oh-my-knowledge 0.26.0 → 0.27.0

package/dist/src/analysis/report-diagnostics.d.ts

@@ -26,7 +26,7 @@ export declare function analyzeResults(report: Report, opts?: AnalyzeResultsOpti
  * Missing fields are bucketed under the `unspecified` key in the relevant
  * distribution map, so users see "I have N samples without difficulty declared".
  *
- * Used by `bench diagnose` CLI to surface coverage gaps. Does NOT participate
+ * Used by `omk improve` to surface coverage gaps. Does NOT participate
  * in grading / judge / verdict. See docs/sample-design-spec.md.
  */
 export declare function buildSampleQualityAggregate(samples: Sample[]): SampleQualityAggregate;

package/dist/src/analysis/report-diagnostics.js

@@ -56,7 +56,7 @@ export function analyzeResults(report, opts = {}) {
  * Missing fields are bucketed under the `unspecified` key in the relevant
  * distribution map, so users see "I have N samples without difficulty declared".
  *
- * Used by `bench diagnose` CLI to surface coverage gaps. Does NOT participate
+ * Used by `omk improve` to surface coverage gaps. Does NOT participate
  * in grading / judge / verdict. See docs/sample-design-spec.md.
  */
 export function buildSampleQualityAggregate(samples) {

package/dist/src/analysis/sample-diagnostics.d.ts

@@ -16,7 +16,7 @@
  * (d) are cost / latency outliers (one sample spikes the budget by itself), or
  * (e) fail across the board (likely broken sample, not a model deficiency).
  *
- * The CLI `omk bench diagnose <reportId>` surfaces these so users can iterate
+ * The CLI `omk improve <reportId>` surfaces these so users can iterate
  * the SAMPLE SET as much as they iterate the SKILL.
  *
  * Output shape: a flat list of issues, each tagged with severity and the
@@ -75,7 +75,7 @@ export declare function normalizeCapability(raw: string): string;
 type DiagnosticLang = 'zh' | 'en';
 export declare function formatSampleIssue(issue: SampleIssue, lang?: DiagnosticLang): string;
 /**
- * Plain-text formatter for `omk bench diagnose <reportId>`.
+ * Plain-text formatter for `omk improve <reportId>`.
  */
 export declare function formatSampleDiagnostics(diag: SampleDiagnosticReport, options?: {
     topN?: number;

package/dist/src/analysis/sample-diagnostics.js

@@ -16,7 +16,7 @@
  * (d) are cost / latency outliers (one sample spikes the budget by itself), or
  * (e) fail across the board (likely broken sample, not a model deficiency).
  *
- * The CLI `omk bench diagnose <reportId>` surfaces these so users can iterate
+ * The CLI `omk improve <reportId>` surfaces these so users can iterate
  * the SAMPLE SET as much as they iterate the SKILL.
  *
  * Output shape: a flat list of issues, each tagged with severity and the
@@ -376,7 +376,7 @@ export function formatSampleIssue(issue, lang = 'zh') {
     }
 }
 /**
- * Plain-text formatter for `omk bench diagnose <reportId>`.
+ * Plain-text formatter for `omk improve <reportId>`.
  */
 export function formatSampleDiagnostics(diag, options = {}) {
     const lines = [];

package/dist/src/authoring/generator.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"generator.d.ts","sourceRoot":"","sources":["../../../src/authoring/generator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,mBAAmB,CAAC;AA8BhD,UAAU,sBAAsB;IAC9B,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,YAAY,CAAC,EAAE,MAAM,CAAC;CACvB;AAED,wBAAsB,eAAe,CAAC,EAAE,YAAY,EAAE,KAAS,EAAE,KAAqB,EAAE,YAAuB,EAAE,EAAE,sBAAsB,GAAG,OAAO,CAAC;IAAE,OAAO,EAAE,MAAM,EAAE,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC,CA2C1L;AAED;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,MAAM,EAAE,GAAG;IAAE,QAAQ,EAAE,MAAM,EAAE,CAAA;CAAE,CAoClF"}
+{"version":3,"file":"generator.d.ts","sourceRoot":"","sources":["../../../src/authoring/generator.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,mBAAmB,CAAC;AA+BhD,UAAU,sBAAsB;IAC9B,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,YAAY,CAAC,EAAE,MAAM,CAAC;CACvB;AAED,wBAAsB,eAAe,CAAC,EAAE,YAAY,EAAE,KAAS,EAAE,KAAqB,EAAE,YAAuB,EAAE,EAAE,sBAAsB,GAAG,OAAO,CAAC;IAAE,OAAO,EAAE,MAAM,EAAE,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC,CA2C1L;AAED;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,wBAAgB,wBAAwB,CAAC,OAAO,EAAE,MAAM,EAAE,GAAG;IAAE,QAAQ,EAAE,MAAM,EAAE,CAAA;CAAE,CAoClF"}

package/dist/src/authoring/generator.js

@@ -1,5 +1,5 @@
 import { createExecutor, DEFAULT_MODEL } from '../executors/index.js';
-const SYSTEM_PROMPT = `你是一个评测用例生成器。你的任务是根据用户提供的 skill（系统提示词）内容，生成高质量的测试用例。
+const SYSTEM_PROMPT = `你是一个评测用例生成器。你的任务是根据用户提供的 skill（系统提示词）内容，生成高质量的评测用例。
 
 每个用例需包含以下字段：
 - sample_id: 唯一标识，格式为 s001, s002, ...
@@ -7,16 +7,17 @@ const SYSTEM_PROMPT = `你是一个评测用例生成器。你的任务是根据
 - context: 可选，附加上下文信息（如代码片段、文档段落等），仅在需要时提供
 - rubric: 评分标准，描述一个好的回答应该具备什么特征（1-2 句话）
 - assertions: 2-3 个断言检查，可选类型：
-  - { "type": "contains", "value": "关键词", "weight": 1 }
-  - { "type": "not_contains", "value": "不应出现的内容", "weight": 0.5 }
-  - { "type": "regex", "pattern": "正则表达式", "weight": 1 }
+  - { "type": "contains", "value": "English keyword or code token", "weight": 1 }
+  - { "type": "not_contains", "value": "English phrase that should not appear", "weight": 0.5 }
+  - { "type": "regex", "pattern": "English|code|token", "weight": 1 }
 
 要求：
-1. 测试用例应覆盖 skill 的不同能力维度
+1. 评测用例应覆盖 skill 的不同能力维度
 2. prompt 要贴近真实用户的使用场景
 3. rubric 要具体，不要泛泛而谈
 4. assertions 要有区分度，能检测出有无 skill 的差异
-5. 断言应检测 skill 文档中的具体细节（如特定参数名、配置值、工作流步骤），而非通用知识。
+5. assertions 的 value / pattern / values / reference 必须使用英文、数字或代码 token，不要使用中文关键词。
+6. 断言应检测 skill 文档中的具体细节（如特定参数名、配置值、工作流步骤），而非通用知识。
    避免使用 baseline 凭常识或搜索文件也能答对的断言（如 not_contains 通用错误写法）。
    优先使用 contains 检测文档独有的术语、参数组合或特定值
 
