oh-my-knowledge 0.32.0 → 0.34.0

package/dist/cli/index.js

@@ -1,6 +1,5 @@
 #!/usr/bin/env node
 import { getCliLang, parseLangFromArgv } from './lib/i18n.js';
-import { checkUpdate } from './lib/update-check.js';
 import { CliExit } from './lib/cli-exit.js';
 // CLI 入口:lang / 版本提醒等共享前置逻辑跑完,把控制权交给 oclif dispatcher。
 // 所有命令统一走 src/cli/commands/* 下的 oclif Command。
@@ -14,6 +13,7 @@ function isShortPath(argv) {
 async function main() {
     const lang = getCliLang(parseLangFromArgv(process.argv));
     if (!isShortPath(process.argv)) {
+        const { checkUpdate } = await import('./lib/update-check.js');
         checkUpdate(lang);
     }
     const { runOclifPath } = await import('./oclif/run.js');
@@ -26,4 +26,3 @@ main().catch((err) => {
     console.error(err);
     process.exit(1);
 });
-//# sourceMappingURL=index.js.map

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

@@ -13,4 +13,3 @@ export declare class CliExit extends Error {
     readonly code: number;
     constructor(code: number);
 }
-//# sourceMappingURL=cli-exit.d.ts.map

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

@@ -17,4 +17,3 @@ export class CliExit extends Error {
         this.name = 'CliExit';
     }
 }
-//# sourceMappingURL=cli-exit.js.map

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

@@ -60,6 +60,8 @@ export interface DoctorFlags {
     model?: string;
     samples?: string;
     timeout?: string;
+    'output-dir'?: string;
+    dimensions?: string;
     'static-only': boolean;
 }
 export type EvalArgs = Record<string, never>;
@@ -130,6 +132,13 @@ export interface EvolveFlags {
     'sample-fix-max-attempts': string;
     'reuse-latest-eval': boolean;
     'improve-mode': string;
+    'holdout-ratio': string;
+    'no-significance-gate': boolean;
+    'significance-alpha': string;
+    'test-ratio': string;
+    'edit-budget': string;
+    'no-edit-budget': boolean;
+    'no-reject-memory': boolean;
 }
 export interface SampleArgs {
     skillPath?: string;
@@ -145,6 +154,8 @@ export interface SampleFlags {
     fix: boolean;
     'reports-dir'?: string;
     treatment?: string;
+    'from-traces': boolean;
+    'observations-dir'?: string;
 }
 export type GoldInitArgs = Record<string, never>;
 export interface GoldInitFlags {
@@ -221,4 +232,3 @@ export interface StudioFlags {
     'no-open': boolean;
     dev: boolean;
 }
-//# sourceMappingURL=cmd-flags.d.ts.map

package/dist/cli/lib/cmd-flags.js

@@ -43,4 +43,3 @@
  * `default: false` 的 boolean 必须在 interface 里标成可选(`field?: boolean`)。
  */
 export {};
-//# sourceMappingURL=cmd-flags.js.map

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

@@ -1,4 +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.update.new_version_available' | 'cli.observe.view_hint' | 'cli.studio.started' | 'cli.studio.stop_hint' | 'cli.studio.open_failed' | 'cli.doctor.no_skill_found' | 'cli.doctor.samples_detected';
+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.studio.started' | 'cli.studio.stop_hint' | 'cli.studio.open_failed' | 'cli.doctor.no_skill_found' | 'cli.doctor.samples_detected';
 export declare const commonDict: Record<CommonMessageKey, CliMessage>;
-//# sourceMappingURL=common.d.ts.map

