oh-my-knowledge 0.47.0 → 0.48.0

package/dist/assets/agent-skills/omk/references/commands.md

@@ -604,6 +604,7 @@ omk sample [skillPath] [flags]
 - `--no-mock` `boolean`:不生成 mocks，eval 时所有工具调用真实执行。
 - `--observations-dir` `option`:observe inbox 目录（from-traces 模式用），默认项目 .omk/observe-inbox。
 - `--reports-dir` `option`:报告目录（fix 模式用），默认 ~/.oh-my-knowledge/reports。
+- `--skill` `option`:仅从指定 skill 的 observe inbox 信号生成草稿（仅 from-traces 模式用）。
 - `--skill-dir` `option` (默认 `skills`):skill 根目录，默认 skills。batch 模式扫此目录。
 - `--treatment` `option`:指定 treatment 名（fix 模式用），默认推断自 skill 路径。
 

package/dist/cli/commands/observe/inbox.js

@@ -3,6 +3,7 @@ import { Flags } from '@oclif/core';
 import { BaseCommand } from '../../oclif/base-command.js';
 import { LANG_FLAG, bilingual } from '../../oclif/i18n.js';
 import { integerStringParser } from '../../oclif/parsers.js';
+import { shellQuoteArg } from '../../../shared/shell-quote.js';
 function pickSkillCount(value, skillName) {
     if (!value || value[skillName] == null)
         return undefined;
@@ -61,6 +62,11 @@ export async function runObserveInbox(_args, flags, lang) {
     if (flags.skill) {
         items = items.filter((item) => item.skillName === flags.skill);
     }
+    const recyclableCount = items.filter((item) => item.severity !== 'noise').length;
+    const sampleCommandBase = `omk sample --from-traces --observations-dir ${shellQuoteArg(dir)}`;
+    const sampleCommand = flags.skill
+        ? `${sampleCommandBase} --skill ${shellQuoteArg(flags.skill)}`
+        : sampleCommandBase;
     if (flags['by-skill']) {
         const reports = flags.skill
             ? loadLatestObservationInboxReports(dir).map((report) => ({
@@ -87,6 +93,11 @@ export async function runObserveInbox(_args, flags, lang) {
         for (const row of rows) {
             console.log(`- ${row.skillName} invocations=${row.invocationCount} sessions=${row.sessionCount} processFindings=${row.observationCount} high=${row.highCount} medium=${row.mediumCount} low=${row.lowCount} noise=${row.noiseCount}${row.latestSeen ? ` latest=${row.latestSeen}` : ''}`);
         }
+        if (recyclableCount > 0) {
+            console.log(lang === 'zh'
+                ? `提示：确认信号后生成回归用例草稿：${sampleCommand}`
+                : `Tip: after confirming signals, draft regression samples: ${sampleCommand}`);
+        }
         return;
     }
     if (flags.explore) {
@@ -122,6 +133,11 @@ export async function runObserveInbox(_args, flags, lang) {
     console.log(lang === 'zh'
         ? 'Tip: omk observe inbox --explore 10 --include-noise  # 显式包含 noise 桶'
         : 'Tip: omk observe inbox --explore 10 --include-noise  # explicitly include the noise bucket');
+    if (recyclableCount > 0) {
+        console.log(lang === 'zh'
+            ? `提示：确认高风险或抽样信号后生成回归用例草稿：${sampleCommand}`
+            : `Tip: after confirming high-risk / sampled signals, draft regression samples: ${sampleCommand}`);
+    }
 }
 function hasLlmEnhancedRuntimeEvidence(evidence) {
     return evidence.userMessages.length > 0

package/dist/cli/commands/observe/ingest.js

@@ -3,6 +3,7 @@ import { Args, Flags } from '@oclif/core';
 import { BaseCommand } from '../../oclif/base-command.js';
 import { LANG_FLAG, bilingual } from '../../oclif/i18n.js';
 import { CliExit } from '../../lib/cli-exit.js';
+import { shellQuoteArg } from '../../../shared/shell-quote.js';
 export default class ObserveIngest extends BaseCommand {
     static description = bilingual({
         zh: '把 trace 目录 ingest 成 observation inbox 报告。',
@@ -71,9 +72,11 @@ export default class ObserveIngest extends BaseCommand {
             report.diagnostics = buildObserveDiagnosticsFromReport(report);
             const path = saveObservationInboxReport(report, outDir);
             console.log(JSON.stringify(report, null, 2));
+            const inboxCommand = `omk observe inbox --input-dir ${shellQuoteArg(outDir)}`;
+            const sampleCommand = `omk sample --from-traces --observations-dir ${shellQuoteArg(outDir)}`;
             process.stderr.write(lang === 'zh'
-                ? `observe inbox 已写入: ${path}\n`
-                : `observe inbox written to: ${path}\n`);
+                ? `observe inbox 已写入：${path}\n下一步：${inboxCommand}\n确认高风险或抽样信号后，可生成回归用例草稿：${sampleCommand}\n`
+                : `observe inbox written to: ${path}\nNext: ${inboxCommand}\nAfter confirming high-risk / sampled signals, draft regression samples: ${sampleCommand}\n`);
         });
     }
 }

