oh-my-knowledge 0.42.0 → 0.43.0

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.
Files changed (38) hide show
  1. package/README.md +1 -1
  2. package/README.zh.md +1 -1
  3. package/dist/artifact-graph/doctor.d.ts +21 -0
  4. package/dist/artifact-graph/doctor.js +569 -0
  5. package/dist/assets/agent-skills/omk/SKILL.md +9 -9
  6. package/dist/assets/agent-skills/omk/references/commands.md +1 -1
  7. package/dist/cli/commands/doctor.js +62 -61
  8. package/dist/cli/commands/eval/index.js +20 -2
  9. package/dist/cli/commands/init.js +1 -0
  10. package/dist/cli/commands/observe/index.d.ts +2 -2
  11. package/dist/cli/commands/observe/index.js +8 -7
  12. package/dist/cli/commands/sample.js +22 -16
  13. package/dist/cli/lib/i18n-dict/common.d.ts +1 -1
  14. package/dist/cli/lib/i18n-dict/common.js +8 -0
  15. package/dist/cli/lib/i18n-dict/help.js +12 -10
  16. package/dist/cli/lib/parse-run-config/samples-discovery.d.ts +5 -7
  17. package/dist/cli/lib/parse-run-config/samples-discovery.js +10 -32
  18. package/dist/cli/lib/parse-run-config.js +4 -4
  19. package/dist/cli/lib/resolve-skill-input.js +10 -12
  20. package/dist/doctor/messages.js +2 -2
  21. package/dist/eval-core/artifact-file-names.d.ts +15 -0
  22. package/dist/eval-core/artifact-file-names.js +46 -0
  23. package/dist/eval-core/evaluation-reporting.js +2 -1
  24. package/dist/eval-core/measurement-dirs.js +13 -7
  25. package/dist/eval-core/report-file-migration.d.ts +10 -0
  26. package/dist/eval-core/report-file-migration.js +90 -0
  27. package/dist/inputs/sample-locator.d.ts +23 -0
  28. package/dist/inputs/sample-locator.js +195 -0
  29. package/dist/inputs/skill-loader.js +7 -17
  30. package/dist/observability/inbox.js +7 -3
  31. package/dist/server/report-server.js +10 -4
  32. package/dist/server/report-store.js +17 -9
  33. package/dist/server/skill-index.js +16 -11
  34. package/dist/types/artifact-graph.d.ts +93 -0
  35. package/dist/types/artifact-graph.js +1 -0
  36. package/dist/types/index.d.ts +1 -0
  37. package/dist/types/index.js +1 -0
  38. package/package.json +1 -1
@@ -1,5 +1,5 @@
1
- import { existsSync, statSync, mkdirSync, readFileSync, readdirSync, unlinkSync, writeFileSync } from 'node:fs';
2
- import { dirname, join, resolve } from 'node:path';
1
+ import { mkdirSync, readFileSync, readdirSync, unlinkSync, writeFileSync } from 'node:fs';
2
+ import { join, resolve } from 'node:path';
3
3
  import { Args, Flags } from '@oclif/core';
4
4
  import { LANG_FLAG, bilingual } from '../oclif/i18n.js';
5
5
  import { BaseCommand } from '../oclif/base-command.js';
@@ -9,50 +9,11 @@ import { tCli } from '../lib/i18n.js';
9
9
  import { makeDoctorProgress } from '../lib/progress.js';
10
10
  import { DEFAULT_DOCTORS_DIR } from '../../eval-core/default-dirs.js';
11
11
  import { indexDoctorWrite, removeDoctorCard } from '../../eval-core/artifact-index.js';
12
+ import { doctorReportFileStem, isReportFileName, reportFilePath } from '../../eval-core/artifact-file-names.js';
13
+ import { migrateLegacyReportFiles } from '../../eval-core/report-file-migration.js';
12
14
  import { projectDoctorsDir, globalDoctorsDir } from '../../eval-core/measurement-dirs.js';
13
- const DEFAULT_SAMPLE_FILENAMES = ['eval-samples.json', 'eval-samples.yaml', 'eval-samples.yml'];
14
- function findSamplesInDir(dir) {
15
- for (const name of DEFAULT_SAMPLE_FILENAMES) {
16
- const candidate = join(dir, name);
17
- if (existsSync(candidate))
18
- return candidate;
19
- }
20
- return null;
21
- }
22
- function sampleSearchDirs(target, cwd) {
23
- const dirs = [];
24
- const add = (dir) => {
25
- const abs = resolve(dir);
26
- if (!dirs.includes(abs))
27
- dirs.push(abs);
28
- };
29
- if (target) {
30
- const absTarget = resolve(target);
31
- if (existsSync(absTarget)) {
32
- const stat = statSync(absTarget);
33
- if (stat.isDirectory()) {
34
- add(absTarget);
35
- add(dirname(absTarget));
36
- add(dirname(dirname(absTarget)));
37
- }
38
- else {
39
- const parent = dirname(absTarget);
40
- add(parent);
41
- add(dirname(parent));
42
- }
43
- }
44
- }
45
- add(cwd);
46
- return dirs;
47
- }
48
- function findDefaultSamplesPath(target, cwd) {
49
- for (const dir of sampleSearchDirs(target, cwd)) {
50
- const samplesPath = findSamplesInDir(dir);
51
- if (samplesPath)
52
- return samplesPath;
53
- }
54
- return null;
55
- }
15
+ import { findDoctorDeprecatedSamplesHint, findDoctorSamplesPath } from '../../inputs/sample-locator.js';
