oh-my-knowledge 0.48.0 → 0.49.0

package/dist/cli/commands/sample.js

@@ -1,38 +1,29 @@
-import { resolve, join, basename, dirname, extname } from 'node:path';
+import { resolve, join, basename, dirname, extname, relative, sep } from 'node:path';
 import { existsSync, mkdirSync, readFileSync, readdirSync, statSync, writeFileSync } from 'node:fs';
-import yaml from 'js-yaml';
 import { Args, Flags } from '@oclif/core';
 import { LANG_FLAG, bilingual } from '../oclif/i18n.js';
 import { BaseCommand } from '../oclif/base-command.js';
 import { integerStringParser } from '../oclif/parsers.js';
 import { CliExit } from '../lib/cli-exit.js';
 import { tCli } from '../lib/i18n.js';
+import { formatSampleGenerationFailureHint } from '../lib/generation-failure-hint.js';
+import { resolveRuntimeSelection } from '../lib/runtime-defaults.js';
 import { projectReportsDir, globalReportsDir } from '../../eval-core/measurement-dirs.js';
-import { loadSamples, parseYaml, listSampleFilesInDir } from '../../inputs/load-samples.js';
+import { loadSamples, listSampleFilesInDir } from '../../inputs/load-samples.js';
+import { getSamplesArray, parseSampleDocument, stringifySampleDocument, writeFixedSamplesToSources, } from '../../inputs/sample-document.js';
 import { defaultFlatSkillSamplesFile, defaultSkillLocalSamplesFile, findFlatSkillSamplesPath, findSkillSamplesPath, } from '../../inputs/sample-locator.js';
 import { hashSample } from '../../eval-core/evaluation-reporting.js';
 import { hashArtifactSource } from '../../inputs/content-hash.js';
-function isRecord(value) {
-    return typeof value === 'object' && value !== null && !Array.isArray(value);
+import { shellQuoteArg } from '../../shared/shell-quote.js';
+function userFacingPath(filePath) {
+    const rel = relative(process.cwd(), filePath);
+    if (rel && rel !== '..' && !rel.startsWith(`..${sep}`))
+        return rel;
+    return filePath;
 }
-function isYamlPath(filePath) {
-    return /\.(ya?ml)$/i.test(filePath);
-}
-function parseSampleDocument(filePath) {
-    const raw = readFileSync(filePath, 'utf-8');
-    return isYamlPath(filePath) ? parseYaml(raw) : JSON.parse(raw);
-}
-function getSamplesArray(document, filePath) {
-    if (Array.isArray(document))
-        return document;
-    if (isRecord(document) && Array.isArray(document.samples))
-        return document.samples;
-    throw new Error(`invalid samples file shape: ${filePath} (expected an array or an object with a 'samples' field)`);
-}
-function stringifySampleDocument(filePath, document) {
-    if (isYamlPath(filePath))
-        return yaml.dump(document, { lineWidth: -1, noRefs: true });
-    return JSON.stringify(document, null, 2);
+export function sampleNextEvalCommand(resolved) {
+    const treatmentPath = resolved.isDirectorySkill ? resolved.skillDir : resolved.skillPath;
+    return `omk eval --control baseline --treatment ${shellQuoteArg(userFacingPath(treatmentPath))}`;
 }
 /** --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
@@ -122,7 +113,7 @@ export function collectSampleDesignFailureIds(report, treatmentName) {
     return ids;
 }
 export function assertFixReportMatchesCurrentInputs(params) {
-    const { report, treatmentName, currentContentHash, samples, sampleIds } = params;
+    const { report, treatmentName, currentContentHash, samples, samplesBaseDir, sampleIds, } = params;
     const lang = params.lang ?? 'zh';
     const issues = [];
     // schemaVersion < 2 的报告:artifactHashes 是旧「仅 SKILL.md 正文文本」哈,与当前「整棵可分发树」哈
@@ -168,7 +159,7 @@ export function assertFixReportMatchesCurrentInputs(params) {
                 missingReportHashes.push(sampleId);
                 continue;
             }
-            const currentSampleHash = hashSample(currentSample);
+            const currentSampleHash = hashSample(currentSample, samplesBaseDir);
             if (expectedSampleHash !== currentSampleHash) {
                 mismatchedSamples.push(sampleId);
             }
@@ -199,36 +190,15 @@ export function assertFixReportMatchesCurrentInputs(params) {
         : 'Re-run omk eval first, then run omk sample --fix again.';
     throw new Error([heading, ...issues, hint].join('\n'));
 }
-export function writeFixedSamplesToSources(loaded, samples, changedIds) {
-    if (changedIds.size === 0)
-        return [];
-    const fixedById = new Map(samples.map((sample) => [sample.sample_id, sample]));
-    const idsByFile = new Map();
-    for (const sampleId of changedIds) {
-        const filePath = loaded.sampleSourceById[sampleId];
-        if (!filePath)
-            throw new Error(`sample ${sampleId} source file not found`);
-        const ids = idsByFile.get(filePath) ?? new Set();
-        ids.add(sampleId);
-        idsByFile.set(filePath, ids);
-    }
-    const written = [];
-    for (const [filePath, ids] of idsByFile.entries()) {
-        const document = parseSampleDocument(filePath);
-        const fileSamples = getSamplesArray(document, filePath);
-        const nextSamples = fileSamples.map((sample) => (ids.has(sample.sample_id) ? (fixedById.get(sample.sample_id) ?? sample) : sample));