package/dist/cli/commands/sample.d.ts

@@ -1,6 +1,7 @@
 import { BaseCommand } from '../oclif/base-command.js';
 import { type CliLang } from '../lib/i18n.js';
 import { type LoadSamplesResult } from '../../inputs/load-samples.js';
+import type { SampleFlags } from '../lib/cmd-flags.js';
 import type { Report, Sample as SampleType } from '../../types/index.js';
 /** --append 合并:已有用例原样保留,新用例逐条接在后面;sample_id 撞已有(或本批已用)时
  *  自动加 `-2`/`-3` 后缀去重。模型每次从 s001 重编号,撞 id 不代表内容重复,所以是改名保留
@@ -24,6 +25,7 @@ export declare function assertFixReportMatchesCurrentInputs(params: {
     lang?: CliLang;
 }): void;
 export declare function writeFixedSamplesToSources(loaded: Pick<LoadSamplesResult, 'sourceFiles' | 'sampleSourceById'>, samples: SampleType[], changedIds: Set<string>): string[];
+export declare function runSampleFromTraces(flags: SampleFlags, lang: CliLang): Promise<void>;
 export default class Sample extends BaseCommand {
     static description: string;
     static examples: {
@@ -48,6 +50,7 @@ export default class Sample extends BaseCommand {
         treatment: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
         'from-traces': import("@oclif/core/interfaces").BooleanFlag<boolean>;
         'observations-dir': import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
+        skill: import("@oclif/core/interfaces").OptionFlag<string | undefined, import("@oclif/core/interfaces").CustomOptions>;
     };
     run(): Promise<void>;
 }

package/dist/cli/commands/sample.js

@@ -346,7 +346,7 @@ async function runSampleFix(args, flags, lang) {
         ? `\n🔧 修复完成: ${result.fixedCount}/${sampleDesignCount} 条已修复 → ${outputTarget}${cost}\n`
         : `\n🔧 Fix complete: ${result.fixedCount}/${sampleDesignCount} fixed → ${outputTarget}${cost}\n`);
 }
-async function runSampleFromTraces(flags, lang) {
+export async function runSampleFromTraces(flags, lang) {
     const { queryObservationInbox, DEFAULT_OBSERVATIONS_DIR } = await import('../../observability/inbox.js');
     const { generateSamplesFromTraces } = await import('../../authoring/generator.js');
     const obsDir = resolve(flags['observations-dir'] ?? DEFAULT_OBSERVATIONS_DIR);
@@ -358,11 +358,14 @@ async function runSampleFromTraces(flags, lang) {
     }
     // Drop noise-tier signals up front: they're exactly what the generator is told to
     // skip, so filtering here avoids feeding junk to the LLM and keeps the no-op path clean.
-    const items = queryObservationInbox(obsDir).filter((it) => it.severity !== 'noise');
+    let items = queryObservationInbox(obsDir).filter((it) => it.severity !== 'noise');
+    if (flags.skill) {
+        items = items.filter((it) => it.skillName === flags.skill);
+    }
     if (items.length === 0) {
         process.stderr.write(lang === 'zh'
-            ? `✅ ${obsDir} 没有可回流的失败信号（噪声级已跳过）\n`
-            : `✅ No recyclable failure signals in ${obsDir} (noise-level skipped)\n`);
+            ? `✅ ${obsDir}${flags.skill ? ` 中 ${flags.skill}` : ''} 没有可回流的失败信号（噪声级已跳过）\n`
+            : `✅ No recyclable failure signals${flags.skill ? ` for ${flags.skill}` : ''} in ${obsDir} (noise-level skipped)\n`);
         return;
     }
     const outPath = join(obsDir, 'sample-drafts.json');
@@ -374,8 +377,8 @@ async function runSampleFromTraces(flags, lang) {
     }
     const count = flags.count !== undefined ? Math.max(1, Number(flags.count) || 5) : undefined;
     process.stderr.write(lang === 'zh'
-        ? `🔭 发现 ${items.length} 个失败信号，正在生成回归用例草稿...\n`
-        : `🔭 Found ${items.length} failure signal(s); generating regression-sample drafts...\n`);
+        ? `🔭 发现 ${items.length} 个${flags.skill ? ` ${flags.skill} 的` : ''}失败信号，正在生成回归用例草稿...\n`
+        : `🔭 Found ${items.length}${flags.skill ? ` ${flags.skill}` : ''} failure signal(s); generating regression-sample drafts...\n`);
     try {
         const { samples, costUSD } = await generateSamplesFromTraces({ items, count, model: flags.model, executorName: flags.executor });
         const cost = costUSD > 0 ? ` $${costUSD.toFixed(4)}` : '';
@@ -401,6 +404,10 @@ async function runSampleFromTraces(flags, lang) {
     }
 }
 async function runSample(args, flags, lang) {
+    if (flags.skill && !flags['from-traces']) {
+        console.error(lang === 'zh' ? '--skill 仅支持 --from-traces 模式。' : '--skill is only supported with --from-traces.');
+        throw new CliExit(2);
+    }
     // --append 目前只在单 skill 生成路径实现;batch / from-traces / fix 不处理它,
     // 静默忽略会误导(用户以为在追加,实际没有)。提前互斥校验,明确报错。
     if (flags.append && (flags.batch || flags['from-traces'] || flags.fix)) {
@@ -695,6 +702,12 @@ export default class Sample extends BaseCommand {
                 en: 'Observe inbox dir (from-traces mode), default project .omk/observe-inbox.',
             }),
         }),
+        skill: Flags.string({
+            description: bilingual({
+                zh: '仅从指定 skill 的 observe inbox 信号生成草稿（仅 from-traces 模式用）。',
+                en: 'Only draft from observe-inbox signals for the specified skill (from-traces mode only).',
+            }),
+        }),
     };
     async run() {
         const { args, flags } = await this.parse(Sample);

