oh-my-knowledge 0.43.0 → 0.45.0
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/dist/artifact-graph/doctor.js +1 -1
- package/dist/artifact-graph/eval.d.ts +16 -0
- package/dist/artifact-graph/eval.js +471 -0
- package/dist/doctor/index.js +25 -10
- package/dist/eval-core/artifact-file-names.d.ts +1 -1
- package/dist/eval-core/artifact-file-names.js +2 -3
- package/dist/eval-core/evaluation-reporting.d.ts +1 -1
- package/dist/eval-core/evaluation-reporting.js +32 -12
- package/dist/inputs/load-samples.js +27 -0
- package/dist/renderer/html-renderer.js +3 -2
- package/dist/renderer/skill-detail-renderer.js +902 -0
- package/dist/renderer/skill-list-renderer.js +2 -7
- package/dist/server/report-server.js +45 -11
- package/dist/server/skill-index.d.ts +6 -0
- package/dist/server/skill-index.js +405 -5
- package/dist/types/artifact-graph.d.ts +1 -1
- package/dist/types/eval.d.ts +10 -0
- package/dist/types/report.d.ts +3 -1
- package/dist/types/skill-index.d.ts +42 -0
- package/package.json +3 -3
|
@@ -42,6 +42,46 @@ export interface SkillObserveSnapshot {
|
|
|
42
42
|
* 历史快照缺此字段时由 segmentCount 兜底推导。 */
|
|
43
43
|
confidence: 'high' | 'low' | 'underpowered';
|
|
44
44
|
}
|
|
45
|
+
export interface SkillGraphStageSnapshot {
|
|
46
|
+
sourceKind: 'doctor' | 'eval' | 'observe';
|
|
47
|
+
sourceId: string;
|
|
48
|
+
graphId: string;
|
|
49
|
+
generatedAt: string;
|
|
50
|
+
graphPath?: string;
|
|
51
|
+
nodeCount: number;
|
|
52
|
+
edgeCount: number;
|
|
53
|
+
}
|
|
54
|
+
export interface SkillGraphNodePreview {
|
|
55
|
+
stableKey?: string;
|
|
56
|
+
nodeKind: string;
|
|
57
|
+
label: string;
|
|
58
|
+
status?: string;
|
|
59
|
+
coverage?: 'declared' | 'undeclared';
|
|
60
|
+
}
|
|
61
|
+
export interface SkillGraphSnapshot {
|
|
62
|
+
/** Studio 聚合 graph sidecar 时实际采用的绑定强度。 */
|
|
63
|
+
bindingStrength: 'content-hash' | 'source-locator' | 'name-only' | 'mixed';
|
|
64
|
+
artifactHash?: string;
|
|
65
|
+
sourceLocator?: string;
|
|
66
|
+
doctor?: SkillGraphStageSnapshot & {
|
|
67
|
+
references: number;
|
|
68
|
+
scripts: number;
|
|
69
|
+
workflows: number;
|
|
70
|
+
workflowNodes: number;
|
|
71
|
+
hardRules: number;
|
|
72
|
+
definitionNodes: SkillGraphNodePreview[];
|
|
73
|
+
};
|
|
74
|
+
eval?: SkillGraphStageSnapshot & {
|
|
75
|
+
variantName?: string;
|
|
76
|
+
samples: number;
|
|
77
|
+
assertions: number;
|
|
78
|
+
failedAssertionEdges: number;
|
|
79
|
+
diagnostics: number;
|
|
80
|
+
measurementNodes: SkillGraphNodePreview[];
|
|
81
|
+
coverageEdges: number;
|
|
82
|
+
declaredCoverageStableKeys: string[];
|
|
83
|
+
};
|
|
84
|
+
}
|
|
45
85
|
export interface SkillIndexEntry {
|
|
46
86
|
skillName: string;
|
|
47
87
|
/** 当前(最新)snapshot — 等价于对应 history 的最后一项,空时为 null。renderer
|
|
@@ -56,6 +96,8 @@ export interface SkillIndexEntry {
|
|
|
56
96
|
/** 综合健康灯。doctor / eval / observe 任一红 → red;任一黄 → yellow;
|
|
57
97
|
* 全绿 → green;皆未跑 → gray。 */
|
|
58
98
|
band: 'green' | 'yellow' | 'red' | 'gray';
|
|
99
|
+
/** doctor/eval/observe graph sidecar 的轻量 Studio 投影。 */
|
|
100
|
+
graph?: SkillGraphSnapshot;
|
|
59
101
|
}
|
|
60
102
|
export interface SkillIndexSummary {
|
|
61
103
|
totalSkills: number;
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "oh-my-knowledge",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.45.0",
|
|
4
4
|
"packageManager": "yarn@4.16.0",
|
|
5
5
|
"description": "Evaluation framework for LLM knowledge inputs — prompts, RAG corpora, skills, agent workflows. Fix the model, vary the artifact. Built-in statistical rigor: bootstrap CI, Krippendorff α, length-debias, saturation curves.",
|
|
6
6
|
"type": "module",
|
|
@@ -92,11 +92,11 @@
|
|
|
92
92
|
"license": "MIT",
|
|
93
93
|
"dependencies": {
|
|
94
94
|
"@anthropic-ai/claude-agent-sdk": "^0.3.143",
|
|
95
|
-
"@anthropic-ai/sdk": "^0.
|
|
95
|
+
"@anthropic-ai/sdk": "^0.105.0",
|
|
96
96
|
"@inquirer/prompts": "^8.4.3",
|
|
97
97
|
"@modelcontextprotocol/sdk": "^1.29.0",
|
|
98
98
|
"@oclif/core": "^4",
|
|
99
|
-
"@openai/codex-sdk": "0.
|
|
99
|
+
"@openai/codex-sdk": "0.141.0",
|
|
100
100
|
"ajv": "^8.18.0",
|
|
101
101
|
"chart.js": "^4.5.1",
|
|
102
102
|
"js-yaml": "^4.1.1",
|