16
+ import { persistDoctorGraphSidecars, removeDoctorGraphSidecars } from '../../artifact-graph/doctor.js';
56
17
  export default class Doctor extends BaseCommand {
57
18
  static description = bilingual({
58
19
  zh: '体检 omk 工作目录,检查 skill 配置 / 依赖 / executor 连通性。',
@@ -186,7 +147,16 @@ export default class Doctor extends BaseCommand {
186
147
  const timeoutSec = flags.timeout != null ? Number(flags.timeout) : defaultTimeoutSec;
187
148
  const timeoutMs = Math.max(1000, Math.floor((Number.isFinite(timeoutSec) ? timeoutSec : defaultTimeoutSec) * 1000));
188
149
  const cwd = process.cwd();
189
- const samplesPath = flags.samples ? resolve(flags.samples) : findDefaultSamplesPath(target, cwd);
150
+ const samplesPath = flags.samples ? resolve(flags.samples) : findDoctorSamplesPath(target, cwd);
151
+ const deprecatedSamplesHint = !flags.samples && !samplesPath
152
+ ? findDoctorDeprecatedSamplesHint(target, cwd)
153
+ : null;
154
+ if (deprecatedSamplesHint) {
155
+ process.stderr.write(tCli('cli.common.deprecated_skill_samples_path', lang, {
156
+ oldPath: deprecatedSamplesHint.oldPath,
157
+ newPath: deprecatedSamplesHint.newPath,
158
+ }));
159
+ }
190
160
  let samples;
191
161
  let requires;
192
162
  if (samplesPath) {
@@ -270,7 +240,7 @@ export default class Doctor extends BaseCommand {
270
240
  }
271
241
  persistDoctorReport(report, flags['output-dir']
272
242
  ? resolve(flags['output-dir'])
273
- : (flags.global ? globalDoctorsDir() : projectDoctorsDir()));
243
+ : (flags.global ? globalDoctorsDir() : projectDoctorsDir()), lang);
274
244
  if (flags.fix) {
275
245
  const existing = report;
276
246
  if (existing.outcome !== 'failed') {
@@ -288,47 +258,76 @@ export default class Doctor extends BaseCommand {
288
258
  // 每个 skill 最多保留多少份历史 doctor 报告(避免无界增长拖慢 studio 启动 +
289
259
  // scanDoctorReports 扫盘成本)。50 = ~每天 1 跑撑 1.5 个月 sparkline,够用。
290
260
  const DOCTOR_HISTORY_MAX_PER_SKILL = 50;
291
- function persistDoctorReport(report, outputDir) {
261
+ function persistDoctorReport(report, outputDir, lang = 'zh') {
292
262
  const dir = outputDir ?? DEFAULT_DOCTORS_DIR;
293
263
  mkdirSync(dir, { recursive: true });
294
- const safeId = report.id.replace(/[/\\:*?"<>|]/g, '_');
264
+ migrateLegacyReportFiles(dir, 'doctor');
295
265
  for (const skill of report.skills) {
296
- const counts = { pass: 0, warn: 0, fail: 0, skipped: 0 };
266
+ const counts = {
267
+ pass: 0,
268
+ warn: 0,
269
+ fail: 0,
270
+ skipped: 0,
271
+ };
297
272
  for (const r of skill.results) {
298
273
  const s = r.status;
299
274
  if (s in counts)
300
275
  counts[s]++;
301
276
  }
277
+ const outcome = skill.status === 'fail' ? 'failed' : skill.status === 'warn' ? 'warnings_only' : 'passed';
302
278
  const perSkill = {
303
279
  ...report,
304
280
  skills: [skill],
305
- ruleStats: { ...counts, total: skill.results.length },
281
+ ruleStats: {
282
+ pass: counts.pass,
283
+ warn: counts.warn,
284
+ fail: counts.fail,
285
+ skipped: counts.skipped,
286
+ total: skill.results.length,
287
+ },
306
288
  totals: {
307
289
  pass: skill.status === 'pass' ? 1 : 0,
308
290
  warn: skill.status === 'warn' ? 1 : 0,
309
291
  fail: skill.status === 'fail' ? 1 : 0,
310
292
  },
311
- outcome: skill.status === 'fail' ? 'failed' : skill.status === 'warn' ? 'warnings_only' : 'passed',
293
+ outcome,
312
294
  };
313
- const safeName = skill.skillName.replace(/[/\\:*?"<>|]/g, '_');
314
- const cardId = `${safeName}-${safeId}`;
315
- const filePath = join(dir, `${cardId}.json`);
295
+ const cardId = doctorReportFileStem(skill.skillName, report.id);
296
+ const filePath = reportFilePath(dir, cardId);
316
297
  writeFileSync(filePath, JSON.stringify(perSkill, null, 2), 'utf8');
317
298
  // 产物发现索引:per-skill 报告落项目本地后,best-effort 追加全局轻卡片,让 studio 跨项目聚合。
318
299
  indexDoctorWrite({
319
300
  id: cardId, path: filePath, skillName: skill.skillName, reportId: report.id, timestamp: report.timestamp,
320
301
  status: skill.status, passCount: counts.pass, warnCount: counts.warn, failCount: counts.fail,
321
302
  }, dir);
303
+ try {
304
+ persistDoctorGraphSidecars({
305
+ report: perSkill,
306
+ skill,
307
+ sourcePath: filePath,
308
+ outputDir: dir,
309
+ fileStem: cardId,
310
+ lang,
311
+ });
312
+ }
313
+ catch (err) {
314
+ const message = err instanceof Error ? err.message : String(err);
315
+ const warning = lang === 'zh'
316