@@ -32,7 +33,7 @@ export async function generateSamples({ skillContent, count = 5, model = DEFAULT
 
 ${skillContent}
 
-请根据这个 skill 生成 ${count} 个测试用例。直接输出 JSON 数组。`;
+请根据这个 skill 生成 ${count} 个评测用例。直接输出 JSON 数组。`;
     const result = await executor({ model, system: SYSTEM_PROMPT, prompt });
     if (!result.ok) {
         throw new Error(`generation failed: ${result.error || 'unknown error'}`);
@@ -57,7 +58,7 @@ ${skillContent}
     // (see sanitizeGeneratedSamples).
     const { stripped } = sanitizeGeneratedSamples(samples);
     if (stripped.length > 0) {
-        process.stderr.write(`[omk gen-samples] LLM-output 含 ${stripped.length} 个非法元数据字段, 已剥离避免污染:\n  - ${stripped.join('\n  - ')}\n`);
+        process.stderr.write(`[omk improve samples] LLM-output 含 ${stripped.length} 个非法元数据字段，已剥离避免污染：\n  - ${stripped.join('\n  - ')}\n`);
     }
     return { samples, costUSD: result.costUSD };
 }

package/dist/src/authoring/generator.js.map

@@ -1 +1 @@
-{"version":3,"file":"generator.js","sourceRoot":"","sources":["../../../src/authoring/generator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAGtE,MAAM,aAAa,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;uCA0BiB,CAAC;AASxC,MAAM,CAAC,KAAK,UAAU,eAAe,CAAC,EAAE,YAAY,EAAE,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,aAAa,EAAE,YAAY,GAAG,QAAQ,EAA0B;IACvI,MAAM,QAAQ,GAAG,cAAc,CAAC,YAAY,CAAC,CAAC;IAE9C,MAAM,MAAM,GAAG;;EAEf,YAAY;;iBAEG,KAAK,sBAAsB,CAAC;IAE3C,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,EAAE,CAAC,CAAC;IAExE,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,sBAAsB,MAAM,CAAC,KAAK,IAAI,eAAe,EAAE,CAAC,CAAC;IAC3E,CAAC;IAED,kEAAkE;IAClE,IAAI,OAAO,GAAG,MAAM,CAAC,MAAO,CAAC,IAAI,EAAE,CAAC;IACpC,MAAM,SAAS,GAAG,OAAO,CAAC,KAAK,CAAC,8BAA8B,CAAC,CAAC;IAChE,IAAI,SAAS,EAAE,CAAC;QACd,OAAO,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAChC,CAAC;IAED,IAAI,OAAiB,CAAC;IACtB,IAAI,CAAC;QACH,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;IAChC,CAAC;IAAC,MAAM,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,mDAAmD,CAAC,CAAC;IACvE,CAAC;IAED,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpD,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;IAC7D,CAAC;IAED,8EAA8E;IAC9E,kCAAkC;IAClC,MAAM,EAAE,QAAQ,EAAE,GAAG,wBAAwB,CAAC,OAAO,CAAC,CAAC;IACvD,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,kCAAkC,QAAQ,CAAC,MAAM,4BAA4B,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CACzG,CAAC;IACJ,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC;AAC9C,CAAC;AAED;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,UAAU,wBAAwB,CAAC,OAAiB;IACxD,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC;IAC7D,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,OAAO,EAAE,eAAe,EAAE,kBAAkB,CAAC,CAAC,CAAC;IACjF,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC;QACvC,kEAAkE;QAClE,0DAA0D;QAC1D,IAAI,OAAO,CAAC,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,CAAC,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAChE,CAAC,CAAC,SAAS,GAAG,IAAI,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC;QACrD,CAAC;QACD,IAAI,OAAO,CAAC,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC1D,MAAM,IAAI,KAAK,CAAC,WAAW,CAAC,mDAAmD,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC;QACrG,CAAC;QAED,IAAI,CAAC,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;YAC/B,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,CAAC;gBACtG,QAAQ,CAAC,IAAI,CAAC,WAAW,CAAC,iBAAiB,OAAO,CAAC,CAAC,UAAU,GAAG,CAAC,CAAC;gBACnE,OAAQ,CAA8B,CAAC,UAAU,CAAC;YACpD,CAAC;QACH,CAAC;QACD,IAAI,CAAC,CAAC,UAAU,KAAK,SAAS,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC,UAAoB,CAAC,EAAE,CAAC;YAChF,QAAQ,CAAC,IAAI,CAAC,WAAW,CAAC,gBAAgB,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,CAAC,EAAE,CAAC,CAAC;YAC1E,OAAQ,CAA8B,CAAC,UAAU,CAAC;QACpD,CAAC;QACD,IAAI,CAAC,CAAC,SAAS,KAAK,SAAS,IAAI,CAAC,OAAO,CAAC,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC,EAAE,CAAC;YACnF,QAAQ,CAAC,IAAI,CAAC,WAAW,CAAC,gBAAgB,OAAO,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC;YACjE,OAAQ,CAA6B,CAAC,SAAS,CAAC;QAClD,CAAC;QACD,IAAI,CAAC,CAAC,UAAU,KAAK,SAAS,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC,UAAoB,CAAC,EAAE,CAAC;YAChF,QAAQ,CAAC,IAAI,CAAC,WAAW,CAAC,gBAAgB,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,CAAC,EAAE,CAAC,CAAC;YAC1E,OAAQ,CAA8B,CAAC,UAAU,CAAC;QACpD,CAAC;QACD,kFAAkF;QAClF,IAAI,CAAC,CAAC,CAAC,UAAU;YAAE,CAAC,CAAC,UAAU,GAAG,eAAe,CAAC;IACpD,CAAC;IACD,OAAO,EAAE,QAAQ,EAAE,CAAC;AACtB,CAAC"}
+{"version":3,"file":"generator.js","sourceRoot":"","sources":["../../../src/authoring/generator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AAGtE,MAAM,aAAa,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;uCA2BiB,CAAC;AASxC,MAAM,CAAC,KAAK,UAAU,eAAe,CAAC,EAAE,YAAY,EAAE,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,aAAa,EAAE,YAAY,GAAG,QAAQ,EAA0B;IACvI,MAAM,QAAQ,GAAG,cAAc,CAAC,YAAY,CAAC,CAAC;IAE9C,MAAM,MAAM,GAAG;;EAEf,YAAY;;iBAEG,KAAK,sBAAsB,CAAC;IAE3C,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,EAAE,CAAC,CAAC;IAExE,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;QACf,MAAM,IAAI,KAAK,CAAC,sBAAsB,MAAM,CAAC,KAAK,IAAI,eAAe,EAAE,CAAC,CAAC;IAC3E,CAAC;IAED,kEAAkE;IAClE,IAAI,OAAO,GAAG,MAAM,CAAC,MAAO,CAAC,IAAI,EAAE,CAAC;IACpC,MAAM,SAAS,GAAG,OAAO,CAAC,KAAK,CAAC,8BAA8B,CAAC,CAAC;IAChE,IAAI,SAAS,EAAE,CAAC;QACd,OAAO,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAChC,CAAC;IAED,IAAI,OAAiB,CAAC;IACtB,IAAI,CAAC;QACH,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;IAChC,CAAC;IAAC,MAAM,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,mDAAmD,CAAC,CAAC;IACvE,CAAC;IAED,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpD,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;IAC7D,CAAC;IAED,8EAA8E;IAC9E,kCAAkC;IAClC,MAAM,EAAE,QAAQ,EAAE,GAAG,wBAAwB,CAAC,OAAO,CAAC,CAAC;IACvD,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,sCAAsC,QAAQ,CAAC,MAAM,2BAA2B,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAC5G,CAAC;IACJ,CAAC;IAED,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC;AAC9C,CAAC;AAED;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,UAAU,wBAAwB,CAAC,OAAiB;IACxD,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,MAAM,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC;IAC7D,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,OAAO,EAAE,eAAe,EAAE,kBAAkB,CAAC,CAAC,CAAC;IACjF,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC;QACvC,kEAAkE;QAClE,0DAA0D;QAC1D,IAAI,OAAO,CAAC,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,CAAC,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAChE,CAAC,CAAC,SAAS,GAAG,IAAI,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC;QACrD,CAAC;QACD,IAAI,OAAO,CAAC,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC1D,MAAM,IAAI,KAAK,CAAC,WAAW,CAAC,mDAAmD,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC;QACrG,CAAC;QAED,IAAI,CAAC,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;YAC/B,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,EAAE,CAAC;gBACtG,QAAQ,CAAC,IAAI,CAAC,WAAW,CAAC,iBAAiB,OAAO,CAAC,CAAC,UAAU,GAAG,CAAC,CAAC;gBACnE,OAAQ,CAA8B,CAAC,UAAU,CAAC;YACpD,CAAC;QACH,CAAC;QACD,IAAI,CAAC,CAAC,UAAU,KAAK,SAAS,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC,UAAoB,CAAC,EAAE,CAAC;YAChF,QAAQ,CAAC,IAAI,CAAC,WAAW,CAAC,gBAAgB,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,CAAC,EAAE,CAAC,CAAC;YAC1E,OAAQ,CAA8B,CAAC,UAAU,CAAC;QACpD,CAAC;QACD,IAAI,CAAC,CAAC,SAAS,KAAK,SAAS,IAAI,CAAC,OAAO,CAAC,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC,EAAE,CAAC;YACnF,QAAQ,CAAC,IAAI,CAAC,WAAW,CAAC,gBAAgB,OAAO,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC;YACjE,OAAQ,CAA6B,CAAC,SAAS,CAAC;QAClD,CAAC;QACD,IAAI,CAAC,CAAC,UAAU,KAAK,SAAS,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC,UAAoB,CAAC,EAAE,CAAC;YAChF,QAAQ,CAAC,IAAI,CAAC,WAAW,CAAC,gBAAgB,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,CAAC,EAAE,CAAC,CAAC;YAC1E,OAAQ,CAA8B,CAAC,UAAU,CAAC;QACpD,CAAC;QACD,kFAAkF;QAClF,IAAI,CAAC,CAAC,CAAC,UAAU;YAAE,CAAC,CAAC,UAAU,GAAG,eAAe,CAAC;IACpD,CAAC;IACD,OAAO,EAAE,QAAQ,EAAE,CAAC;AACtB,CAAC"}