-        const nextDocument = Array.isArray(document)
-            ? nextSamples
-            : { ...document, samples: nextSamples };
-        writeFileSync(filePath, stringifySampleDocument(filePath, nextDocument));
-        written.push(filePath);
-    }
-    return written;
-}
+export { writeFixedSamplesToSources };
 async function runSampleFix(args, flags, lang) {
     const { fixSamples } = await import('../../authoring/sample-fixer.js');
     const { createFileStore, createOverlayReportStore } = await import('../../server/report-store.js');
     const model = flags.model;
+    const executorName = flags.executor;
+    if (!model || !executorName) {
+        throw new Error('internal error: sample fix requires runtime selection before execution');
+    }
     const skillPath = args.skillPath;
     if (!skillPath) {
         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');
@@ -294,6 +264,7 @@ async function runSampleFix(args, flags, lang) {
             treatmentName,
             currentContentHash,
             samples,
+            samplesBaseDir: loadedSamples.baseDir,
             sampleIds: sampleDesignIds,
             lang,
         });
@@ -304,7 +275,7 @@ async function runSampleFix(args, flags, lang) {
     }
     process.stderr.write(lang === 'zh' ? `🔧 发现 ${sampleDesignCount} 条 sample_design 失败，开始修复...\n` : `🔧 Found ${sampleDesignCount} sample_design failure(s), fixing...\n`);
     const { createExecutor } = await import('../../executors/index.js');
-    const exec = createExecutor(flags.executor ?? 'claude');
+    const exec = createExecutor(executorName);
     const executorFn = async (opts) => {
         const result = await exec({
             model: opts.model,
@@ -313,7 +284,12 @@ async function runSampleFix(args, flags, lang) {
             timeoutMs: opts.timeoutMs,
             lean: opts.lean,
         });
-        return { ok: result.ok, text: result.output ?? '', costUSD: result.costUSD };
+        return {
+            ok: result.ok,
+            text: result.output ?? '',
+            costUSD: result.costUSD,
+            costReported: result.costReportedByExecutor !== false,
+        };
     };
     const result = await fixSamples({
         skillContent,
@@ -336,7 +312,9 @@ async function runSampleFix(args, flags, lang) {
             process.stderr.write(lang === 'zh' ? `  ⚠ ${f.sampleId} 未修改${f.error ? `: ${f.error}` : ''}\n` : `  ⚠ ${f.sampleId} unchanged${f.error ? `: ${f.error}` : ''}\n`);
         }
     }
-    const cost = result.costUSD > 0 ? ` $${result.costUSD.toFixed(4)}` : '';
+    const cost = result.costReported
+        ? ` $${result.costUSD.toFixed(4)}`
+        : ` ${lang === 'zh' ? '成本未完整上报' : 'cost not fully reported'}`;
     const outputTarget = writtenFiles.length === 0
         ? samplesInput
         : writtenFiles.length === 1
@@ -349,6 +327,11 @@ async function runSampleFix(args, 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 model = flags.model;
+    const executorName = flags.executor;
+    if (!model || !executorName) {
+        throw new Error('internal error: sample generation requires runtime selection before execution');
+    }
     const obsDir = resolve(flags['observations-dir'] ?? DEFAULT_OBSERVATIONS_DIR);
     if (!existsSync(obsDir)) {
         console.error(lang === 'zh'
@@ -380,7 +363,7 @@ export async function runSampleFromTraces(flags, lang) {
         ? `🔭 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 { samples, costUSD } = await generateSamplesFromTraces({ items, count, model, executorName });
         const cost = costUSD > 0 ? ` $${costUSD.toFixed(4)}` : '';
         if (samples.length === 0) {
             // The model conservatively skipped every signal (noise / unreproducible). That's a
@@ -399,7 +382,9 @@ export async function runSampleFromTraces(flags, lang) {
     catch (err) {
         if (err instanceof CliExit)
             throw err;
-        console.error(lang === 'zh' ? `生成失败: ${err.message}` : `Generation failed: ${err.message}`);
+        const message = err.message;
+        console.error((lang === 'zh' ? `生成失败: ${message}` : `Generation failed: ${message}`)
+            + formatSampleGenerationFailureHint(message, flags.executor, lang));
         throw new CliExit(1);
     }
 }
@@ -427,6 +412,10 @@ async function runSample(args, flags, lang) {
         ? Math.max(1, Number(flags.count) || 5)
         : undefined;
     const model = flags.model;
+    const executorName = flags.executor;
+    if (!executorName) {
+        throw new Error('internal error: sample generation requires runtime selection before execution');
+    }
     const focus = flags.focus || undefined;
     if (focus) {
         process.stderr.write(tCli('cli.gen.focus_applied', lang, { focus }));
@@ -439,6 +428,7 @@ async function runSample(args, flags, lang) {
         }
         const entries = readdirSync(skillDir);
         let generated = 0;
+        let failed = 0;
         for (const entry of entries) {
             let name;
             let skillPath;
@@ -477,7 +467,7 @@ async function runSample(args, flags, lang) {