+ ? `⚠️ doctor graph sidecar 写入失败:${message}\n`
317
+ : `⚠️ failed to write doctor graph sidecar: ${message}\n`;
318
+ process.stderr.write(warning);
319
+ }
322
320
  pruneDoctorHistory(dir, skill.skillName, DOCTOR_HISTORY_MAX_PER_SKILL);
323
321
  }
324
322
  }
325
- // 写入新报告后调用:扫 dir 里属于该 skill 的所有 single-skill doctor JSON,
323
+ // 写入新报告后调用:扫 dir 里属于该 skill 的所有 single-skill doctor report,
326
324
  // 按 timestamp 倒排,保留 maxKeep 份最近的,其余删。按 content 匹配 skillName 不
327
- // 看文件名,所以同时清理新 `{name}-{id}.json` 与遗留 `{name}.json` 两种命名。
325
+ // 看文件名,所以清理逻辑不依赖 readdir 顺序或 stem 推断 skill 名。
328
326
  export function pruneDoctorHistory(dir, skillName, maxKeep) {
327
+ migrateLegacyReportFiles(dir, 'doctor');
329
328
  const candidates = [];
330
329
  for (const file of readdirSync(dir)) {
331
- if (!file.endsWith('.json'))
330
+ if (!isReportFileName(file))
332
331
  continue;
333
332
  try {
334
333
  const data = JSON.parse(readFileSync(join(dir, file), 'utf-8'));
@@ -337,20 +336,22 @@ export function pruneDoctorHistory(dir, skillName, maxKeep) {
337
336
  continue;
338
337
  if (data.skills[0].skillName !== skillName)
339
338
  continue;
340
- candidates.push({ file, timestamp: data.timestamp });
339
+ candidates.push({ file, graphStem: doctorReportFileStem(skillName, data.id), timestamp: data.timestamp });
341
340
  }
342
341
  catch { /* skip corrupt / unrelated json */ }
343
342
  }
344
343
  if (candidates.length <= maxKeep)
345
344
  return;
346
345
  candidates.sort((a, b) => b.timestamp.localeCompare(a.timestamp));
347
- for (const { file } of candidates.slice(maxKeep)) {
346
+ for (const { file, graphStem } of candidates.slice(maxKeep)) {
348
347
  try {
349
348
  unlinkSync(join(dir, file));
350
349
  }
351
350
  catch { /* ignore */ }
352
351
  // 连带删卡片:否则被 prune 掉的报告会经 listDoctorCards 合并在本项目 studio「复活」(正文已删、卡片还在)。
353
352
  // 卡片 id = 文件 stem(`{name}-{id}`),与 indexDoctorWrite 写入口径一致。
354
- removeDoctorCard(file.replace(/\.json$/, ''));
353
+ const doctorStem = file.replace(/\.report\.json$/, '');
354
+ removeDoctorCard(doctorStem);
355
+ removeDoctorGraphSidecars(dir, graphStem);
355
356
  }
356
357
  }
@@ -10,6 +10,7 @@ import { computeRunTally } from '../../lib/run-tally.js';
10
10
  import { DEFAULT_BOOTSTRAP_SAMPLES } from '../../../eval-core/bootstrap.js';
11
11
  import { DEFAULT_GATE_THRESHOLD } from '../../../eval-core/verdict.js';
12
12
  import { EVALUATION_REPORT_SCHEMA_VERSION } from '../../../eval-core/evaluation-reporting.js';
13
+ import { findSingleTreatmentDeprecatedSamplesHint, hasUsableSamplesPath, } from '../../../inputs/sample-locator.js';
13
14
  function isDryRunReport(report) {
14
15
  return Boolean(report && typeof report === 'object' && report.dryRun === true);
15
16
  }
@@ -180,6 +181,23 @@ async function announceSavedReport({ report, filePath, reportsDir, values, lang,
180
181
  }
181
182
  async function runEval(_args, flags, lang) {
182
183
  const { values, config, evalConfig } = parseRunConfig({ ...flags });
184
+ if (!values.batch && !hasUsableSamplesPath(config.samplesPath)) {
185
+ const treatmentRaw = typeof values.treatment === 'string' ? values.treatment : '';
186
+ const treatments = treatmentRaw.split(',').map((v) => v.trim()).filter(Boolean);
187
+ const deprecatedSamplesHint = !values.samples && !evalConfig?.samples && treatments.length === 1
188
+ ? findSingleTreatmentDeprecatedSamplesHint(treatments[0], config.skillDir, process.cwd())
189
+ : null;
190
+ if (deprecatedSamplesHint) {
191
+ process.stderr.write(tCli('cli.common.deprecated_skill_samples_path', lang, {
192
+ oldPath: deprecatedSamplesHint.oldPath,
193
+ newPath: deprecatedSamplesHint.newPath,
194
+ }));
195
+ }
196
+ console.error(tCli('cli.common.error_prefix', lang, {
197
+ message: tCli('cli.common.samples_not_found', lang, { path: config.samplesPath }),
198
+ }));
199
+ throw new CliExit(1);
200
+ }
183
201
  const { runEvaluation, runMultiple, runBatchEvaluation } = await import('../../../eval-workflows/run-evaluation.js');
184
202
  config.onProgress = makeOnProgress(lang);
185
203
  const repeatRaw = values.repeat;
@@ -377,8 +395,8 @@ export default class Eval extends BaseCommand {
377
395
  }),
378
396
  samples: Flags.string({
379
397
  description: bilingual({
380
- zh: '用例文件路径。默认 eval-samples.json,也接受 .yaml/.yml;自动发现 --skill-dir 下的 <skill>/.omk/samples.json。',
381
- en: 'Samples file path. Defaults to eval-samples.json (also .yaml/.yml); auto-discovers <skill>/.omk/samples.json under --skill-dir.',