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

@@ -36,8 +36,24 @@ export const commonDict = {
         en: '⚠ Failed to load samples file ({path}): {message}\n',
     },
     'cli.update.new_version_available': {
-        zh: '\n💡 新版本可用: {old} → {new}, 运行 npm update {pkg} -g 升级\n\n',
-        en: '\n💡 New version available: {old} → {new}, run npm update {pkg} -g to upgrade\n\n',
+        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',
+    },
+    'cli.update.box_title': {
+        zh: '↑ omk 有新版本',
+        en: '↑ omk update available',
+    },
+    'cli.update.box_version_line': {
+        zh: '当前 {old} → 最新 {new}',
+        en: 'current {old} → latest {new}',
+    },
+    'cli.update.box_upgrade_line': {
+        zh: '升级：{cmd}',
+        en: 'Upgrade: {cmd}',
+    },
+    'cli.update.box_silence_line': {
+        zh: '静默：设 {env} 关闭提醒',
+        en: 'Silence: set {env} to disable',
     },
     'cli.observe.view_hint': {
         zh: '分析 JSON 已写入 output-dir；后续可用 omk observe 持续生成日报。',
@@ -64,4 +80,3 @@ export const commonDict = {
         en: '✓ Using eval samples file: {path}',
     },
 };
-//# sourceMappingURL=common.js.map

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

@@ -1,4 +1,3 @@
 import type { CliMessage } from './types.js';
-export type EvolveMessageKey = 'cli.evolve.specify_skill_path' | 'cli.evolve.section_header' | 'cli.evolve.round_baseline' | 'cli.evolve.round_baseline_reused' | 'cli.evolve.round_error' | 'cli.evolve.round_done' | 'cli.evolve.summary' | 'cli.evolve.best_path' | 'cli.evolve.versions_saved' | 'cli.evolve.report_link';
+export type EvolveMessageKey = 'cli.evolve.specify_skill_path' | 'cli.evolve.section_header' | 'cli.evolve.round_baseline' | 'cli.evolve.round_baseline_reused' | 'cli.evolve.round_error' | 'cli.evolve.round_done' | 'cli.evolve.summary' | 'cli.evolve.best_path' | 'cli.evolve.versions_saved' | 'cli.evolve.report_link' | 'cli.evolve.holdout_active' | 'cli.evolve.holdout_disabled' | 'cli.evolve.gate_underpowered' | 'cli.evolve.generalization' | 'cli.evolve.test_disabled' | 'cli.evolve.reject_not_significant';
 export declare const evolveDict: Record<EvolveMessageKey, CliMessage>;
-//# sourceMappingURL=evolve.d.ts.map

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

@@ -39,5 +39,28 @@ export const evolveDict = {
         zh: '📊 查看报告：omk studio（报告 ID：{id}）\n',
         en: '📊 View report: omk studio (report id: {id})\n',
     },
+    'cli.evolve.holdout_active': {
+        zh: '🔀 holdout 切分：训练 {train} / 验收 {holdout}，accept 看 holdout 分（防 train-on-test）\n',
+        en: '🔀 Holdout split: {train} train / {holdout} holdout; accepted on holdout score (guards train-on-test)\n',
+    },
+    'cli.evolve.holdout_disabled': {
+        zh: '⚠️ holdout-ratio={ratio} 样本太少无法切分，已回退全集打分\n',
+        en: '⚠️ holdout-ratio={ratio} too few samples to split; using full-set scoring\n',
+    },
+    'cli.evolve.gate_underpowered': {
+        zh: '⚠️ 决策样本量不足以做显著性门禁，本轮退回点估计接受；增大用例集或降 --holdout-ratio 可启用严格门\n',
+        en: '⚠️ Too few decision samples for the significance gate; fell back to point-estimate accept. Add samples or lower --holdout-ratio to enable the strict gate\n',
+    },
+    'cli.evolve.generalization': {
+        zh: '🔒 锁定 test 集（{count} 条，全程不参与选择）：泛化分 {score}\n',
+        en: '🔒 Locked test set ({count} samples, never used for selection): generalization {score}\n',
+    },
+    'cli.evolve.reject_not_significant': {
+        zh: '（差异不显著）',
+        en: ' (not significant)',
+    },
+    'cli.evolve.test_disabled': {
+        zh: '⚠️ test-ratio={ratio} 样本太少无法做三路切分，已回退两路 holdout，本次无泛化分\n',
+        en: '⚠️ test-ratio={ratio} too few samples for a 3-way split; fell back to 2-way holdout, no generalization score this run\n',
+    },
 };
-//# sourceMappingURL=evolve.js.map