package/dist/src/cli/cli-exit.d.ts

@@ -0,0 +1,15 @@
+/**
+ * 命令退出信号。execute() / parse-strict / requireEvaluationReport 等想以
+ * 特定 exit code 终止时,throw 这个,不要直接 process.exit。
+ *
+ * dispatcher (`main()` 顶层 catch) 负责把 CliExit 转成 process.exit(code) —
+ * 这样 execute() 可以在单测里被 try/catch 捕获,不 kill 整个测试进程。
+ *
+ * 异步回调内的「子进程 exit code 透传」。
+ * 不走这层,继续 process.exit — 那是子进程退出后的 cleanup,不在 main 调用栈。
+ */
+export declare class CliExit extends Error {
+    readonly code: number;
+    constructor(code: number);
+}
+//# sourceMappingURL=cli-exit.d.ts.map

package/dist/src/cli/cli-exit.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli-exit.d.ts","sourceRoot":"","sources":["../../../src/cli/cli-exit.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AACH,qBAAa,OAAQ,SAAQ,KAAK;aACJ,IAAI,EAAE,MAAM;gBAAZ,IAAI,EAAE,MAAM;CAIzC"}

package/dist/src/cli/cli-exit.js

@@ -0,0 +1,19 @@
+/**
+ * 命令退出信号。execute() / parse-strict / requireEvaluationReport 等想以
+ * 特定 exit code 终止时,throw 这个,不要直接 process.exit。
+ *
+ * dispatcher (`main()` 顶层 catch) 负责把 CliExit 转成 process.exit(code) —
+ * 这样 execute() 可以在单测里被 try/catch 捕获,不 kill 整个测试进程。
+ *
+ * 异步回调内的「子进程 exit code 透传」。
+ * 不走这层,继续 process.exit — 那是子进程退出后的 cleanup,不在 main 调用栈。
+ */
+export class CliExit extends Error {
+    code;
+    constructor(code) {
+        super(`CliExit(${code})`);
+        this.code = code;
+        this.name = 'CliExit';
+    }
+}
+//# sourceMappingURL=cli-exit.js.map

package/dist/src/cli/cli-exit.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli-exit.js","sourceRoot":"","sources":["../../../src/cli/cli-exit.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AACH,MAAM,OAAO,OAAQ,SAAQ,KAAK;IACJ;IAA5B,YAA4B,IAAY;QACtC,KAAK,CAAC,WAAW,IAAI,GAAG,CAAC,CAAC;QADA,SAAI,GAAJ,IAAI,CAAQ;QAEtC,IAAI,CAAC,IAAI,GAAG,SAAS,CAAC;IACxB,CAAC;CACF"}