398
+ zh: '用例文件路径。默认项目级 eval-samples.json,也接受 .yaml/.yml;单 treatment 时可自动发现 <skill>/.omk/。',
399
+ en: 'Samples path. Defaults to project-level eval-samples.json (also .yaml/.yml); single-treatment runs can auto-discover <skill>/.omk/.',
382
400
  }),
383
401
  }),
384
402
  'skill-dir': Flags.string({
@@ -8,6 +8,7 @@ import { tCli } from '../lib/i18n.js';
8
8
  const INIT_OMK_GITIGNORE = `# omk 测量 bulk + doctor --fix 备份(项目本地)——不入库;前导 / 锚定 .omk/ 顶层,不误伤嵌套同名目录。
9
9
  /observe-health/
10
10
  /doctors/
11
+ /graphs/
11
12
  /observe-inbox/
12
13
  /reports/
13
14
  /backups/
@@ -2,8 +2,8 @@ import { BaseCommand } from '../../oclif/base-command.js';
2
2
  import type { SkillHealthReport } from '../../../observability/skill-health-analyzer.js';
3
3
  /**
4
4
  * observe-health 报告落盘:id / 文件名加 4 位随机段,根治「同秒两次 omk observe 直接覆盖、数据丢失」的 bug。
5
- * 保留 `-observe-health.json` 后缀 —— resolveObserveHealthDir 靠它判项目优先、listAnalyses 靠它取 id stem、
6
- * loadAnalysis 靠 `${id}.json` 读真身。落盘后 best-effort 追加全局轻卡片,让 studio 跨项目聚合。
5
+ * 文件名统一为 `{run}.report.json`:目录表达 observe-health 域,文件表达这是可读报告。
6
+ * 落盘后 best-effort 追加全局轻卡片,让 studio 跨项目聚合。
7
7
  */
8
8
  export declare function persistObserveHealthReport(report: SkillHealthReport, outDir: string): {
9
9
  id: string;
@@ -1,5 +1,5 @@
1
1
  import { mkdirSync, writeFileSync } from 'node:fs';
2
- import { resolve, join } from 'node:path';
2
+ import { resolve } from 'node:path';
3
3
  import { Args, Flags } from '@oclif/core';
4
4
  import { LANG_FLAG, bilingual } from '../../oclif/i18n.js';
5
5
  import { BaseCommand } from '../../oclif/base-command.js';
@@ -8,17 +8,18 @@ import { tCli } from '../../lib/i18n.js';
8
8
  import { parseLastWindow } from '../../lib/shared.js';
9
9
  import { projectObserveHealthDir, globalObserveHealthDir } from '../../../eval-core/measurement-dirs.js';
10
10
  import { indexObserveWrite } from '../../../eval-core/artifact-index.js';
11
+ import { reportFilePath, runFileSuffix } from '../../../eval-core/artifact-file-names.js';
12
+ import { migrateLegacyReportFiles } from '../../../eval-core/report-file-migration.js';
11
13
  /**
12
14
  * observe-health 报告落盘:id / 文件名加 4 位随机段,根治「同秒两次 omk observe 直接覆盖、数据丢失」的 bug。
13
- * 保留 `-observe-health.json` 后缀 —— resolveObserveHealthDir 靠它判项目优先、listAnalyses 靠它取 id stem、
14
- * loadAnalysis 靠 `${id}.json` 读真身。落盘后 best-effort 追加全局轻卡片,让 studio 跨项目聚合。
15
+ * 文件名统一为 `{run}.report.json`:目录表达 observe-health 域,文件表达这是可读报告。
16
+ * 落盘后 best-effort 追加全局轻卡片,让 studio 跨项目聚合。
15
17
  */
16
18
  export function persistObserveHealthReport(report, outDir) {
17
19
  mkdirSync(outDir, { recursive: true });
18
- const timestamp = new Date().toISOString().replace(/[:.]/g, '-').slice(0, 19);
19
- const rand = Math.random().toString(36).slice(2, 6);
20
- const id = `${timestamp}-${rand}-observe-health`;
21
- const jsonPath = join(outDir, `${id}.json`);
20
+ migrateLegacyReportFiles(outDir, 'observe-health');
21
+ const id = runFileSuffix();
22
+ const jsonPath = reportFilePath(outDir, id);
22
23
  writeFileSync(jsonPath, JSON.stringify(report, null, 2));
23
24
  indexObserveWrite(report, jsonPath, outDir, id);
24
25
  return { id, jsonPath };
@@ -9,6 +9,7 @@ import { CliExit } from '../lib/cli-exit.js';
9
9
  import { tCli } from '../lib/i18n.js';
10
10
  import { projectReportsDir, globalReportsDir } from '../../eval-core/measurement-dirs.js';
11
11
  import { loadSamples, parseYaml, listSampleFilesInDir } from '../../inputs/load-samples.js';
12
+ import { defaultFlatSkillSamplesFile, defaultSkillLocalSamplesFile, findFlatSkillSamplesPath, findSkillSamplesPath, } from '../../inputs/sample-locator.js';
12
13
  import { hashSample } from '../../eval-core/evaluation-reporting.js';
13
14
  import { hashArtifactSource } from '../../inputs/content-hash.js';
14
15
  function isRecord(value) {
@@ -233,23 +234,23 @@ async function runSampleFix(args, flags, lang) {
233
234
  console.error(lang === 'zh' ? '请指定 skill 路径,如: omk sample skills/my-skill/SKILL.md --fix' : 'Specify skill path: omk sample skills/my-skill/SKILL.md --fix');
234
235
  throw new CliExit(1);
235
236
  }
236
- const resolvedSkillPath = resolve(skillPath);
237
- if (!existsSync(resolvedSkillPath)) {
238
- console.error(lang === 'zh' ? `skill 文件不存在: ${resolvedSkillPath}` : `Skill file not found: ${resolvedSkillPath}`);
237
+ const { resolveSkillInput } = await import('../lib/resolve-skill-input.js');