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

@@ -1,4 +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.review_hint' | 'cli.gen.failed' | 'cli.gen.focus_applied';
 export declare const genDict: Record<GenMessageKey, CliMessage>;
-//# sourceMappingURL=gen.d.ts.map

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

@@ -60,4 +60,3 @@ export const genDict = {
         en: '🎯 Focus scenarios (--focus): {focus}\n',
     },
 };
-//# sourceMappingURL=gen.js.map

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

@@ -1,4 +1,3 @@
 import type { CliMessage } from './types.js';
 export type HelpMessageKey = 'cli.help.observe' | 'cli.help.observe_ingest' | 'cli.help.observe_inbox' | 'cli.help.observe_show' | 'cli.help.evolve' | 'cli.help.sample' | 'cli.help.studio';
 export declare const helpDict: Record<HelpMessageKey, CliMessage>;
-//# sourceMappingURL=help.d.ts.map

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

@@ -278,4 +278,3 @@ Examples:
 `,
     },
 };
-//# sourceMappingURL=help.js.map

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

@@ -1,4 +1,3 @@
 import type { CliMessage } from './types.js';
 export type InitMessageKey = 'cli.init.scaffolded' | 'cli.init.next_steps_title' | 'cli.init.next_step_edit_samples' | 'cli.init.next_step_edit_skills' | 'cli.init.next_step_run' | 'cli.init.note_codex_executor';
 export declare const initDict: Record<InitMessageKey, CliMessage>;
-//# sourceMappingURL=init.d.ts.map

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

@@ -24,4 +24,3 @@ export const initDict = {
         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.',
     },
 };
-//# sourceMappingURL=init.js.map

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

@@ -0,0 +1,3 @@
+import type { CliMessage } from './types.js';
+export type InstallMessageKey = 'cli.install.asset_missing' | 'cli.install.unknown_input' | 'cli.install.unknown_target' | 'cli.install.invalid_target_combo' | 'cli.install.no_detected_targets' | 'cli.install.target_exists' | 'cli.install.plan' | 'cli.install.plan_skill' | 'cli.install.installed' | 'cli.install.kind_unsupported' | 'cli.install.copied' | 'cli.install.adopted' | 'cli.install.registered' | 'cli.install.target_overlaps_source' | 'cli.install.path_not_found' | 'cli.install.skillmd_missing' | 'cli.install.skillmd_is_symlink' | 'cli.install.not_a_skill' | 'cli.install.not_a_git_repo' | 'cli.install.git_skill_not_found' | 'cli.install.git_unsafe_path' | 'cli.install.next_hint';
+export declare const installDict: Record<InstallMessageKey, CliMessage>;