             }
             try {
                 const skillContent = readFileSync(skillPath, 'utf-8');
-                const { samples, costUSD } = await generateSamples({ skillContent, count, model, focus, noMock: flags['no-mock'], executorName: flags.executor });
+                const { samples, costUSD } = await generateSamples({ skillContent, count, model, focus, noMock: flags['no-mock'], executorName });
                 mkdirSync(dirname(samplesPath), { recursive: true });
                 writeFileSync(samplesPath, JSON.stringify(samples, null, 2));
                 const cost = costUSD > 0 ? ` $${costUSD.toFixed(4)}` : '';
@@ -487,11 +477,17 @@ async function runSample(args, flags, lang) {
                 generated++;
             }
             catch (err) {
+                failed++;
+                const message = err.message;
                 process.stderr.write(tCli('cli.gen.skill_failed', lang, {
-                    name, message: err.message,
+                    name, message: `${message}${formatSampleGenerationFailureHint(message, flags.executor, lang)}`,
                 }));
             }
         }
+        if (failed > 0) {
+            console.error(tCli('cli.gen.batch_failed_summary', lang, { generated, failed }));
+            throw new CliExit(1);
+        }
         if (generated === 0) {
             console.log(tCli('cli.gen.batch_none_needed', lang));
         }
@@ -531,7 +527,7 @@ async function runSample(args, flags, lang) {
         }
         // 已有用例文件:默认报错保护;--append 时追加(下面合并),不报错。
         if (existingFile && !flags.append) {
-            console.error(tCli('cli.gen.samples_already_exists', lang));
+            console.error(tCli('cli.gen.samples_already_exists', lang, { command: sampleNextEvalCommand(resolved) }));
             throw new CliExit(1);
         }
         if (count !== undefined) {
@@ -541,7 +537,7 @@ async function runSample(args, flags, lang) {
             process.stderr.write(tCli('cli.gen.single_generating_auto', lang));
         }
         try {
-            const { samples, costUSD } = await generateSamples({ skillContent, count, model, focus, noMock: flags['no-mock'], executorName: flags.executor });
+            const { samples, costUSD } = await generateSamples({ skillContent, count, model, focus, noMock: flags['no-mock'], executorName });
             const cost = costUSD > 0 ? ` $${costUSD.toFixed(4)}` : '';
             if (existingFile && flags.append) {
                 // 追加:读已有 → 合并(撞 id 去重)→ 保留原 json/yaml 格式与 wrapper 写回。
@@ -560,12 +556,15 @@ async function runSample(args, flags, lang) {
                     n: samples.length, path: outputPath, cost,
                 }));
             }
-            console.log(tCli('cli.gen.review_hint', lang));
+            console.log(tCli('cli.gen.review_hint', lang, { command: sampleNextEvalCommand(resolved) }));
         }
         catch (err) {
             if (err instanceof CliExit)
                 throw err;
-            console.error(tCli('cli.gen.failed', lang, { message: err.message }));
+            const message = err.message;
+            console.error(tCli('cli.gen.failed', lang, {
+                message: `${message}${formatSampleGenerationFailureHint(message, flags.executor, lang)}`,
+            }));
             throw new CliExit(1);
         }
     }
@@ -632,15 +631,14 @@ export default class Sample extends BaseCommand {
         }),
         model: Flags.string({
             description: bilingual({
-                zh: '生成 LLM model 名，默认 sonnet。',
-                en: 'Generation LLM model name, default sonnet.',
+                zh: '生成 LLM model 名。Codex 自动读取本机配置；也可用 OMK_MODEL 设置环境偏好。',
+                en: 'Generation LLM model name. Codex reads the local configured model; OMK_MODEL sets an environment preference.',
             }),
-            default: 'sonnet',
         }),
         executor: Flags.string({
             description: bilingual({
-                zh: '执行器名，默认 claude（同 omk eval / doctor / evolve）。指定 codex 等其它执行器时，记得连带传一个该执行器能识别的 --model。',
-                en: 'Executor name, default claude (same as omk eval / doctor / evolve). When using another executor like codex, also pass a --model it recognizes.',
+                zh: '执行器名。Codex 任务内自动用 codex；也可用 OMK_EXECUTOR 设置环境偏好。',
+                en: 'Executor name. Defaults to codex inside Codex tasks; OMK_EXECUTOR sets an environment preference.',
             }),
         }),
         'skill-dir': Flags.string({
@@ -713,7 +711,13 @@ export default class Sample extends BaseCommand {
         const { args, flags } = await this.parse(Sample);
         const lang = this.lang;
         await this.runWithCliExit(async () => {
-            await runSample(args, { ...flags, lang }, lang);
+            const runtime = resolveRuntimeSelection({ executor: flags.executor, model: flags.model }, { lang });
+            await runSample(args, {
+                ...flags,
+                executor: runtime.executor,
+                model: runtime.model,
+                lang,
+            }, lang);
         });
     }
 }