package/dist/cli/lib/cmd-flags.d.ts

@@ -162,6 +162,7 @@ export interface SampleFlags {
     treatment?: string;
     'from-traces': boolean;
     'observations-dir'?: string;
+    skill?: string;
 }
 export type GoldInitArgs = Record<string, never>;
 export interface GoldInitFlags {

package/dist/observability/inbox-view-model.d.ts

@@ -5,6 +5,7 @@ import { type ResolvedObservationReviewSession } from './resolved-review.js';
 import { type ObservationSkillChain } from './skill-chain.js';
 import { type ResolvedSkillStandards, type SkillDerivedStandards } from './soft-standards/index.js';
 export interface ObservationInboxViewModel {
+    observationsDir?: string;
     activeSkill?: string;
     allItems: ObservationInboxItem[];
     items: ObservationInboxItem[];

package/dist/observability/inbox-view-model.js

@@ -76,6 +76,7 @@ export function buildObservationInboxViewModel(observationsDir, options = {}) {
         }
     }
     return {
+        observationsDir,
         activeSkill,
         allItems,
         items,

package/dist/renderer/observation-inbox-renderer.js

@@ -4,9 +4,10 @@ import { findNegativeFeedbackMatches, findPositiveFeedbackMatches, findUserCorre
 import { durationMsBetween } from '../shared/time.js';
 import { OBSERVATION_INBOX_STYLES } from './observation-inbox/styles.js';
 import { experienceSkillAnchor, formatTimeRange as formatTimeRangeImpl, formatTimestamp, renderArtifactVersion, renderField, renderJson, renderFeedbackAttributionLabel, skillAnchor, truncateText, } from './observation-inbox/helpers.js';
+import { shellQuoteArg } from '../shared/shell-quote.js';
 export { renderFeedbackAttributionLabel } from './observation-inbox/helpers.js';
 export function renderObservationInboxPage(model, lang = DEFAULT_LANG) {
-    const { activeSkill, allItems, items, reports, experienceReports, skillInvocationCounts, skillSessionCounts, skillInvocationLastSeen, skillToolCallCounts, skillChains, skillDerivedStandards, totalSkillInvocations, severitySkillCounts, skillCount, reportCount, latestSeenLabel, reviewState, resolvedReviewSessions, } = model;
+    const { observationsDir, activeSkill, allItems, items, reports, experienceReports, skillInvocationCounts, skillSessionCounts, skillInvocationLastSeen, skillToolCallCounts, skillChains, skillDerivedStandards, totalSkillInvocations, severitySkillCounts, skillCount, reportCount, latestSeenLabel, reviewState, resolvedReviewSessions, } = model;
     const experience = experienceReports.find((report) => report.skills.length > 0 || report.sessions.length > 0 || report.invocations.length > 0);
     const pageTitle = activeSkill ? `Observe Inbox · ${activeSkill}` : 'Observe Inbox';
     const experienceToolCountsBySkill = new Map();
@@ -5167,6 +5168,22 @@ export function renderObservationInboxPage(model, lang = DEFAULT_LANG) {
     const empty = items.length === 0 && !experience
         ? `<p style="color:var(--text-muted);margin-top:24px">${activeSkill ? `当前 skill 没有可展示的调用或过程发现：${e(activeSkill)}` : (lang === 'zh' ? '暂无 inbox item。运行 omk observe ingest <sessions-dir> 生成。' : 'No inbox items. Run omk observe ingest <sessions-dir> first.')}</p>`