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

@@ -0,0 +1,90 @@
+export const installDict = {
+    'cli.install.asset_missing': {
+        zh: '未找到打包内置资源：{path}。请先运行 yarn build，或确认 npm 包包含 dist/assets/agent-skills/omk。',
+        en: 'Packaged built-in asset not found: {path}. Run yarn build first, or verify the npm package includes dist/assets/agent-skills/omk.',
+    },
+    'cli.install.unknown_input': {
+        zh: 'install 接受内置 id omk-agent-skill、本地 skill 路径（如 ./skills/review、./review.md），或 git:<ref>:<name>（当前仓库某 ref 的 skill）。无法识别的输入：{input}',
+        en: 'install accepts the built-in id omk-agent-skill, a local skill path (e.g. ./skills/review or ./review.md), or git:<ref>:<name> (a skill at a ref of the current repo). Unrecognized input: {input}',
+    },
+    'cli.install.unknown_target': {
+        zh: '未知安装目标：{target}。可用值：auto, codex, claude, all；或用 --dest <dir> 指定 skill 根目录。',
+        en: 'Unknown install target: {target}. Use auto, codex, claude, all; or pass --dest <dir> for a custom skill root.',
+    },
+    'cli.install.invalid_target_combo': {
+        zh: '安装目标组合不合法：{target}。auto 和 all 必须单独使用；如需多个明确目标，请写 codex,claude。',
+        en: 'Invalid install target combination: {target}. auto and all must be used alone; use codex,claude for multiple explicit targets.',
+    },
+    'cli.install.no_detected_targets': {
+        zh: '未检测到本机支持的 agent skill 目录。请用 --to codex / --to claude / --to all 明确目标，或用 --dest <dir> 指定 skill 根目录。',
+        en: 'No supported local agent skill directory was detected. Pass --to codex / --to claude / --to all, or pass --dest <dir> for a custom skill root.',
+    },
+    'cli.install.target_exists': {
+        zh: '目标已存在：{path}。如要覆盖，请加 --force。',
+        en: 'Target already exists: {path}. Pass --force to overwrite.',
+    },
+    'cli.install.plan': {
+        zh: '将安装 omk Agent Skill 到：{path}',
+        en: 'Will install the omk Agent Skill to: {path}',
+    },
+    'cli.install.plan_skill': {
+        zh: '将安装 skill {name} 到：{path}',
+        en: 'Will install skill {name} to: {path}',
+    },
+    'cli.install.installed': {
+        zh: '已安装 omk Agent Skill：{path}',
+        en: 'Installed omk Agent Skill: {path}',
+    },
+    'cli.install.kind_unsupported': {
+        zh: '暂不支持安装 kind 为 {kind} 的知识输入；当前仅支持 skill（prompt / agent / workflow 的 artifact 模型尚未定义）。',
+        en: 'Installing a {kind} knowledge input is not yet supported; only skill is supported today (the artifact model for prompt / agent / workflow is not yet defined).',
+    },
+    'cli.install.copied': {
+        zh: '已安装 skill {name}：{path}',
+        en: 'Installed skill {name}: {path}',
+    },
+    'cli.install.adopted': {
+        zh: '已就地接管 skill {name}（已在目标位置，未改动文件）：{path}',
+        en: 'Adopted skill {name} in place (already at target, files untouched): {path}',
+    },
+    'cli.install.registered': {
+        zh: '已登记受管记录 {id}：{store}',
+        en: 'Registered managed record {id}: {store}',
+    },
+    'cli.install.target_overlaps_source': {
+        zh: '安装目标与源相互嵌套，拒绝执行（会删掉源或自我复制）。源：{source}；目标：{target}。请换一个不与源重叠的 --dest。',
+        en: 'Install target overlaps the source (would delete the source or copy into itself); refused. Source: {source}; target: {target}. Use a --dest that does not overlap the source.',
+    },
+    'cli.install.path_not_found': {
+        zh: '找不到要安装的 skill 路径：{path}',
+        en: 'Skill path to install not found: {path}',
+    },
+    'cli.install.skillmd_missing': {
+        zh: '目录下没有 SKILL.md，不是一个 directory-skill：{path}',
+        en: 'No SKILL.md in the directory; not a directory-skill: {path}',
+    },
+    'cli.install.not_a_skill': {
+        zh: '不是 skill 文件（需要 .md 或含 SKILL.md 的目录）：{path}',
+        en: 'Not a skill file (expected a .md file or a directory containing SKILL.md): {path}',
+    },
+    'cli.install.skillmd_is_symlink': {
+        zh: 'SKILL.md 是指向 {target} 的软链，分发会跳过软链、装出来的 skill 会缺 SKILL.md。请直接 install 真源位置：{target}',
+        en: 'SKILL.md is a symlink to {target}; symlinks are skipped during distribution, so the installed skill would be missing its SKILL.md. Install the real source instead: {target}',
+    },
+    'cli.install.not_a_git_repo': {
+        zh: '当前目录不在 git 仓库内，无法解析 git: 源。请在仓库内执行，或改用本地路径。',
+        en: 'Current directory is not inside a git repo; cannot resolve a git: source. Run inside the repo, or use a local path.',
+    },
+    'cli.install.git_skill_not_found': {
+        zh: '在 git ref {ref} 下找不到 skill {name}（既无 {name}/SKILL.md 也无 {name}.md）。',
+        en: 'Skill {name} not found at git ref {ref} (neither {name}/SKILL.md nor {name}.md).',
+    },
+    'cli.install.git_unsafe_path': {
+        zh: 'git tree 含越界路径 {path}（.. / 绝对路径 / 空段），拒绝物化以防写出临时目录。该 skill 的 git tree 可能被手工构造，请核查来源。',
+        en: 'git tree contains an out-of-bounds path {path} (.. / absolute / empty segment); refusing to materialize to avoid escaping the temp dir. The skill git tree may be hand-crafted; verify the source.',
+    },
+    'cli.install.next_hint': {
+        zh: '现在可以在 coding agent 中说「用 omk 评测这个 skill」。',
+        en: 'You can now ask your coding agent: "use omk to evaluate this skill".',
+    },
+};