package/dist/cli/lib/codex-model-hint.d.ts

@@ -0,0 +1,9 @@
+export interface CodexModelSuggestion {
+    model: string;
+    fromConfig: boolean;
+    configPath?: string;
+}
+export declare function getCodexModelSuggestion(env?: NodeJS.ProcessEnv): CodexModelSuggestion;
+export declare function codexModelHint(lang: 'zh' | 'en', env?: NodeJS.ProcessEnv): string;
+export declare function codexModelFlagValue(env?: NodeJS.ProcessEnv): string;
+export declare function codexExecutorFlags(env?: NodeJS.ProcessEnv): string;

package/dist/cli/lib/codex-model-hint.js

@@ -0,0 +1,45 @@
+import { existsSync, readFileSync } from 'node:fs';
+import { homedir } from 'node:os';
+import { join } from 'node:path';
+const CODEX_MODEL_PLACEHOLDER = '<codex-model>';
+function parseTopLevelCodexModel(configText) {
+    for (const line of configText.split(/\r?\n/)) {
+        if (/^\s*\[/.test(line))
+            return null;
+        const match = line.match(/^\s*model\s*=\s*(?:"([^"]+)"|'([^']+)')\s*(?:#.*)?$/);
+        const model = match?.[1] ?? match?.[2];
+        if (model)
+            return model;
+    }
+    return null;
+}
+export function getCodexModelSuggestion(env = process.env) {
+    const codexHome = env.CODEX_HOME || join(homedir(), '.codex');
+    const configPath = join(codexHome, 'config.toml');
+    if (existsSync(configPath)) {
+        try {
+            const model = parseTopLevelCodexModel(readFileSync(configPath, 'utf-8'));
+            if (model)
+                return { model, fromConfig: true, configPath };
+        }
+        catch { /* best-effort hint only */ }
+    }
+    return { model: CODEX_MODEL_PLACEHOLDER, fromConfig: false, configPath };
+}
+export function codexModelHint(lang, env = process.env) {
+    const suggestion = getCodexModelSuggestion(env);
+    if (suggestion.fromConfig) {
+        return lang === 'zh'
+            ? `已按本机 Codex 配置 model=${suggestion.model} 填入。`
+            : `Filled from local Codex config model=${suggestion.model}.`;
+    }
+    return lang === 'zh'
+        ? `把 ${CODEX_MODEL_PLACEHOLDER} 换成本机 Codex 可用模型；可查看 ${suggestion.configPath} 的 model。`
+        : `Replace ${CODEX_MODEL_PLACEHOLDER} with a model your local Codex can run; check model in ${suggestion.configPath}.`;
+}
+export function codexModelFlagValue(env = process.env) {
+    return getCodexModelSuggestion(env).model;
+}
+export function codexExecutorFlags(env = process.env) {
+    return `--executor codex --model ${codexModelFlagValue(env)}`;
+}

package/dist/cli/lib/generation-failure-hint.d.ts

@@ -0,0 +1,2 @@
+import { type CliLang } from './i18n.js';
+export declare function formatSampleGenerationFailureHint(message: string, executorName: string | undefined, lang: CliLang, env?: NodeJS.ProcessEnv): string;

package/dist/cli/lib/generation-failure-hint.js

@@ -0,0 +1,61 @@
+import { tCli } from './i18n.js';
+import { codexExecutorFlags, codexModelFlagValue, codexModelHint } from './codex-model-hint.js';
+import { looksLikeLlmSetupFailure, looksLikeModelUnavailableFailure } from './llm-failure-classifier.js';
+const CLAUDE_SAMPLE_EXECUTORS = new Set(['claude', 'claude-sdk']);
+const CODEX_SAMPLE_EXECUTORS = new Set(['codex', 'codex-sdk']);
+const OPENAI_API_SAMPLE_EXECUTORS = new Set(['openai-api']);
+const ANTHROPIC_API_SAMPLE_EXECUTORS = new Set(['anthropic-api']);
+export function formatSampleGenerationFailureHint(message, executorName, lang, env = process.env) {
+    const executor = executorName?.trim();
+    if (!executor)
+        return '';
+    if (!looksLikeLlmSetupFailure(message))
+        return '';
+    if (CLAUDE_SAMPLE_EXECUTORS.has(executor)) {
+        return tCli('cli.gen.claude_auth_hint', lang, {
+            codexFlags: codexExecutorFlags(env),
+            codexModelHint: codexModelHint(lang, env),
+            openaiFlags: '--executor openai-api --model <openai-model>',
+        });
+    }
+    if (CODEX_SAMPLE_EXECUTORS.has(executor)) {
+        if (looksLikeModelUnavailableFailure(message)) {
+            return tCli('cli.gen.codex_model_hint', lang, {
+                codexFlags: codexExecutorFlags(env),
+                codexExec: `codex exec -m ${codexModelFlagValue(env)} "hi"`,
+                claudeFlags: '--executor claude --model sonnet',
+                openaiFlags: '--executor openai-api --model <openai-model>',
+                codexModelHint: codexModelHint(lang, env),
+            });
+        }
+        return tCli('cli.gen.codex_auth_hint', lang, {
+            claudeFlags: '--executor claude --model sonnet',
+            openaiFlags: '--executor openai-api --model <openai-model>',
+        });
+    }
+    if (OPENAI_API_SAMPLE_EXECUTORS.has(executor)) {
+        if (looksLikeModelUnavailableFailure(message)) {
+            return tCli('cli.gen.openai_api_model_hint', lang, {
+                claudeFlags: '--executor claude --model sonnet',
+                codexFlags: codexExecutorFlags(env),
+                codexModelHint: codexModelHint(lang, env),
+            });
+        }
+        return tCli('cli.gen.openai_api_auth_hint', lang, {
+            claudeFlags: '--executor claude --model sonnet',
+            codexFlags: codexExecutorFlags(env),
+            codexModelHint: codexModelHint(lang, env),
+        });
+    }
+    if (ANTHROPIC_API_SAMPLE_EXECUTORS.has(executor)) {
+        if (looksLikeModelUnavailableFailure(message)) {
+            return tCli('cli.gen.anthropic_api_model_hint', lang, {
+                claudeFlags: '--executor claude --model sonnet',
+            });
+        }
+        return tCli('cli.gen.anthropic_api_auth_hint', lang, {
+            claudeFlags: '--executor claude --model sonnet',
+        });
+    }
+    return '';
+}