package/dist/src/cli/commands/_shared.d.ts

@@ -0,0 +1,12 @@
+import { type CliLang } from '../i18n.js';
+import type { EvaluationReport, ReportDocument } from '../../types/index.js';
+export interface EvalResult {
+    report: ReportDocument;
+    filePath: string | null;
+}
+export interface ReportServer {
+    start: () => Promise<string>;
+}
+export declare function requireEvaluationReport(report: ReportDocument | null, id: string, lang: CliLang): EvaluationReport;
+export declare function parseLastWindow(spec: string): string | null;
+//# sourceMappingURL=_shared.d.ts.map

package/dist/src/cli/commands/_shared.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"_shared.d.ts","sourceRoot":"","sources":["../../../../src/cli/commands/_shared.ts"],"names":[],"mappings":"AACA,OAAO,EAAQ,KAAK,OAAO,EAAE,MAAM,YAAY,CAAC;AAChD,OAAO,KAAK,EAAE,gBAAgB,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAE7E,MAAM,WAAW,UAAU;IACzB,MAAM,EAAE,cAAc,CAAC;IACvB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;CACzB;AAED,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,MAAM,OAAO,CAAC,MAAM,CAAC,CAAC;CAC9B;AAED,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,cAAc,GAAG,IAAI,EAAE,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,GAAG,gBAAgB,CAYlH;AAED,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAQ3D"}

package/dist/src/cli/commands/_shared.js

@@ -0,0 +1,26 @@
+import { CliExit } from '../cli-exit.js';
+import { tCli } from '../i18n.js';
+export function requireEvaluationReport(report, id, lang) {
+    if (!report) {
+        console.error(tCli('cli.common.report_not_found', lang, { id }));
+        throw new CliExit(1);
+    }
+    if (report.kind === 'batch-evaluation') {
+        console.error(lang === 'zh'
+            ? `报告 ${id} 是 BatchEvaluationReport。该命令需要单次 EvaluationReport；请使用其中的 child reportId。`
+            : `Report ${id} is a BatchEvaluationReport. This command requires an EvaluationReport; use a child reportId from the batch.`);
+        throw new CliExit(1);
+    }
+    return report;
+}
+export function parseLastWindow(spec) {
+    // "7d" / "24h" / "30m" → ISO timestamp (from = now - spec)
+    const m = /^(\d+)([dhm])$/.exec(spec);
+    if (!m)
+        return null;
+    const n = Number(m[1]);
+    const unit = m[2];
+    const ms = unit === 'd' ? n * 86400_000 : unit === 'h' ? n * 3600_000 : n * 60_000;
+    return new Date(Date.now() - ms).toISOString();
+}
+//# sourceMappingURL=_shared.js.map

package/dist/src/cli/commands/_shared.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"_shared.js","sourceRoot":"","sources":["../../../../src/cli/commands/_shared.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AACzC,OAAO,EAAE,IAAI,EAAgB,MAAM,YAAY,CAAC;AAYhD,MAAM,UAAU,uBAAuB,CAAC,MAA6B,EAAE,EAAU,EAAE,IAAa;IAC9F,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,6BAA6B,EAAE,IAAI,EAAE,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC;QACjE,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IACD,IAAI,MAAM,CAAC,IAAI,KAAK,kBAAkB,EAAE,CAAC;QACvC,OAAO,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI;YACzB,CAAC,CAAC,MAAM,EAAE,0EAA0E;YACpF,CAAC,CAAC,UAAU,EAAE,8GAA8G,CAAC,CAAC;QAChI,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,IAAY;IAC1C,2DAA2D;IAC3D,MAAM,CAAC,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACtC,IAAI,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACpB,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACvB,MAAM,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IAClB,MAAM,EAAE,GAAG,IAAI,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,SAAS,CAAC,CAAC,CAAC,IAAI,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC;IACnF,OAAO,IAAI,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;AACjD,CAAC"}

package/dist/src/cli/commands/doctor.d.ts

@@ -0,0 +1,2 @@
+export declare function execute(argv: string[]): Promise<void>;
+//# sourceMappingURL=doctor.d.ts.map

package/dist/src/cli/commands/doctor.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"doctor.d.ts","sourceRoot":"","sources":["../../../../src/cli/commands/doctor.ts"],"names":[],"mappings":"AAoDA,wBAAsB,OAAO,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAwF3D"}