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

@@ -1,4 +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_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_child_report_missing' | 'cli.run.eval_complete' | 'cli.run.tally' | 'cli.run.report_saved' | '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 declare const runDict: Record<RunMessageKey, CliMessage>;
-//# sourceMappingURL=run.d.ts.map

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

@@ -140,4 +140,3 @@ export const runDict = {
         en: '⚠️  --skip-connectivity enabled: LLM connectivity check skipped. Verify executor / judge are reachable by other means.',
     },
 };
-//# sourceMappingURL=run.js.map

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

@@ -2,4 +2,3 @@ export interface CliMessage {
     zh: string;
     en: string;
 }
-//# sourceMappingURL=types.d.ts.map

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

@@ -1,2 +1 @@
 export {};
-//# sourceMappingURL=types.js.map

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

@@ -55,9 +55,9 @@ import { type EvolveMessageKey } from './i18n-dict/evolve.js';
 import { type GenMessageKey } from './i18n-dict/gen.js';
 import { type HelpMessageKey } from './i18n-dict/help.js';
 import { type InitMessageKey } from './i18n-dict/init.js';
+import { type InstallMessageKey } from './i18n-dict/install.js';
 import { type RunMessageKey } from './i18n-dict/run.js';
 import type { CliMessage } from './i18n-dict/types.js';
 export type { CliMessage } from './i18n-dict/types.js';
-export type CliMessageKey = CommonMessageKey | EvolveMessageKey | GenMessageKey | HelpMessageKey | InitMessageKey | RunMessageKey;
+export type CliMessageKey = CommonMessageKey | EvolveMessageKey | GenMessageKey | HelpMessageKey | InitMessageKey | InstallMessageKey | RunMessageKey;
 export declare const CLI_DICT: Record<CliMessageKey, CliMessage>;
-//# sourceMappingURL=i18n-dict.d.ts.map

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

@@ -55,6 +55,7 @@ import { evolveDict } from './i18n-dict/evolve.js';
 import { genDict } from './i18n-dict/gen.js';
 import { helpDict } from './i18n-dict/help.js';
 import { initDict } from './i18n-dict/init.js';
+import { installDict } from './i18n-dict/install.js';
 import { runDict } from './i18n-dict/run.js';
 export const CLI_DICT = {
     ...commonDict,
@@ -62,6 +63,6 @@ export const CLI_DICT = {
     ...genDict,
     ...helpDict,
     ...initDict,
+    ...installDict,
     ...runDict,
 };
-//# sourceMappingURL=i18n-dict.js.map

package/dist/cli/lib/i18n.d.ts

@@ -26,4 +26,3 @@ export declare function tBoth(key: CliMessageKey, params?: Record<string, string
     zh: string;
     en: string;
 };
-//# sourceMappingURL=i18n.d.ts.map

package/dist/cli/lib/i18n.js

@@ -55,4 +55,3 @@ export function tBoth(key, params) {
         en: tCli(key, 'en', params),
     };
 }
-//# sourceMappingURL=i18n.js.map