package/dist/cli/lib/i18n-dict/common.d.ts

@@ -1,3 +1,3 @@
 import type { CliMessage } from './types.js';
-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.progress_skill_start' | 'cli.doctor.progress_skill_done';
+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.common.samples_not_found_hint' | '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.progress_skill_start' | 'cli.doctor.progress_skill_done';
 export declare const commonDict: Record<CommonMessageKey, CliMessage>;

package/dist/cli/lib/i18n-dict/common.js

@@ -43,6 +43,10 @@ export const commonDict = {
         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。',
         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.',
     },
+    'cli.common.samples_not_found_hint': {
+        zh: '下一步：先运行 {command} 生成用例，人工 review 后再重跑 omk eval。',
+        en: 'Next: run {command} to generate samples, review them, then re-run omk eval.',
+    },
     'cli.update.new_version_available': {
         zh: '\n💡 新版本可用：{old} → {new}，运行 npm i -g oh-my-knowledge@latest 升级\n\n',
         en: '\n💡 New version available: {old} → {new}, run npm i -g oh-my-knowledge@latest to upgrade\n\n',

package/dist/cli/lib/i18n-dict/gen.d.ts

@@ -1,3 +1,3 @@
 import type { CliMessage } from './types.js';
-export type GenMessageKey = 'cli.gen.skill_skipped_existing' | 'cli.gen.skill_generating' | 'cli.gen.skill_generating_auto' | 'cli.gen.skill_done' | 'cli.gen.skill_failed' | 'cli.gen.batch_none_needed' | 'cli.gen.batch_summary' | 'cli.gen.specify_skill_path' | 'cli.gen.samples_already_exists' | 'cli.gen.single_generating' | 'cli.gen.single_generating_auto' | 'cli.gen.single_done' | 'cli.gen.append_done' | 'cli.gen.append_single_only' | 'cli.gen.review_hint' | 'cli.gen.failed' | 'cli.gen.focus_applied';
+export type GenMessageKey = 'cli.gen.skill_skipped_existing' | 'cli.gen.skill_generating' | 'cli.gen.skill_generating_auto' | 'cli.gen.skill_done' | 'cli.gen.skill_failed' | 'cli.gen.batch_none_needed' | 'cli.gen.batch_failed_summary' | 'cli.gen.batch_summary' | 'cli.gen.specify_skill_path' | 'cli.gen.samples_already_exists' | 'cli.gen.single_generating' | 'cli.gen.single_generating_auto' | 'cli.gen.single_done' | 'cli.gen.append_done' | 'cli.gen.append_single_only' | 'cli.gen.review_hint' | 'cli.gen.claude_auth_hint' | 'cli.gen.codex_auth_hint' | 'cli.gen.codex_model_hint' | 'cli.gen.openai_api_auth_hint' | 'cli.gen.openai_api_model_hint' | 'cli.gen.anthropic_api_auth_hint' | 'cli.gen.anthropic_api_model_hint' | 'cli.gen.failed' | 'cli.gen.focus_applied';
 export declare const genDict: Record<GenMessageKey, CliMessage>;