package/dist/src/cli/commands/doctor.js

@@ -0,0 +1,137 @@
+import { existsSync, statSync } from 'node:fs';
+import { dirname, join, resolve } from 'node:path';
+import { CliExit } from '../cli-exit.js';
+import { tCli, langFromArgv } from '../i18n.js';
+import { COMMON_OPTIONS } from '../parse-run-config.js';
+import { parseArgsStrictOrExit } from '../parse-strict.js';
+const DEFAULT_SAMPLE_FILENAMES = ['eval-samples.json', 'eval-samples.yaml', 'eval-samples.yml'];
+function findSamplesInDir(dir) {
+    for (const name of DEFAULT_SAMPLE_FILENAMES) {
+        const candidate = join(dir, name);
+        if (existsSync(candidate))
+            return candidate;
+    }
+    return null;
+}
+function sampleSearchDirs(target, cwd) {
+    const dirs = [];
+    const add = (dir) => {
+        const abs = resolve(dir);
+        if (!dirs.includes(abs))
+            dirs.push(abs);
+    };
+    if (target) {
+        const absTarget = resolve(target);
+        if (existsSync(absTarget)) {
+            const stat = statSync(absTarget);
+            if (stat.isDirectory()) {
+                add(absTarget);
+                add(dirname(absTarget));
+                add(dirname(dirname(absTarget)));
+            }
+            else {
+                const parent = dirname(absTarget);
+                add(parent);
+                add(dirname(parent));
+            }
+        }
+    }
+    add(cwd);
+    return dirs;
+}
+function findDefaultSamplesPath(target, cwd) {
+    for (const dir of sampleSearchDirs(target, cwd)) {
+        const samplesPath = findSamplesInDir(dir);
+        if (samplesPath)
+            return samplesPath;
+    }
+    return null;
+}
+export async function execute(argv) {
+    const lang = langFromArgv(argv);
+    const { values, positionals } = parseArgsStrictOrExit({
+        args: argv,
+        allowPositionals: true,
+        options: {
+            ...COMMON_OPTIONS,
+            json: { type: 'boolean', default: false },
+            gate: { type: 'boolean', default: false },
+            executor: { type: 'string' },
+            model: { type: 'string' },
+            samples: { type: 'string' },
+            timeout: { type: 'string' },
+        },
+    });
+    const target = positionals[0] ?? null;
+    const executorName = values.executor ?? 'claude';
+    const model = values.model ?? 'sonnet';
+    const timeoutRaw = values.timeout;
+    const timeoutSec = timeoutRaw != null ? Number(timeoutRaw) : 8;
+    const timeoutMs = Math.max(1000, Math.floor((Number.isFinite(timeoutSec) ? timeoutSec : 8) * 1000));
+    const cwd = process.cwd();
+    const samplesPath = values.samples ? resolve(values.samples) : findDefaultSamplesPath(target, cwd);
+    let samples;
+    let requires;
+    if (samplesPath) {
+        try {
+            const { loadSamples } = await import('../../inputs/load-samples.js');
+            const loaded = loadSamples(samplesPath);
+            samples = loaded.samples;
+            requires = loaded.requires;
+        }
+        catch (err) {
+            const msg = err instanceof Error ? err.message : String(err);
+            process.stderr.write(tCli('cli.common.warn_load_samples_failed', lang, { path: samplesPath, message: msg }));
+        }
+    }
+    const { runDoctor } = await import('../../doctor/index.js');
+    const { renderDoctorReportText, renderDoctorReportJson } = await import('../../doctor/renderer.js');
+    let report;
+    try {
+        report = await runDoctor({
+            target,
+            cwd,
+            executorName,
+            model,
+            timeoutMs,
+            lang,
+            samples,
+            requires,
+        });
+    }
+    catch (err) {
+        // CliExit 透传(防御性,目前 runDoctor 不抛 CliExit,但保持四个 catch 一致)。
+        if (err instanceof CliExit)
+            throw err;
+        const msg = err instanceof Error ? err.message : String(err);
+        console.error(tCli('cli.doctor.no_skill_found', lang, { path: target ?? cwd }));
+        console.error(`(${msg})`);
+        throw new CliExit(1);
+    }
+    if (report.skills.length === 0) {
+        console.error(tCli('cli.doctor.no_skill_found', lang, { path: target ?? cwd }));
+        throw new CliExit(1);
+    }
+    const isJson = values.json;
+    const isGate = values.gate;
+    if (isJson) {
+        console.log(renderDoctorReportJson(report));
+    }
+    else if (isGate) {
+        // gate 模式: 静默 stdout, fail 时简短 stderr 摘要(供 CI 抓 exit code)
+        if (report.outcome === 'failed') {
+            const summary = lang === 'zh'
+                ? `doctor failed: ${report.totals.fail} 个 skill 未通过 (${report.totals.warn} warn / ${report.totals.pass} pass)`
+                : `doctor failed: ${report.totals.fail} skills did not pass (${report.totals.warn} warn / ${report.totals.pass} pass)`;
+            console.error(summary);
+        }
+    }
+    else {
+        if (samplesPath && samples) {
+            process.stderr.write(tCli('cli.doctor.samples_detected', lang, { path: samplesPath }) + '\n');
+        }
+        renderDoctorReportText(report, lang);
+    }
+    throw new CliExit(report.outcome === 'failed' ? 1 : 0);
+}
+//# sourceMappingURL=doctor.js.map

package/dist/src/cli/commands/doctor.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"doctor.js","sourceRoot":"","sources":["../../../../src/cli/commands/doctor.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AAC/C,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACnD,OAAO,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AACzC,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAChD,OAAO,EAAE,cAAc,EAAE,MAAM,wBAAwB,CAAC;AACxD,OAAO,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAI3D,MAAM,wBAAwB,GAAG,CAAC,mBAAmB,EAAE,mBAAmB,EAAE,kBAAkB,CAAU,CAAC;AAEzG,SAAS,gBAAgB,CAAC,GAAW;IACnC,KAAK,MAAM,IAAI,IAAI,wBAAwB,EAAE,CAAC;QAC5C,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;QAClC,IAAI,UAAU,CAAC,SAAS,CAAC;YAAE,OAAO,SAAS,CAAC;IAC9C,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,SAAS,gBAAgB,CAAC,MAAqB,EAAE,GAAW;IAC1D,MAAM,IAAI,GAAa,EAAE,CAAC;IAC1B,MAAM,GAAG,GAAG,CAAC,GAAW,EAAQ,EAAE;QAChC,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC;QACzB,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC;YAAE,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC1C,CAAC,CAAC;IACF,IAAI,MAAM,EAAE,CAAC;QACX,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;QAClC,IAAI,UAAU,CAAC,SAAS,CAAC,EAAE,CAAC;YAC1B,MAAM,IAAI,GAAG,QAAQ,CAAC,SAAS,CAAC,CAAC;YACjC,IAAI,IAAI,CAAC,WAAW,EAAE,EAAE,CAAC;gBACvB,GAAG,CAAC,SAAS,CAAC,CAAC;gBACf,GAAG,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC;gBACxB,GAAG,CAAC,OAAO,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC;YACnC,CAAC;iBAAM,CAAC;gBACN,MAAM,MAAM,GAAG,OAAO,CAAC,SAAS,CAAC,CAAC;gBAClC,GAAG,CAAC,MAAM,CAAC,CAAC;gBACZ,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC;YACvB,CAAC;QACH,CAAC;IACH,CAAC;IACD,GAAG,CAAC,GAAG,CAAC,CAAC;IACT,OAAO,IAAI,CAAC;AACd,CAAC;AAED,SAAS,sBAAsB,CAAC,MAAqB,EAAE,GAAW;IAChE,KAAK,MAAM,GAAG,IAAI,gBAAgB,CAAC,MAAM,EAAE,GAAG,CAAC,EAAE,CAAC;QAChD,MAAM,WAAW,GAAG,gBAAgB,CAAC,GAAG,CAAC,CAAC;QAC1C,IAAI,WAAW;YAAE,OAAO,WAAW,CAAC;IACtC,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,OAAO,CAAC,IAAc;IAC1C,MAAM,IAAI,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;IAChC,MAAM,EAAE,MAAM,EAAE,WAAW,EAAE,GAAG,qBAAqB,CAAC;QACpD,IAAI,EAAE,IAAI;QACV,gBAAgB,EAAE,IAAI;QACtB,OAAO,EAAE;YACP,GAAG,cAAc;YACjB,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE;YACzC,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE;YACzC,QAAQ,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;YAC5B,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;YACzB,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;YAC3B,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;SAC5B;KACF,CAAC,CAAC;IAEH,MAAM,MAAM,GAAkB,WAAW,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC;IACrD,MAAM,YAAY,GAAI,MAAM,CAAC,QAA+B,IAAI,QAAQ,CAAC;IACzE,MAAM,KAAK,GAAI,MAAM,CAAC,KAA4B,IAAI,QAAQ,CAAC;IAC/D,MAAM,UAAU,GAAG,MAAM,CAAC,OAA6B,CAAC;IACxD,MAAM,UAAU,GAAG,UAAU,IAAI,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC/D,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC;IACpG,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,EAAE,CAAC;IAC1B,MAAM,WAAW,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,OAAiB,CAAC,CAAC,CAAC,CAAC,sBAAsB,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC7G,IAAI,OAA6B,CAAC;IAClC,IAAI,QAA4C,CAAC;IACjD,IAAI,WAAW,EAAE,CAAC;QAChB,IAAI,CAAC;YACH,MAAM,EAAE,WAAW,EAAE,GAAG,MAAM,MAAM,CAAC,8BAA8B,CAAC,CAAC;YACrE,MAAM,MAAM,GAAG,WAAW,CAAC,WAAW,CAAC,CAAC;YACxC,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;YACzB,QAAQ,GAAG,MAAM,CAAC,QAAQ,CAAC;QAC7B,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC7D,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,qCAAqC,EAAE,IAAI,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC;QAC/G,CAAC;IACH,CAAC;IAED,MAAM,EAAE,SAAS,EAAE,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,CAAC;IAC5D,MAAM,EAAE,sBAAsB,EAAE,sBAAsB,EAAE,GAAG,MAAM,MAAM,CAAC,0BAA0B,CAAC,CAAC;IAEpG,IAAI,MAAM,CAAC;IACX,IAAI,CAAC;QACH,MAAM,GAAG,MAAM,SAAS,CAAC;YACvB,MAAM;YACN,GAAG;YACH,YAAY;YACZ,KAAK;YACL,SAAS;YACT,IAAI;YACJ,OAAO;YACP,QAAQ;SACT,CAAC,CAAC;IACL,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,0DAA0D;QAC1D,IAAI,GAAG,YAAY,OAAO;YAAE,MAAM,GAAG,CAAC;QACtC,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QAC7D,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,2BAA2B,EAAE,IAAI,EAAE,EAAE,IAAI,EAAE,MAAM,IAAI,GAAG,EAAE,CAAC,CAAC,CAAC;QAChF,OAAO,CAAC,KAAK,CAAC,IAAI,GAAG,GAAG,CAAC,CAAC;QAC1B,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IAED,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC/B,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,2BAA2B,EAAE,IAAI,EAAE,EAAE,IAAI,EAAE,MAAM,IAAI,GAAG,EAAE,CAAC,CAAC,CAAC;QAChF,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IAED,MAAM,MAAM,GAAG,MAAM,CAAC,IAAe,CAAC;IACtC,MAAM,MAAM,GAAG,MAAM,CAAC,IAAe,CAAC;IAEtC,IAAI,MAAM,EAAE,CAAC;QACX,OAAO,CAAC,GAAG,CAAC,sBAAsB,CAAC,MAAM,CAAC,CAAC,CAAC;IAC9C,CAAC;SAAM,IAAI,MAAM,EAAE,CAAC;QAClB,2DAA2D;QAC3D,IAAI,MAAM,CAAC,OAAO,KAAK,QAAQ,EAAE,CAAC;YAChC,MAAM,OAAO,GAAG,IAAI,KAAK,IAAI;gBAC3B,CAAC,CAAC,kBAAkB,MAAM,CAAC,MAAM,CAAC,IAAI,iBAAiB,MAAM,CAAC,MAAM,CAAC,IAAI,WAAW,MAAM,CAAC,MAAM,CAAC,IAAI,QAAQ;gBAC9G,CAAC,CAAC,kBAAkB,MAAM,CAAC,MAAM,CAAC,IAAI,yBAAyB,MAAM,CAAC,MAAM,CAAC,IAAI,WAAW,MAAM,CAAC,MAAM,CAAC,IAAI,QAAQ,CAAC;YACzH,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QACzB,CAAC;IACH,CAAC;SAAM,CAAC;QACN,IAAI,WAAW,IAAI,OAAO,EAAE,CAAC;YAC3B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,6BAA6B,EAAE,IAAI,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;QAChG,CAAC;QACD,sBAAsB,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;IACvC,CAAC;IAED,MAAM,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACzD,CAAC"}

