}
|
|
45
|
+
export interface FileWorkflowStateStorePruneResult {
|
|
46
|
+
directory: string;
|
|
47
|
+
dryRun: boolean;
|
|
48
|
+
deletedWorkflowStateKeys: string[];
|
|
49
|
+
keptWorkflowStateKeys: string[];
|
|
50
|
+
}
|
|
51
|
+
export interface SqliteWorkflowStateServiceOptions {
|
|
52
|
+
db: SqliteDatabaseLike;
|
|
53
|
+
tableName?: string;
|
|
54
|
+
}
|
|
55
|
+
export interface PostgresWorkflowStateServiceOptions {
|
|
56
|
+
client: PostgresClientLike;
|
|
57
|
+
tableName?: string;
|
|
58
|
+
}
|
|
59
|
+
export type WorkflowStateRecordMigrationTarget = typeof WORKFLOW_STATE_RECORD_SCHEMA_VERSION;
|
|
60
|
+
export declare const normalizeWorkflowStateRecord: (value: unknown) => WorkflowStateRecord;
|
|
61
|
+
export declare const migrateWorkflowStateRecord: (value: unknown, targetVersion?: WorkflowStateRecordMigrationTarget) => WorkflowStateRecord;
|
|
62
|
+
export declare const createInMemoryWorkflowStateService: () => WorkflowStateService;
|
|
63
|
+
export declare const createFileWorkflowStateService: (options: FileWorkflowStateServiceOptions) => WorkflowStateService;
|
|
64
|
+
export declare const pruneFileWorkflowStateStore: (options: FileWorkflowStateStorePruneOptions) => Promise<FileWorkflowStateStorePruneResult>;
|
|
65
|
+
export declare const createSqliteWorkflowStateService: (options: SqliteWorkflowStateServiceOptions) => WorkflowStateService;
|
|
66
|
+
export declare const createPostgresWorkflowStateService: (options: PostgresWorkflowStateServiceOptions) => WorkflowStateService;
|
|
67
|
+
//# sourceMappingURL=workflow-state-service.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"workflow-state-service.d.ts","sourceRoot":"","sources":["../src/workflow-state-service.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAa,kBAAkB,EAAE,kBAAkB,EAAuB,MAAM,YAAY,CAAC;AACzG,OAAO,EAAgE,KAAK,yBAAyB,EAAE,KAAK,cAAc,EAAE,MAAM,eAAe,CAAC;AAKlJ,eAAO,MAAM,oCAAoC,EAAG,CAAU,CAAC;AAE/D,MAAM,WAAW,mBAAmB;IAClC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,sBAAsB;IACrC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,MAAM,CAAC,EAAE,cAAc,CAAC;CACzB;AAED,MAAM,WAAW,sBAAuB,SAAQ,mBAAmB;IACjE,KAAK,EAAE,yBAAyB,CAAC;IACjC,gBAAgB,CAAC,EAAE,MAAM,CAAC;CAC3B;AAED,MAAM,WAAW,mBAAoB,SAAQ,mBAAmB;IAC9D,aAAa,EAAE,OAAO,oCAAoC,CAAC;IAC3D,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,yBAAyB,CAAC;IACjC,MAAM,EAAE,cAAc,CAAC;IACvB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,oBAAoB;IACnC,iBAAiB,CAAC,KAAK,EAAE,sBAAsB,GAAG,OAAO,CAAC,mBAAmB,CAAC,GAAG,mBAAmB,CAAC;IACrG,iBAAiB,CAAC,KAAK,EAAE,mBAAmB,GAAG,OAAO,CAAC,mBAAmB,GAAG,SAAS,CAAC,GAAG,mBAAmB,GAAG,SAAS,CAAC;IAC1H,kBAAkB,CAAC,KAAK,EAAE,sBAAsB,GAAG,OAAO,CAAC,mBAAmB,EAAE,CAAC,GAAG,mBAAmB,EAAE,CAAC;IAC1G,mBAAmB,CAAC,KAAK,EAAE,mBAAmB,GAAG,OAAO,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC;CACvE;AAED,MAAM,WAAW,+BAA+B;IAC9C,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,kCAAmC,SAAQ,+BAA+B;IACzF,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,OAAO,CAAC;CAClB;AAED,MAAM,WAAW,iCAAiC;IAChD,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,OAAO,CAAC;IAChB,wBAAwB,EAAE,MAAM,EAAE,CAAC;IACnC,qBAAqB,EAAE,MAAM,EAAE,CAAC;CACjC;AAED,MAAM,WAAW,iCAAiC;IAChD,EAAE,EAAE,kBAAkB,CAAC;IACvB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,mCAAmC;IAClD,MAAM,EAAE,kBAAkB,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,MAAM,kCAAkC,GAAG,OAAO,oCAAoC,CAAC;AAQ7F,eAAO,MAAM,4BAA4B,GAAI,OAAO,OAAO,KAAG,mBAkC7D,CAAC;AAEF,eAAO,MAAM,0BAA0B,GACrC,OAAO,OAAO,EACd,gBAAe,kCAAyE,KACvF,mBAKF,CAAC;AA2GF,eAAO,MAAM,kCAAkC,QAAO,oBAwBrD,CAAC;AAEF,eAAO,MAAM,8BAA8B,GAAI,SAAS,+BAA+B,KAAG,oBAwDzF,CAAC;AAEF,eAAO,MAAM,2BAA2B,GACtC,SAAS,kCAAkC,KAC1C,OAAO,CAAC,iCAAiC,CAgD3C,CAAC;AAEF,eAAO,MAAM,gCAAgC,GAAI,SAAS,iCAAiC,KAAG,oBAkI7F,CAAC;AAEF,eAAO,MAAM,kCAAkC,GAAI,SAAS,mCAAmC,KAAG,oBA4HjG,CAAC"}
|