package/dist/cli/lib/i18n-dict/gen.js

@@ -23,17 +23,21 @@ export const genDict = {
         zh: '没有需要生成的 eval-samples (所有 skill 都已有配对文件)',
         en: 'No eval-samples need generating (all skills already have paired files)',
     },
+    'cli.gen.batch_failed_summary': {
+        zh: '\n生成未完成：已生成 {generated} 份，失败 {failed} 份。请按上面的错误提示修复后重试；全部生成成功后再运行 omk eval --batch --dry-run。',
+        en: '\nGeneration incomplete: generated {generated}, failed {failed}. Fix the errors above and retry; once everything succeeds, run omk eval --batch --dry-run.',
+    },
     'cli.gen.batch_summary': {
-        zh: '\n共生成 {n} 份 eval-samples, 请审查后运行: omk eval --batch',
-        en: '\nGenerated {n} eval-samples files. Review them, then run: omk eval --batch',
+        zh: '\n共生成 {n} 份 eval-samples。下一步：\n  1. 人工审查生成的评测用例，删掉不可信样本，补边界、反例\n  2. 预览任务：omk eval --batch --dry-run\n  3. 跑评测：omk eval --batch',
+        en: '\nGenerated {n} eval-samples files. Next steps:\n  1. Review the generated samples; drop weak cases and add boundary / counterexamples\n  2. Preview the task plan: omk eval --batch --dry-run\n  3. Run the eval: omk eval --batch',
     },
     'cli.gen.specify_skill_path': {
         zh: '请指定 skill 文件路径, 例如: omk sample skills/my-skill.md',
         en: 'Please specify a skill file path, e.g.: omk sample skills/my-skill.md',
     },
     'cli.gen.samples_already_exists': {
-        zh: 'eval-samples.json 已存在。如需覆盖请先删除该文件。',
-        en: 'eval-samples.json already exists. Delete it first if you want to overwrite.',
+        zh: 'eval-samples 已存在。要补场景请加 --append（常配 --focus）；要继续评测，运行：{command}',
+        en: 'eval-samples already exist. To add scenarios, use --append (often with --focus); to continue, run: {command}',
     },
     'cli.gen.single_generating': {
         zh: '🔄 正在生成 {count} 条评测用例...\n',
@@ -56,8 +60,36 @@ export const genDict = {
         en: '--append currently supports single-skill mode only; it cannot be combined with --batch / --from-traces / --fix.\n',
     },
     'cli.gen.review_hint': {
-        zh: '\n请审查生成的评测用例后运行: omk eval',
-        en: '\nReview the generated test cases, then run: omk eval',
+        zh: '\n下一步：\n  1. 人工审查生成的评测用例，删掉不可信样本，补边界、反例\n  2. 预览任务：{command} --dry-run\n  3. 跑评测：{command}',
+        en: '\nNext steps:\n  1. Review the generated samples; drop weak cases and add boundary / counterexamples\n  2. Preview the task plan: {command} --dry-run\n  3. Run the eval: {command}',
+    },
+    'cli.gen.claude_auth_hint': {
+        zh: '\n提示：当前 sample 生成使用 Claude 系列执行器。先确认 Claude Code 已安装并完成登录；如果你在 Codex 环境里，可以改用：{codexFlags}（{codexModelHint}）；如果要走 OpenAI API，可以改用：{openaiFlags}，并设置 OPENAI_API_KEY。',
+        en: '\nHint: sample generation is using a Claude-based executor. First confirm Claude Code is installed and authenticated; in a Codex environment, switch to: {codexFlags} ({codexModelHint}); to use the OpenAI API path, switch to: {openaiFlags}, and set OPENAI_API_KEY.',
+    },
+    'cli.gen.codex_auth_hint': {
+        zh: '\n提示：当前 sample 生成使用 Codex 系列执行器。先确认 Codex CLI / SDK 已安装并完成登录；如果你有 Claude Code 可用，可以改用：{claudeFlags}；如果要走 OpenAI API，可以改用：{openaiFlags}，并设置 OPENAI_API_KEY。',
+        en: '\nHint: sample generation is using a Codex-based executor. First confirm the Codex CLI / SDK is installed and authenticated; if Claude Code is available, switch to: {claudeFlags}; to use the OpenAI API path, switch to: {openaiFlags}, and set OPENAI_API_KEY.',
+    },
+    'cli.gen.codex_model_hint': {
+        zh: '\n提示：当前 sample 生成使用 Codex 系列执行器，但模型名看起来不可用。可以先按本机 Codex 配置重试：{codexFlags}（{codexModelHint}）；也可以先运行 `{codexExec}` 验证模型是否可用。若只是想先跑通，可以改用：{claudeFlags}；或走 OpenAI API：{openaiFlags}，并设置 OPENAI_API_KEY。',
+        en: '\nHint: sample generation is using a Codex-based executor, but the model name appears unavailable. Retry with the local Codex config model: {codexFlags} ({codexModelHint}); you can also run `{codexExec}` to verify the model. To just get a first run through, switch to: {claudeFlags}; or use the OpenAI API path: {openaiFlags}, and set OPENAI_API_KEY.',
+    },
+    'cli.gen.openai_api_auth_hint': {
+        zh: '\n提示：当前 sample 生成使用 OpenAI API 执行器。请检查 OPENAI_API_KEY / OPENAI_BASE_URL 是否可用，并确认模型名对当前端点可用；如果只是想先跑通，也可以改用：{claudeFlags}，或：{codexFlags}（{codexModelHint}）。',
+        en: '\nHint: sample generation is using the OpenAI API executor. Check OPENAI_API_KEY / OPENAI_BASE_URL and confirm the model is available on that endpoint; to just get a first run through, you can also switch to: {claudeFlags}, or: {codexFlags} ({codexModelHint}).',
+    },
+    'cli.gen.openai_api_model_hint': {
+        zh: '\n提示：当前 sample 生成使用 OpenAI API 执行器，但模型名看起来对当前端点不可用。请检查 --model、OPENAI_BASE_URL 与账号权限是否匹配；如果只是想先跑通，也可以改用：{claudeFlags}，或：{codexFlags}（{codexModelHint}）。',
+        en: '\nHint: sample generation is using the OpenAI API executor, but the model name appears unavailable on the current endpoint. Check --model, OPENAI_BASE_URL, and account access; to just get a first run through, you can also switch to: {claudeFlags}, or: {codexFlags} ({codexModelHint}).',
+    },
+    'cli.gen.anthropic_api_auth_hint': {
+        zh: '\n提示：当前 sample 生成使用 Anthropic API 执行器。请检查 ANTHROPIC_API_KEY / ANTHROPIC_BASE_URL 是否可用，并确认模型名对当前端点可用；如果你有 Claude Code 可用，也可以改用：{claudeFlags}。',
+        en: '\nHint: sample generation is using the Anthropic API executor. Check ANTHROPIC_API_KEY / ANTHROPIC_BASE_URL and confirm the model is available on that endpoint; if Claude Code is available, you can also switch to: {claudeFlags}.',
+    },
+    'cli.gen.anthropic_api_model_hint': {
+        zh: '\n提示：当前 sample 生成使用 Anthropic API 执行器，但模型名看起来对当前端点不可用。请检查 --model、ANTHROPIC_BASE_URL 与账号权限是否匹配；如果你有 Claude Code 可用，也可以改用：{claudeFlags}。',
+        en: '\nHint: sample generation is using the Anthropic API executor, but the model name appears unavailable on the current endpoint. Check --model, ANTHROPIC_BASE_URL, and account access; if Claude Code is available, you can also switch to: {claudeFlags}.',
     },
     'cli.gen.failed': {
         zh: '生成失败: {message}',