238
+ let resolvedInput;
239
+ try {
240
+ resolvedInput = resolveSkillInput(skillPath, lang);
241
+ }
242
+ catch (err) {
243
+ console.error(err instanceof Error ? err.message : String(err));
239
244
  throw new CliExit(1);
240
245
  }
241
- const isDir = basename(resolvedSkillPath) === 'SKILL.md';
242
- const skillDir = isDir ? dirname(resolvedSkillPath) : dirname(resolvedSkillPath);
243
- const samplesInput = isDir
244
- ? join(skillDir, '.omk')
245
- : resolve('eval-samples.json');
246
+ const samplesInput = resolvedInput.samplesPath;
246
247
  if (!existsSync(samplesInput)) {
247
248
  console.error(lang === 'zh' ? `samples 路径不存在: ${samplesInput},先运行 omk sample 生成` : `Samples path not found: ${samplesInput}, run omk sample first`);
248
249
  throw new CliExit(1);
249
250
  }
250
- const defaultTreatmentName = isDir
251
- ? basename(skillDir)
252
- : basename(resolvedSkillPath, extname(resolvedSkillPath));
251
+ const defaultTreatmentName = resolvedInput.isDirectorySkill
252
+ ? basename(resolvedInput.skillDir)
253
+ : basename(resolvedInput.skillPath, extname(resolvedInput.skillPath));
253
254
  const treatmentName = flags.treatment ?? defaultTreatmentName;
254
255
  process.stderr.write(lang === 'zh' ? `🔍 正在查找 ${treatmentName} 的最新评测报告...\n` : `🔍 Scanning latest report for ${treatmentName}...\n`);
255
256
  // 显式 --reports-dir 固定该目录;默认 overlay(项目 .omk/reports 盖全局),findByVariant 记录优先看项目、
@@ -277,7 +278,7 @@ async function runSampleFix(args, flags, lang) {
277
278
  throw new CliExit(1);
278
279
  }
279
280
  const samples = loadedSamples.samples;
280
- const skillContent = readFileSync(resolvedSkillPath, 'utf-8');
281
+ const skillContent = readFileSync(resolvedInput.skillPath, 'utf-8');
281
282
  const sampleDesignIds = collectSampleDesignFailureIds(report, treatmentName);
282
283
  const sampleDesignCount = sampleDesignIds.size;
283
284
  if (sampleDesignCount === 0) {
@@ -286,7 +287,7 @@ async function runSampleFix(args, flags, lang) {
286
287
  }
287
288
  // 当前内容指纹走整树哈,与 eval 报告口径一致:dir-skill(用户传 .../SKILL.md)哈整棵 skill 目录、
288
289
  // 单文件 .md 哈单文件字节。
289
- const currentContentHash = hashArtifactSource(isDir ? skillDir : resolvedSkillPath, isDir);
290
+ const currentContentHash = hashArtifactSource(resolvedInput.isDirectorySkill ? resolvedInput.skillDir : resolvedInput.skillPath, resolvedInput.isDirectorySkill);
290
291
  try {
291
292
  assertFixReportMatchesCurrentInputs({
292
293
  report,
@@ -435,24 +436,29 @@ async function runSample(args, flags, lang) {
435
436
  let name;
436
437
  let skillPath;
437
438
  let samplesPath;
439
+ let existingSamplesPath;
438
440
  const fullPath = join(skillDir, entry);
439
441
  if (entry.endsWith('.md') && !entry.endsWith('.eval-samples.json')) {
440
442
  name = entry.slice(0, -3);
441
443
  skillPath = fullPath;
442
- samplesPath = join(skillDir, `${name}.eval-samples.json`);
444
+ samplesPath = defaultFlatSkillSamplesFile(skillDir, name);
445
+ existingSamplesPath = findFlatSkillSamplesPath(skillDir, name);
443
446
  }
444
447
  else if (statSync(fullPath).isDirectory()) {
445
448
  const skillMd = join(fullPath, 'SKILL.md');
446
449
  if (!existsSync(skillMd))
447
450
  continue;
451
+ if (existsSync(join(skillDir, `${entry}.md`)))
452
+ continue;
448
453
  name = entry;
449
454
  skillPath = skillMd;
450
- samplesPath = join(fullPath, '.omk', 'samples.json');
455
+ samplesPath = defaultSkillLocalSamplesFile(fullPath);
456
+ existingSamplesPath = findSkillSamplesPath(fullPath);
451
457
  }
452
458
  else {
453
459
  continue;
454
460
  }
455
- if (existsSync(samplesPath)) {
461
+ if (existingSamplesPath) {
456
462
  process.stderr.write(tCli('cli.gen.skill_skipped_existing', lang, { name }));
457
463
  continue;
458
464
  }
@@ -1,3 +1,3 @@
1
1
  import type { CliMessage } from './types.js';
2
- export type CommonMessageKey = 'cli.common.unknown_domain' | 'cli.common.error_prefix' | 'cli.common.skill_dir_not_found' | 'cli.common.skill_file_not_found' | 'cli.common.skill_dir_no_skill_md' | 'cli.common.report_not_found' | 'cli.common.no_judge_model' | 'cli.common.judge_models_single_only' | 'cli.common.warn_load_samples_failed' | 'cli.update.new_version_available' | 'cli.update.box_title' | 'cli.update.box_version_line' | 'cli.update.box_upgrade_line' | 'cli.update.box_silence_line' | 'cli.observe.view_hint' | 'cli.observe.observation_recorded' | 'cli.observe.production_gap' | 'cli.studio.started' | 'cli.studio.stop_hint' | 'cli.studio.open_failed' | 'cli.doctor.no_skill_found' | 'cli.doctor.samples_detected' | 'cli.doctor.progress_skill_start' | 'cli.doctor.progress_skill_done';