package/dist/src/cli/commands/eval-debias.d.ts

@@ -0,0 +1,2 @@
+export declare function execute(argv: string[]): Promise<void>;
+//# sourceMappingURL=eval-debias.d.ts.map

package/dist/src/cli/commands/eval-debias.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"eval-debias.d.ts","sourceRoot":"","sources":["../../../../src/cli/commands/eval-debias.ts"],"names":[],"mappings":"AAQA,wBAAsB,OAAO,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CAiF3D"}

package/dist/src/cli/commands/eval-debias.js

@@ -0,0 +1,88 @@
+import { resolve } from 'node:path';
+import { CliExit } from '../cli-exit.js';
+import { tCli, langFromArgv } from '../i18n.js';
+import { COMMON_OPTIONS, DEFAULT_REPORTS_DIR } from '../parse-run-config.js';
+import { parseArgsStrictOrExit } from '../parse-strict.js';
+import { requireEvaluationReport } from './_shared.js';
+export async function execute(argv) {
+    const lang = langFromArgv(argv);
+    const [kind, reportId, ...rest] = argv;
+    if (!kind) {
+        console.log(usage(lang));
+        throw new CliExit(1);
+    }
+    if (kind !== 'length') {
+        console.error(`Unknown debias validation kind: ${kind}. Use "length".`);
+        throw new CliExit(1);
+    }
+    if (!reportId) {
+        console.error('Usage: omk eval debias length <reportId>');
+        throw new CliExit(1);
+    }
+    const { values } = parseArgsStrictOrExit({
+        args: rest,
+        options: {
+            ...COMMON_OPTIONS,
+            'reports-dir': { type: 'string', default: DEFAULT_REPORTS_DIR },
+            samples: { type: 'string' },
+            variant: { type: 'string' },
+            'judge-models': { type: 'string' },
+            'bootstrap-samples': { type: 'string', default: '1000' },
+            seed: { type: 'string' },
+        },
+    });
+    const { parseJudgeModelsArgOrExit } = await import('../parse-run-config.js');
+    const cliJudgeModels = values['judge-models'] !== undefined
+        ? parseJudgeModelsArgOrExit(values['judge-models'])
+        : undefined;
+    if (cliJudgeModels && cliJudgeModels.length > 1) {
+        console.error(tCli('cli.common.judge_models_single_only', lang, { cmd: 'eval debias' }));
+        throw new CliExit(2);
+    }
+    const { createFileStore } = await import('../../server/report-store.js');
+    const store = createFileStore(resolve(values['reports-dir']));
+    const report = requireEvaluationReport(await store.get(reportId), reportId, lang);
+    const samplesPath = values.samples ?? report.meta?.request?.samplesPath;
+    if (!samplesPath) {
+        console.error('Cannot find samples path. Pass --samples <path> or ensure report has request.samplesPath.');
+        throw new CliExit(1);
+    }
+    const { loadSamples } = await import('../../inputs/load-samples.js');
+    const { samples } = loadSamples(samplesPath);
+    const debiasJudges = cliJudgeModels
+        ?? (report.meta?.judgeModels?.[0]
+            ? [{ executor: report.meta.judgeModels[0].executor, model: report.meta.judgeModels[0].model }]
+            : []);
+    if (debiasJudges.length === 0) {
+        console.error(tCli('cli.common.no_judge_model', lang));
+        throw new CliExit(1);
+    }
+    process.stderr.write(lang === 'zh'
+        ? '\n⚠ eval debias 会重判所有 sample × variant，judge 成本大约翻倍。\n'
+        : '\n⚠ eval debias will re-judge all sample × variant pairs; judge cost will roughly double.\n');
+    const { createExecutor } = await import('../../executors/index.js');
+    const judgeExecutor = createExecutor(debiasJudges[0].executor);
+    const judgeModel = debiasJudges[0].model;
+    const { validateLengthDebias, formatDebiasValidate } = await import('../../grading/debias-validate.js');
+    const seedVal = values.seed != null ? Number(values.seed) : undefined;
+    const bsRaw = Number(values['bootstrap-samples']) || 1000;
+    const result = await validateLengthDebias({
+        report,
+        samples,
+        judgeExecutor,
+        judgeModel,
+        variant: values.variant,
+        bootstrapSamples: Math.max(100, bsRaw),
+        seed: Number.isFinite(seedVal) ? seedVal : undefined,
+        onProgress: ({ sample_id, completed, total }) => {
+            process.stderr.write(`  judging ${completed}/${total}: ${sample_id}\n`);
+        },
+    });
+    console.log(formatDebiasValidate(result));
+}
+function usage(lang) {
+    return lang === 'zh'
+        ? '用法：omk eval debias length <reportId> [--samples <path>] [--judge-models <executor:model>]'
+        : 'Usage: omk eval debias length <reportId> [--samples <path>] [--judge-models <executor:model>]';
+}
+//# sourceMappingURL=eval-debias.js.map