package/dist/cli/lib/i18n-dict/help.js

@@ -133,9 +133,9 @@ omk evolve——多轮自动迭代改进 skill
 选项：
   --rounds <n>                        迭代轮数（默认：5）
   --target <score>                    目标分数
-  --model <name>                      任务执行模型，每轮跑 eval samples 的被测模型（默认：sonnet）
-  --improve-model <name>              skill 改写模型，每轮根据反馈改写 skill 的模型（默认：sonnet）
-  --judge-models <executor:model>     单评委配置（默认：claude:haiku）
+  --model <name>                      任务执行模型，默认跟随 runtime；Codex 读取本机配置
+  --improve-model <name>              skill 改写模型，默认沿用任务执行模型
+  --judge-models <executor:model>     单评委配置（默认跟随执行器；Codex 沿用被测模型）
 
 示例：
   omk evolve skills/code-review/SKILL.md
@@ -151,9 +151,9 @@ Usage:
 Options:
   --rounds <n>                        Iteration rounds (default: 5)
   --target <score>                    Target score
-  --model <name>                      Task executor model — runs eval samples each round (default: sonnet)
-  --improve-model <name>              Skill rewriter model — rewrites the skill each round (default: sonnet)
-  --judge-models <executor:model>     Single judge config (default: claude:haiku)
+  --model <name>                      Task executor model; follows runtime (Codex reads local config)
+  --improve-model <name>              Skill rewriter model; defaults to the task model
+  --judge-models <executor:model>     Single judge config (follows executor; Codex reuses task model)
 
 Examples:
   omk evolve skills/code-review/SKILL.md

package/dist/cli/lib/i18n-dict/init.d.ts

@@ -1,3 +1,3 @@
 import type { CliMessage } from './types.js';
-export type InitMessageKey = 'cli.init.scaffolded' | 'cli.init.next_steps_title' | 'cli.init.next_step_run' | 'cli.init.next_step_executor' | 'cli.init.next_step_customize' | 'cli.init.note_codex_executor';
+export type InitMessageKey = 'cli.init.scaffolded' | 'cli.init.next_steps_title' | 'cli.init.next_step_run' | 'cli.init.next_step_executor' | 'cli.init.next_step_report' | 'cli.init.next_step_customize' | 'cli.init.note_skill_injection';
 export declare const initDict: Record<InitMessageKey, CliMessage>;

package/dist/cli/lib/i18n-dict/init.js

@@ -11,19 +11,23 @@ export const initDict = {
     // 跑出第一份报告是冷启动最该先发生的事;「换成你自己的」放到跑通之后。这也消除了
     // 主 README「不用改任何文件」与旧 init「先编辑」的矛盾。
     'cli.init.next_step_run': {
-        zh: '  1. 直接跑通（无需先改任何文件）：omk eval --control code-review-v1 --treatment code-review-v2',
-        en: '  1. Run it as-is (no edits needed): omk eval --control code-review-v1 --treatment code-review-v2',
+        zh: '  1. 直接跑通（无需先改任何文件）：{command}',
+        en: '  1. Run it as-is (no edits needed): {command}',
+    },
+    'cli.init.next_step_report': {
+        zh: '  2. 看报告里的 verdict 和“下一步”：PROGRESS 才能发布；UNDERPOWERED / NOISE 先扩样到约 20 条以上后重跑。',
+        en: '  2. Read the report verdict and Next line: PROGRESS can ship; UNDERPOWERED / NOISE means grow to roughly 20+ samples and re-run.',
     },
     'cli.init.next_step_executor': {
-        zh: '     默认执行器与评委用 claude CLI，需先装好并登录；想换别的模型或离线跑（无需 API key）见文档「执行器」。',
-        en: '     The default executor and judge use the claude CLI (install and log in first); to use another model or run offline (no API key) see the Executors docs.',
+        zh: '     executor / judge 会按运行环境选择；Codex 任务自动使用本机 Codex 配置。也可用 OMK_EXECUTOR / OMK_MODEL 固定环境偏好，详见 https://oh-my-knowledge.pages.dev/zh/reference/executors。',
+        en: '     The executor / judge follow the runtime environment; Codex tasks use the local Codex configuration automatically. OMK_EXECUTOR / OMK_MODEL pin environment preferences. See https://oh-my-knowledge.pages.dev/reference/executors.',
     },
     'cli.init.next_step_customize': {
-        zh: '  2. 跑通后，把 skills/code-review-v1/SKILL.md 和 skills/code-review-v2/SKILL.md 与 eval-samples.json 换成你自己的 skill 和用例',
-        en: '  2. Once it runs, replace skills/code-review-v1/SKILL.md and skills/code-review-v2/SKILL.md and eval-samples.json with your own skills and cases',
+        zh: '  3. 跑通后，替换为你自己的 skill 和 eval-samples.json；还没有用例时先运行 omk sample <skill-path>。',
+        en: '  3. Once it runs, replace the starter skills and eval-samples.json with your own; if you have no samples yet, run omk sample <skill-path>.',
     },
-    'cli.init.note_codex_executor': {
-        zh: '\n注: omk 评测时把 SKILL.md 整文(含 frontmatter)作为 system prompt 注入——跨 executor 一致(claude / codex / openai-api / gemini 都走同一条路径,不依赖任何 executor 的 native skill auto-discovery 或 Skill 工具机制)。frontmatter 在 prompt 头部对 model 行为无显著影响。\n模板带 Claude Code 兼容的 frontmatter(name + description)是为了让同一份 directory-skill 也能 deploy 到 Claude Code:把整个目录复制到 ~/.claude/skills/code-review-v1/(整目录,不是单个 SKILL.md),Claude SDK 才能识别。这是 omk 评测之外的 bonus,一份文件双向 dogfood。',
-        en: '\nNote: during omk evaluation the full SKILL.md (frontmatter included) is injected as the system prompt — uniformly across executors (claude / codex / openai-api / gemini all take the same path; omk does not rely on any executor\'s native skill auto-discovery or Skill tool). Frontmatter has no measurable impact on model behavior in this position.\nThe template ships with Claude Code-compatible frontmatter (name + description) so the same directory-skill can also be deployed to Claude Code: copy the whole directory to ~/.claude/skills/code-review-v1/ (the directory, not just SKILL.md) so Claude SDK can recognize it. That is a bonus beyond omk evaluation — one source, two-way dogfood.',
+    'cli.init.note_skill_injection': {
+        zh: '     注：omk eval 会把 SKILL.md 作为 system prompt 注入；模板 frontmatter 只是方便同一目录复用为 agent skill。',
+        en: '     Note: omk eval injects SKILL.md as the system prompt; the starter frontmatter only helps reuse the same directory as an agent skill.',
     },
 };