2
+ export type CommonMessageKey = 'cli.common.unknown_domain' | 'cli.common.error_prefix' | 'cli.common.skill_dir_not_found' | 'cli.common.skill_file_not_found' | 'cli.common.skill_dir_no_skill_md' | 'cli.common.report_not_found' | 'cli.common.no_judge_model' | 'cli.common.judge_models_single_only' | 'cli.common.warn_load_samples_failed' | 'cli.common.deprecated_skill_samples_path' | 'cli.common.samples_not_found' | 'cli.update.new_version_available' | 'cli.update.box_title' | 'cli.update.box_version_line' | 'cli.update.box_upgrade_line' | 'cli.update.box_silence_line' | 'cli.observe.view_hint' | 'cli.observe.observation_recorded' | 'cli.observe.production_gap' | 'cli.studio.started' | 'cli.studio.stop_hint' | 'cli.studio.open_failed' | 'cli.doctor.no_skill_found' | 'cli.doctor.samples_detected' | 'cli.doctor.progress_skill_start' | 'cli.doctor.progress_skill_done';
3
3
  export declare const commonDict: Record<CommonMessageKey, CliMessage>;
@@ -35,6 +35,14 @@ export const commonDict = {
35
35
  zh: '⚠ 加载 samples 文件失败 ({path}): {message}\n',
36
36
  en: '⚠ Failed to load samples file ({path}): {message}\n',
37
37
  },
38
+ 'cli.common.deprecated_skill_samples_path': {
39
+ zh: '⚠ 发现旧的目录 skill 用例位置:{oldPath}。目录 skill 的私有用例已改放到 {newPath},请迁移。\n',
40
+ en: '⚠ Found deprecated directory skill samples path: {oldPath}. Directory skill samples now live at {newPath}; please migrate.\n',
41
+ },
42
+ 'cli.common.samples_not_found': {
43
+ zh: '未找到评测用例:{path}。请通过 --samples 指定文件,或创建项目级 eval-samples.json;单 treatment 目录 skill 请使用 <skill>/.omk/samples.json。',
44
+ en: 'Eval samples not found: {path}. Pass --samples, create project-level eval-samples.json, or use <skill>/.omk/samples.json for a single-treatment directory skill.',
45
+ },
38
46
  'cli.update.new_version_available': {
39
47
  zh: '\n💡 新版本可用:{old} → {new},运行 npm i -g oh-my-knowledge@latest 升级\n\n',
40
48
  en: '\n💡 New version available: {old} → {new}, run npm i -g oh-my-knowledge@latest to upgrade\n\n',
@@ -170,20 +170,21 @@ omk sample——生成或补齐 eval-samples 评测用例
170
170
  omk sample --batch [--skill-dir <dir>] [options]
171
171
 
172
172
  输出位置(默认):
173
- <skill>/SKILL.md → <skill>/.omk/samples.json(omk 标准约定)
174
- 其他 .md 路径 → 当前目录的 eval-samples.json(兜底)
173
+ 目录 skill(<skill>/SKILL.md) → <skill>/.omk/samples.json(omk 标准约定)
174
+ 扁平 .md(单次) → 当前目录的 eval-samples.json(项目级兜底)
175
+ 扁平 .md(--batch) → <skill-dir>/<name>.eval-samples.json(兼容 paired 布局)
175
176
 
176
177
  选项:
177
178
  --count <n> 强制生成 N 条(不指定时由 LLM 按 skill 类型自动判断:
178
179
  工作流型 6-8 条 / 原子型 4-6 条 / 混合型 5-7 条)
179
180
  --model <name> 生成模型(默认:opus;lean+effort-low 已自动开,想省钱可改 sonnet/haiku)
180
181
  --focus <text> 自然语言指定希望覆盖的场景(追加到 prompt,优先级高于自由发挥)
181
- --batch 为 skill 目录下缺少 eval-samples 的 skill 批量生成
182
+ --batch 为 skill 目录下缺少 samples 的 skill 批量生成
182
183
  --skill-dir <path> skill 目录(batch 使用,默认:skills)
183
184
 
184
185
  示例:
185
- omk improve samples skills/req-tool.md
186
- omk improve samples skills/req-tool.md --count 8 \\
186
+ omk sample skills/req-tool.md
187
+ omk sample skills/req-tool.md --count 8 \\
187
188
  --focus "重点覆盖 tag 查询走 PROJECT 空 → WORKSPACE 兜底的多步流程,以及 search 失败的错误路径"
188
189
  `,
189
190
  en: `
@@ -194,20 +195,21 @@ Usage:
194
195
  omk sample --batch [--skill-dir <dir>] [options]
195
196
 
196
197
  Output path (default):
197
- <skill>/SKILL.md → <skill>/.omk/samples.json (omk standard layout)
198
- other .md paths → ./eval-samples.json in current directory (fallback)
198
+ directory skill (<skill>/SKILL.md) → <skill>/.omk/samples.json (omk standard layout)
199
+ flat .md (single) → ./eval-samples.json in current directory (project fallback)
200
+ flat .md (--batch) → <skill-dir>/<name>.eval-samples.json (compatible paired layout)
199
201
 
200
202
  Options:
201
203
  --count <n> Force N samples (omit to let LLM auto-decide by skill type:
202
204
  workflow 6-8 / atomic 4-6 / mixed 5-7)
203
205
  --model <name> Generation model (default: opus; lean+effort-low applied; pass --model sonnet/haiku to save cost)
204
206
  --focus <text> Natural-language scenario hints appended to the prompt (overrides freeform diversity)
205
- --batch Generate for skills that are missing eval-samples
207