package/dist/src/cli/commands/eval-debias.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"eval-debias.js","sourceRoot":"","sources":["../../../../src/cli/commands/eval-debias.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,OAAO,EAAE,MAAM,gBAAgB,CAAC;AACzC,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAChD,OAAO,EAAE,cAAc,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AAC7E,OAAO,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAE3D,OAAO,EAAE,uBAAuB,EAAE,MAAM,cAAc,CAAC;AAEvD,MAAM,CAAC,KAAK,UAAU,OAAO,CAAC,IAAc;IAC1C,MAAM,IAAI,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC;IAChC,MAAM,CAAC,IAAI,EAAE,QAAQ,EAAE,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC;IACvC,IAAI,CAAC,IAAI,EAAE,CAAC;QACV,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;QACzB,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IACD,IAAI,IAAI,KAAK,QAAQ,EAAE,CAAC;QACtB,OAAO,CAAC,KAAK,CAAC,mCAAmC,IAAI,iBAAiB,CAAC,CAAC;QACxE,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IACD,IAAI,CAAC,QAAQ,EAAE,CAAC;QACd,OAAO,CAAC,KAAK,CAAC,0CAA0C,CAAC,CAAC;QAC1D,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IAED,MAAM,EAAE,MAAM,EAAE,GAAG,qBAAqB,CAAC;QACvC,IAAI,EAAE,IAAI;QACV,OAAO,EAAE;YACP,GAAG,cAAc;YACjB,aAAa,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,mBAAmB,EAAE;YAC/D,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;YAC3B,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;YAC3B,cAAc,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;YAClC,mBAAmB,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE;YACxD,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;SACzB;KACF,CAAC,CAAC;IAEH,MAAM,EAAE,yBAAyB,EAAE,GAAG,MAAM,MAAM,CAAC,wBAAwB,CAAC,CAAC;IAC7E,MAAM,cAAc,GAAI,MAAM,CAAC,cAAc,CAAwB,KAAK,SAAS;QACjF,CAAC,CAAC,yBAAyB,CAAC,MAAM,CAAC,cAAc,CAAW,CAAC;QAC7D,CAAC,CAAC,SAAS,CAAC;IACd,IAAI,cAAc,IAAI,cAAc,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAChD,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,qCAAqC,EAAE,IAAI,EAAE,EAAE,GAAG,EAAE,aAAa,EAAE,CAAC,CAAC,CAAC;QACzF,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IAED,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,8BAA8B,CAAC,CAAC;IACzE,MAAM,KAAK,GAAgB,eAAe,CAAC,OAAO,CAAC,MAAM,CAAC,aAAa,CAAW,CAAC,CAAC,CAAC;IACrF,MAAM,MAAM,GAAG,uBAAuB,CAAC,MAAM,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,CAAC;IAClF,MAAM,WAAW,GAAI,MAAM,CAAC,OAA8B,IAAI,MAAM,CAAC,IAAI,EAAE,OAAO,EAAE,WAAW,CAAC;IAChG,IAAI,CAAC,WAAW,EAAE,CAAC;QACjB,OAAO,CAAC,KAAK,CAAC,2FAA2F,CAAC,CAAC;QAC3G,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IAED,MAAM,EAAE,WAAW,EAAE,GAAG,MAAM,MAAM,CAAC,8BAA8B,CAAC,CAAC;IACrE,MAAM,EAAE,OAAO,EAAE,GAAG,WAAW,CAAC,WAAW,CAAC,CAAC;IAC7C,MAAM,YAAY,GAAkB,cAAc;WAC7C,CAAC,MAAM,CAAC,IAAI,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC;YAC7B,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE,KAAK,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,KAAK,EAAE,CAAC;YAC9F,CAAC,CAAC,EAAE,CAAC,CAAC;IACZ,IAAI,YAAY,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC9B,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,2BAA2B,EAAE,IAAI,CAAC,CAAC,CAAC;QACvD,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,CAAC;IACvB,CAAC;IAED,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI;QAChC,CAAC,CAAC,wDAAwD;QAC1D,CAAC,CAAC,6FAA6F,CAAC,CAAC;IAEnG,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,0BAA0B,CAAC,CAAC;IACpE,MAAM,aAAa,GAAG,cAAc,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC;IAC/D,MAAM,UAAU,GAAG,YAAY,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC;IACzC,MAAM,EAAE,oBAAoB,EAAE,oBAAoB,EAAE,GAAG,MAAM,MAAM,CAAC,kCAAkC,CAAC,CAAC;IACxG,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,IAAI,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACtE,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,mBAAmB,CAAC,CAAC,IAAI,IAAI,CAAC;IAC1D,MAAM,MAAM,GAAG,MAAM,oBAAoB,CAAC;QACxC,MAAM;QACN,OAAO;QACP,aAAa;QACb,UAAU;QACV,OAAO,EAAE,MAAM,CAAC,OAA6B;QAC7C,gBAAgB,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC;QACtC,IAAI,EAAE,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS;QACpD,UAAU,EAAE,CAAC,EAAE,SAAS,EAAE,SAAS,EAAE,KAAK,EAAE,EAAE,EAAE;YAC9C,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,aAAa,SAAS,IAAI,KAAK,KAAK,SAAS,IAAI,CAAC,CAAC;QAC1E,CAAC;KACF,CAAC,CAAC;IACH,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,MAAM,CAAC,CAAC,CAAC;AAC5C,CAAC;AAED,SAAS,KAAK,CAAC,IAAiB;IAC9B,OAAO,IAAI,KAAK,IAAI;QAClB,CAAC,CAAC,2FAA2F;QAC7F,CAAC,CAAC,+FAA+F,CAAC;AACtG,CAAC"}