         : '';
+    const recyclableObservationCount = allItems.filter((item) => item.severity !== 'noise').reduce((sum, item) => sum + item.occurrences, 0);
+    const sampleFromTracesBaseCommand = `omk sample --from-traces --observations-dir ${shellQuoteArg(observationsDir || '.omk/observe-inbox')}`;
+    const sampleFromTracesCommand = activeSkill
+        ? `${sampleFromTracesBaseCommand} --skill ${shellQuoteArg(activeSkill)}`
+        : sampleFromTracesBaseCommand;
+    const observeLoopCta = recyclableObservationCount > 0
+        ? `<section data-observe-feedback-loop style="margin-top:14px;border:1px solid var(--border);border-radius:8px;background:var(--bg-muted);padding:12px 14px;font-size:13px;line-height:1.55">
+        <div style="display:flex;justify-content:space-between;gap:12px;align-items:flex-start;flex-wrap:wrap">
+          <div>
+            <div style="font-weight:700;color:var(--text-primary)">${lang === 'zh' ? '把已确认的 observe gap 回流成 eval sample' : 'Recycle confirmed observe gaps into eval samples'}</div>
+            <div style="color:var(--text-muted);margin-top:3px">${lang === 'zh' ? `当前有 ${recyclableObservationCount} 个非噪声信号。先 review 高风险或抽样信号，确认可复现后生成草稿。` : `${recyclableObservationCount} non-noise signal(s) are available. Review high-risk / sampled signals first, then draft reproducible cases.`}</div>
+          </div>
+          <code style="display:block;max-width:100%;overflow:auto;white-space:nowrap;padding:6px 8px;border:1px solid var(--border);border-radius:6px;background:var(--bg-surface);color:var(--text-primary)">${e(sampleFromTracesCommand)}</code>
+        </div>
+      </section>`
+        : '';
     const v0SummarySection = `
       <section class="observe-summary-grid" style="display:grid;grid-template-columns:repeat(4,minmax(0,1fr));gap:10px;margin-top:14px">
         <div style="padding:12px;border:1px solid var(--border);border-radius:8px;background:var(--bg-surface)">
@@ -5222,6 +5239,7 @@ export function renderObservationInboxPage(model, lang = DEFAULT_LANG) {
         <div class="metric-guide-body">${metricGuideHtml}</div>
       </aside>
       ${empty}
+      ${observeLoopCta}
       ${experienceSection}
       <div data-v0-observation-view style="display:none">
       <div class="report-version-divider" aria-label="1.0 和 2.0 报告分隔">

package/dist/shared/shell-quote.d.ts

@@ -0,0 +1,2 @@
+/** Format one POSIX shell argument for user-facing copy/paste commands. */
+export declare function shellQuoteArg(value: string): string;

package/dist/shared/shell-quote.js

@@ -0,0 +1,7 @@
+const SAFE_SHELL_ARG_RE = /^[A-Za-z0-9_./:@%+=,-]+$/;
+/** Format one POSIX shell argument for user-facing copy/paste commands. */
+export function shellQuoteArg(value) {
+    if (value && SAFE_SHELL_ARG_RE.test(value))
+        return value;
+    return `'${value.replace(/'/g, "'\\''")}'`;
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-knowledge",
-  "version": "0.47.0",
+  "version": "0.48.0",
   "packageManager": "yarn@4.16.0",
   "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.",
   "type": "module",