+ --batch Generate for skills that are missing samples
206
208
  --skill-dir <path> Skill directory for batch mode (default: skills)
207
209
 
208
210
  Examples:
209
- omk improve samples skills/req-tool.md
210
- omk improve samples skills/req-tool.md --count 8 \\
211
+ omk sample skills/req-tool.md
212
+ omk sample skills/req-tool.md --count 8 \\
211
213
  --focus "Cover PROJECT-empty → WORKSPACE-fallback multi-step tag lookup and the search-failure error path"
212
214
  `,
213
215
  },
@@ -1,18 +1,16 @@
1
1
  /**
2
2
  * 当 CLI 未传 `--samples` 时,自动发现 sample 路径。
3
3
  *
4
- * 发现顺序(单 treatment 模式):
5
- * 1. treatment 是绝对 / 相对路径(文件 / 目录) → 找它所在目录的 `.omk/`
6
- * 2. fallback 找目录里的 `eval-samples.{json,yaml,yml}`
7
- * 3. fallback `<skillDir>/<treatmentName>/.omk/`
8
- * 4. 终极兜底 cwd 下的 `eval-samples.{json,yaml,yml}`
4
+ * 发现顺序:
5
+ * 1. 单 treatment → 找该 skill 私有 samples(`<skill>/.omk/`)或扁平 skill paired 文件
6
+ * 2. 终极兜底 cwd 下的项目级 `eval-samples.{json,yaml,yml}`
9
7
  *
10
8
  * `loadSamples` 在下游自己处理「文件 vs 目录」—— 目录模式 glob `*.{json,yaml,yml}`
11
9
  * 合并,跳过 reserved 前缀(`report-` / `health-` / `_`)。所以一个 skill 可以把
12
10
  * sample 拆到多文件(`workflow.json` + `platform.json`),也可以单 `samples.json`,
13
11
  * 两种都行。
14
12
  *
15
- * 多 treatment 评测必须显式传 `--samples`(因为「找哪个 skill 的 bundled samples」
16
- * 没有唯一答案)。本函数仅处理单 treatment 与零 treatment 情况。
13
+ * 多 treatment 评测不会触发 skill-local 自动发现(因为「找哪个 skill 的 bundled samples」
14
+ * 没有唯一答案),只回到项目级 samples 兜底。
17
15
  */
18
16
  export declare function discoverSamplesPath(values: Record<string, unknown>, skillDir: string): string;
@@ -1,50 +1,28 @@
1
1
  /**
2
2
  * 当 CLI 未传 `--samples` 时,自动发现 sample 路径。
3
3
  *
4
- * 发现顺序(单 treatment 模式):
5
- * 1. treatment 是绝对 / 相对路径(文件 / 目录) → 找它所在目录的 `.omk/`
6
- * 2. fallback 找目录里的 `eval-samples.{json,yaml,yml}`
7
- * 3. fallback `<skillDir>/<treatmentName>/.omk/`
8
- * 4. 终极兜底 cwd 下的 `eval-samples.{json,yaml,yml}`
4
+ * 发现顺序:
5
+ * 1. 单 treatment → 找该 skill 私有 samples(`<skill>/.omk/`)或扁平 skill paired 文件
6
+ * 2. 终极兜底 cwd 下的项目级 `eval-samples.{json,yaml,yml}`
9
7
  *
10
8
  * `loadSamples` 在下游自己处理「文件 vs 目录」—— 目录模式 glob `*.{json,yaml,yml}`
11
9
  * 合并,跳过 reserved 前缀(`report-` / `health-` / `_`)。所以一个 skill 可以把
12
10
  * sample 拆到多文件(`workflow.json` + `platform.json`),也可以单 `samples.json`,
13
11
  * 两种都行。
14
12
  *
15
- * 多 treatment 评测必须显式传 `--samples`(因为「找哪个 skill 的 bundled samples」
16
- * 没有唯一答案)。本函数仅处理单 treatment 与零 treatment 情况。
13
+ * 多 treatment 评测不会触发 skill-local 自动发现(因为「找哪个 skill 的 bundled samples」
14
+ * 没有唯一答案),只回到项目级 samples 兜底。
17
15
  */
18
- import { resolve, join, dirname } from 'node:path';
19
- import { existsSync, statSync } from 'node:fs';
16
+ import { findProjectSamplesFile, findSingleTreatmentSamplesPath } from '../../../inputs/sample-locator.js';
20
17
  export function discoverSamplesPath(values, skillDir) {
21
18
  const treatmentRaw = values.treatment;
22
19
  const treatments = treatmentRaw
23
20
  ? treatmentRaw.split(',').map((v) => v.trim()).filter(Boolean)
24
21
  : [];
25
22
  if (treatments.length === 1) {
26
- const expr = treatments[0];
27
- const resolved = resolve(expr);
28
- if (existsSync(resolved)) {
29
- const treatmentDir = statSync(resolved).isDirectory() ? resolved : dirname(resolved);
30
- const omkDir = join(treatmentDir, '.omk');
31
- if (existsSync(omkDir))
32
- return omkDir;
33
- for (const name of ['eval-samples.json', 'eval-samples.yaml', 'eval-samples.yml']) {
34
- if (existsSync(join(treatmentDir, name)))
35
- return join(treatmentDir, name);
36
- }
37
- }
38
- const omkDir = join(skillDir, expr, '.omk');
39
- if (existsSync(omkDir))
40
- return omkDir;
23
+ const samplesPath = findSingleTreatmentSamplesPath(treatments[0], skillDir, process.cwd());
24
+ if (samplesPath)
25
+ return samplesPath;
41
26
  }
42
- let cwdFile = 'eval-samples.json';
43
- if (!existsSync(resolve(cwdFile))) {
44
- if (existsSync(resolve('eval-samples.yaml')))
45
- cwdFile = 'eval-samples.yaml';
46
- else if (existsSync(resolve('eval-samples.yml')))
47
- cwdFile = 'eval-samples.yml';
48
- }
49
- return cwdFile;
27
+ return findProjectSamplesFile(process.cwd()) ?? 'eval-samples.json';
50
28
  }
@@ -41,10 +41,10 @@ export function parseRunConfig(values) {
41
41
  ? loadEvalConfig(values.config)
42
42
  : null;
43
43
  const skillDir = resolve(values['skill-dir'] ?? 'skills');
44
- // 2) Resolve samples path: CLI > config > <skillDir>/<treatment>/.omk/ discovery > cwd default.