package/dist/cli/lib/i18n-dict/run.d.ts

@@ -1,3 +1,3 @@
 import type { CliMessage } from './types.js';
-export type RunMessageKey = 'cli.progress.preflight_starting' | 'cli.progress.sample_retry' | 'cli.progress.sample_error' | 'cli.progress.sample_executing' | 'cli.progress.sample_exec_done' | 'cli.progress.output_preview' | 'cli.progress.judging' | 'cli.progress.judged' | 'cli.progress.skipped' | 'cli.progress.sample_done' | 'cli.progress.sample_failed_done' | 'cli.run.invalid_repeat' | 'cli.run.invalid_holdout_ratio' | 'cli.run.invalid_judge_repeat' | 'cli.run.no_debias_length_active' | 'cli.run.invalid_bootstrap_samples' | 'cli.run.bootstrap_samples_too_large' | 'cli.run.dry_run_no_scores' | 'cli.run.skill_section' | 'cli.run.run_section' | 'cli.run.batch_complete' | 'cli.run.batch_verdict_header' | 'cli.run.batch_verdict_next_step' | 'cli.run.batch_child_report_missing' | 'cli.run.eval_complete' | 'cli.run.tally' | 'cli.run.report_saved' | 'cli.run.evidence_recorded' | 'cli.run.evidence_recorded_unbound' | 'cli.run.report_only_gate_skipped' | 'cli.run.report_server_running' | 'cli.run.report_server_view' | 'cli.run.report_server_stop' | 'cli.run.no_serve_in_non_tty' | 'cli.run.no_serve_view_hint' | 'cli.run.gold_load_failed' | 'cli.run.gold_load_issue' | 'cli.run.contamination_warning' | 'cli.run.skip_connectivity_warning';
+export type RunMessageKey = 'cli.progress.preflight_starting' | 'cli.progress.sample_retry' | 'cli.progress.sample_error' | 'cli.progress.sample_executing' | 'cli.progress.sample_exec_done' | 'cli.progress.output_preview' | 'cli.progress.judging' | 'cli.progress.judged' | 'cli.progress.skipped' | 'cli.progress.sample_done' | 'cli.progress.sample_failed_done' | 'cli.run.invalid_repeat' | 'cli.run.invalid_holdout_ratio' | 'cli.run.invalid_judge_repeat' | 'cli.run.no_debias_length_active' | 'cli.run.invalid_bootstrap_samples' | 'cli.run.bootstrap_samples_too_large' | 'cli.run.dry_run_no_scores' | 'cli.run.skill_section' | 'cli.run.run_section' | 'cli.run.batch_complete' | 'cli.run.batch_verdict_header' | 'cli.run.batch_verdict_next_step' | 'cli.run.batch_child_report_missing' | 'cli.run.eval_complete' | 'cli.run.tally' | 'cli.run.report_saved' | 'cli.run.evidence_recorded' | 'cli.run.evidence_recorded_promotable' | 'cli.run.evidence_recorded_unbound' | 'cli.run.report_only_gate_skipped' | 'cli.run.report_server_running' | 'cli.run.report_server_view' | 'cli.run.report_server_stop' | 'cli.run.no_serve_in_non_tty' | 'cli.run.no_serve_view_hint' | 'cli.run.gold_load_failed' | 'cli.run.gold_load_issue' | 'cli.run.contamination_warning' | 'cli.run.codex_fallback_hint' | 'cli.run.codex_auth_hint' | 'cli.run.codex_model_hint' | 'cli.run.openai_api_auth_hint' | 'cli.run.openai_api_model_hint' | 'cli.run.anthropic_api_auth_hint' | 'cli.run.anthropic_api_model_hint' | 'cli.run.skip_connectivity_warning';
 export declare const runDict: Record<RunMessageKey, CliMessage>;