package/dist/cli/lib/parse-run-config/judge-models.d.ts

@@ -21,4 +21,3 @@
 import type { JudgeConfig } from '../../../types/index.js';
 export declare function parseJudgeModelsArg(raw: string): JudgeConfig[];
 export declare function parseJudgeModelsArgOrExit(raw: string): JudgeConfig[];
-//# sourceMappingURL=judge-models.d.ts.map

package/dist/cli/lib/parse-run-config/judge-models.js

@@ -52,4 +52,3 @@ export function parseJudgeModelsArgOrExit(raw) {
         throw new CliExit(2);
     }
 }
-//# sourceMappingURL=judge-models.js.map

package/dist/cli/lib/parse-run-config/samples-discovery.d.ts

@@ -16,4 +16,3 @@
  * 没有唯一答案)。本函数仅处理单 treatment 与零 treatment 情况。
  */
 export declare function discoverSamplesPath(values: Record<string, unknown>, skillDir: string): string;
-//# sourceMappingURL=samples-discovery.d.ts.map

package/dist/cli/lib/parse-run-config/samples-discovery.js

@@ -17,7 +17,6 @@
  */
 import { resolve, join, dirname } from 'node:path';
 import { existsSync, statSync } from 'node:fs';
-import { parseVariantCwd } from '../../../inputs/skill-loader.js';
 export function discoverSamplesPath(values, skillDir) {
     const treatmentRaw = values.treatment;
     const treatments = treatmentRaw
@@ -36,8 +35,7 @@ export function discoverSamplesPath(values, skillDir) {
                     return join(treatmentDir, name);
             }
         }
-        const tname = parseVariantCwd(expr).name;
-        const omkDir = join(skillDir, tname, '.omk');
+        const omkDir = join(skillDir, expr, '.omk');
         if (existsSync(omkDir))
             return omkDir;
     }
@@ -50,4 +48,3 @@ export function discoverSamplesPath(values, skillDir) {
     }
     return cwdFile;
 }
-//# sourceMappingURL=samples-discovery.js.map

package/dist/cli/lib/parse-run-config/variant-resolution.d.ts

@@ -9,10 +9,14 @@
  *   - 仅 batch 模式 → 留空,batch workflow 自己生成(baseline vs 每个 skill)
  *   - 都没有 → throw,附带 skill-dir 下候选 variant 提示,引导显式声明
  *
- * 之后做 variant name 唯一性检查 —— 同名 variant 不能同时在 --control 和
- * --treatment,也不能 --treatment 内重复。否则 ensemble agreement / report 字段
- * 都会按 name 去重,跑了但聚不到。
+ * 之后做 variant 唯一性检查 —— 同一个物理 variant 不能同时在 --control 和
+ * --treatment,也不能 --treatment 内重复。按 `variantIdentity`(规范化后的物理身份,
+ * 路径 resolve、目录与 SKILL.md 折叠、cwd 纳入)判重,而非派生短名,也不是裸 expr 字面量:
+ *   - `v1/greeter.md` 与 `v2/greeter.md` 短名都是 greeter 却是两个 variant —— 不该误判重复,
+ *     最终唯一名由 resolveArtifacts 的 ensureUniqueVariantNames 消歧。
+ *   - `./x.md` 与 `x.md`、`dir` 与 `dir/SKILL.md` 是同一份 skill —— 必须判为重复,
+ *     否则会被解析成两个 artifact 后用 `x` / `x#2` 消歧,变成同一份 skill 自比,
+ *     悄悄废掉「control 不能等于 treatment」的测量保护。
  */
 import type { EvalConfig, VariantSpec } from '../../../types/index.js';
 export declare function resolveVariantSpecs(values: Record<string, unknown>, evalConfig: EvalConfig | null, skillDir: string): VariantSpec[];
-//# sourceMappingURL=variant-resolution.d.ts.map