45
- // The .omk/ discovery only fires when exactly one --treatment is given, so omk knows
46
- // which skill's bundled samples to use. The dir form (loadSamples handles both file + dir)
47
- // means a skill can split samples across multiple files (workflow.json / platform.json / ...).
44
+ // 2) Resolve samples path: CLI > config > single-treatment skill-local discovery > cwd project default.
45
+ // Skill-local discovery only fires when exactly one --treatment is given, so omk knows
46
+ // which skill's bundled samples to use. The `<skill>/.omk/` dir form (loadSamples handles
47
+ // both file + dir) means a skill can split samples across multiple files.
48
48
  const cliSamples = values.samples;
49
49
  let samplesFile;
50
50
  if (cliSamples) {
@@ -1,10 +1,11 @@
1
1
  import { resolve, join, dirname, basename } from 'node:path';
2
2
  import { existsSync, statSync } from 'node:fs';
3
3
  import { tCli } from './i18n.js';
4
+ import { findFlatSkillSamplesPath, findProjectSamplesFile, findSkillSamplesPath, skillLocalSamplesDir, } from '../../inputs/sample-locator.js';
4
5
  // 统一 skill 入参解析:既接受 SKILL.md 文件(老式 + flat skill),也接受 directory-skill
5
- // 目录(skills/foo/ 自动找 foo/SKILL.md)。samples 发现优先级:.omk/ 目录(loadSamples
6
- // 支持目录模式,自动 glob 多文件) > eval-samples.json > .yaml > .yml,fallback 是
7
- // .omk/ 目录路径(给上游 "samples 不存在" 的提示一个稳定路径)。
6
+ // 目录(skills/foo/ 自动找 foo/SKILL.md)。目录-skill 以 `<skill>/.omk/` 为标准
7
+ // samples 命名空间;扁平 .md 兼容 paired sidecar,找不到时回到项目级
8
+ // `eval-samples.json`。
8
9
  //
9
10
  // 错误用 tCli 走 i18n,调用方直接 console.error err.message 给用户看,zh/en 都要正确。
10
11
  export function resolveSkillInput(input, lang) {
@@ -26,16 +27,13 @@ export function resolveSkillInput(input, lang) {
26
27
  skillPath = resolved;
27
28
  skillDir = dirname(resolved);
28
29
  }
29
- const omkDir = join(skillDir, '.omk');
30
- const candidates = [
31
- ...(existsSync(omkDir) ? [omkDir] : []),
32
- join(skillDir, 'eval-samples.json'),
33
- join(skillDir, 'eval-samples.yaml'),
34
- join(skillDir, 'eval-samples.yml'),
35
- ];
36
- // fallback 到 .omk/ 目录:上游会报 "no sample files found in directory" 引导用户创建
37
- const samplesPath = candidates.find(existsSync) ?? omkDir;
38
30
  // 形态以解析后的 skillPath 命名为准:目录-skill 的 skillPath 总是 `.../SKILL.md`。
39
31
  const isDirectorySkill = basename(skillPath) === 'SKILL.md';
32
+ const flatSkillName = basename(skillPath).replace(/\.md$/i, '');
33
+ const samplesPath = isDirectorySkill
34
+ ? findSkillSamplesPath(skillDir) ?? skillLocalSamplesDir(skillDir)
35
+ : findFlatSkillSamplesPath(skillDir, flatSkillName)
36
+ ?? findProjectSamplesFile(process.cwd())
37
+ ?? 'eval-samples.json';
40
38
  return { skillPath, skillDir, samplesPath, isDirectorySkill };
41
39
  }
@@ -133,8 +133,8 @@ export const DOCTOR_MESSAGES = {
133
133
  en: 'directory-skill missing SKILL.md entry file',
134
134
  },
135
135
  'cli.doctor.skill_metadata.hint.frontmatter': {
136
- zh: 'front-matter 用 YAML 语法,key: value 或 - item 形式。可参考 examples/multi-skills 下的 skill 写法',
137
- en: 'front-matter uses YAML syntax (key: value or - item). See examples/multi-skills for reference',
136
+ zh: 'front-matter 用 YAML 语法,key: value 或 - item 形式。可参考 examples/skill-map-showcase/skills/release-readiness 的目录式 skill 写法',
137
+ en: 'front-matter uses YAML syntax (key: value or - item). See examples/skill-map-showcase/skills/release-readiness for a directory-skill reference',
138
138
  },
139
139
  'cli.doctor.skill_metadata.hint.hardrules': {
140
140
  zh: 'hardRules 必须写在 SKILL.md front-matter 中,格式为 hardRules: [{ id, rule, expectedBehavior }];id 要稳定且唯一,expectedBehavior 写可观察行为。',