package/dist/src/cli/commands/eval-gold.d.ts

@@ -0,0 +1,2 @@
+export declare function execute(argv: string[]): Promise<void>;
+//# sourceMappingURL=eval-gold.d.ts.map

package/dist/src/cli/commands/eval-gold.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"eval-gold.d.ts","sourceRoot":"","sources":["../../../../src/cli/commands/eval-gold.ts"],"names":[],"mappings":"AAQA,wBAAsB,OAAO,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,CA+G3D"}

package/dist/src/cli/commands/eval-gold.js

@@ -0,0 +1,137 @@
+import { resolve } from 'node:path';
+import { CliExit } from '../cli-exit.js';
+import { langFromArgv } from '../i18n.js';
+import { COMMON_OPTIONS, DEFAULT_REPORTS_DIR } from '../parse-run-config.js';
+import { parseArgsStrictOrExit } from '../parse-strict.js';
+import { requireEvaluationReport } from './_shared.js';
+export async function execute(argv) {
+    const lang = langFromArgv(argv);
+    const [sub, ...rest] = argv;
+    if (!sub) {
+        console.log(usage(lang));
+        throw new CliExit(1);
+    }
+    if (sub === 'init') {
+        const { values } = parseArgsStrictOrExit({
+            args: rest,
+            options: {
+                ...COMMON_OPTIONS,
+                out: { type: 'string', default: './gold-dataset' },
+                annotator: { type: 'string' },
+            },
+        });
+        const { initGoldDataset } = await import('../../grading/gold-cli.js');
+        try {
+            const written = initGoldDataset(values.out, {
+                annotator: values.annotator,
+            });
+            console.log(lang === 'zh'
+                ? `已在 ${values.out} 创建 ${written.length} 个文件：`
+                : `Created ${written.length} files in ${values.out}:`);
+            for (const p of written)
+                console.log(`  ${p}`);
+            console.log(lang === 'zh'
+                ? '\n下一步：编辑 annotations.yaml 加入真实标注，然后运行 omk eval gold validate'
+                : '\nNext step: edit annotations.yaml with real annotations, then run omk eval gold validate');
+        }
+        catch (err) {
+            console.error(err.message);
+            throw new CliExit(1);
+        }
+        return;
+    }
+    if (sub === 'validate') {
+        const { positionals } = parseArgsStrictOrExit({
+            args: rest,
+            allowPositionals: true,
+            options: { ...COMMON_OPTIONS },
+        });
+        const dir = positionals[0];
+        if (!dir) {
+            console.error('Usage: omk eval gold validate <dir>');
+            throw new CliExit(1);
+        }
+        const { validateGoldDataset } = await import('../../grading/gold-cli.js');
+        const result = validateGoldDataset(dir);
+        if (result.ok) {
+            console.log(lang === 'zh'
+                ? `✓ gold dataset OK，共 ${result.sampleCount} 条标注`
+                : `✓ gold dataset OK — ${result.sampleCount} annotations`);
+            return;
+        }
+        console.error(`✗ gold dataset has ${result.issues.length} issue(s):`);
+        for (const msg of result.issues)
+            console.error(`  - ${msg}`);
+        throw new CliExit(1);
+    }
+    if (sub === 'compare') {
+        const reportId = rest[0];
+        if (!reportId) {
+            console.error('Usage: omk eval gold compare <reportId> --gold-dir <dir>');
+            throw new CliExit(1);
+        }
+        const { values } = parseArgsStrictOrExit({
+            args: rest.slice(1),
+            options: {
+                ...COMMON_OPTIONS,
+                'gold-dir': { type: 'string' },
+                variant: { type: 'string' },
+                'reports-dir': { type: 'string', default: DEFAULT_REPORTS_DIR },
+                'bootstrap-samples': { type: 'string', default: '1000' },
+                seed: { type: 'string' },
+            },
+        });
+        const goldDir = values['gold-dir'];
+        if (!goldDir) {
+            console.error('--gold-dir is required');
+            throw new CliExit(1);
+        }
+        const { loadGoldDataset } = await import('../../grading/gold-dataset.js');
+        const { compareGoldToReport, formatGoldCompare } = await import('../../grading/gold-cli.js');
+        const { createFileStore } = await import('../../server/report-store.js');
+        const { dataset, issues } = loadGoldDataset(goldDir);
+        if (!dataset) {
+            console.error('Cannot load gold dataset:');
+            for (const i of issues)
+                console.error(`  - ${i.message}`);
+            throw new CliExit(1);
+        }
+        for (const i of issues)
+            console.error(`warn: ${i.message}`);
+        const store = createFileStore(resolve(values['reports-dir']));
+        const report = requireEvaluationReport(await store.get(reportId), reportId, lang);
+        const samples = Math.max(100, Number(values['bootstrap-samples']) || 1000);
+        const seedVal = values.seed != null ? Number(values.seed) : undefined;
+        const result = compareGoldToReport({
+            report,
+            gold: dataset,
+            variant: values.variant,
+            samples,
+            seed: Number.isFinite(seedVal) ? seedVal : undefined,
+        });
+        console.log(formatGoldCompare(result, dataset));
+        return;
+    }
+    console.error(`Unknown subcommand: eval gold ${sub}. Use init / validate / compare.`);
+    throw new CliExit(1);
+}
+function usage(lang) {
+    return lang === 'zh'
+        ? [
+            'omk eval gold — 管理 human-gold 标注集',
+            '',
+            '用法：',
+            '  omk eval gold init [--out <dir>] [--annotator <name>]',
+            '  omk eval gold validate <dir>',
+            '  omk eval gold compare <reportId> --gold-dir <dir>',
+        ].join('\n')
+        : [
+            'omk eval gold — manage human-gold annotation datasets',
+            '',
+            'Usage:',
+            '  omk eval gold init [--out <dir>] [--annotator <name>]',
+            '  omk eval gold validate <dir>',
+            '  omk eval gold compare <reportId> --gold-dir <dir>',
+        ].join('\n');
+}
+//# sourceMappingURL=eval-gold.js.map