package/dist/cli/lib/parse-run-config/variant-resolution.js

@@ -9,27 +9,59 @@
  *   - 仅 batch 模式 → 留空,batch workflow 自己生成(baseline vs 每个 skill)
  *   - 都没有 → throw,附带 skill-dir 下候选 variant 提示,引导显式声明
  *
- * 之后做 variant name 唯一性检查 —— 同名 variant 不能同时在 --control 和
- * --treatment,也不能 --treatment 内重复。否则 ensemble agreement / report 字段
- * 都会按 name 去重,跑了但聚不到。
+ * 之后做 variant 唯一性检查 —— 同一个物理 variant 不能同时在 --control 和
+ * --treatment,也不能 --treatment 内重复。按 `variantIdentity`(规范化后的物理身份,
+ * 路径 resolve、目录与 SKILL.md 折叠、cwd 纳入)判重,而非派生短名,也不是裸 expr 字面量:
+ *   - `v1/greeter.md` 与 `v2/greeter.md` 短名都是 greeter 却是两个 variant —— 不该误判重复,
+ *     最终唯一名由 resolveArtifacts 的 ensureUniqueVariantNames 消歧。
+ *   - `./x.md` 与 `x.md`、`dir` 与 `dir/SKILL.md` 是同一份 skill —— 必须判为重复,
+ *     否则会被解析成两个 artifact 后用 `x` / `x#2` 消歧,变成同一份 skill 自比,
+ *     悄悄废掉「control 不能等于 treatment」的测量保护。
  */
-import { discoverVariants, parseVariantCwd } from '../../../inputs/skill-loader.js';
+import { discoverVariants, variantExprToSkillName, variantIdentity, parseVariantCwd } from '../../../inputs/skill-loader.js';
 import { configVariantsToSpecs } from '../../../inputs/eval-config.js';
+/** CLI 的 --control / --treatment 只收 artifact 身份。`name@cwd` 语法已移除:撞到就抛迁移错误,
+ *  引导用户改用 --control-cwd / --treatment-cwd 或 eval.yaml 的 variant.cwd。git 修订语法 `@{...}`
+ *  由 parseVariantCwd 保护、不误判。cwd 由调用方在边界注入(见 eval-runner)。 */
+function cliVariantSpec(rawExpr, role, cwd) {
+    // 探测旧 name@cwd 形态报错。注:含合法 `@`(非 `@{`)的路径(如 /x/@dir/skill.md)会被一并
+    // 误判 —— 属 pre-existing 限制(这类路径本就不被 @cwd 支持),报错文案仍指向迁移指引。
+    if (parseVariantCwd(rawExpr).cwd !== undefined) {
+        throw new Error(`「name@cwd」语法已移除: "${rawExpr}"。请改用 --${role}-cwd <dir> 声明 runtime context，`
+            + `或在 eval.yaml 的 variant 上用结构化 cwd: 字段。`);
+    }
+    return { name: variantExprToSkillName(rawExpr), role, expr: rawExpr, ...(cwd ? { cwd } : {}) };
+}
 export function resolveVariantSpecs(values, evalConfig, skillDir) {
     const controlExpr = values.control;
     const treatmentExprs = values.treatment
         ? values.treatment.split(',').map((v) => v.trim()).filter(Boolean)
         : [];
+    const controlCwd = values['control-cwd'];
+    // 逗号列表,与 treatment 按序对齐;空位(空串)= 该 treatment 无 cwd。不 filter,保留位次。
+    const treatmentCwds = values['treatment-cwd']
+        ? values['treatment-cwd'].split(',').map((v) => v.trim())
+        : [];
     let variantSpecs;
     if (controlExpr || treatmentExprs.length > 0) {
         // CLI roles present → CLI entirely replaces config.variants (no merging).
+        if (controlCwd !== undefined && !controlExpr) {
+            throw new Error('--control-cwd 需要配 --control 一起用。');
+        }
+        if (treatmentCwds.length > 0 && treatmentCwds.length !== treatmentExprs.length) {
+            throw new Error(`--treatment-cwd 的数量(${treatmentCwds.length})必须与 --treatment(${treatmentExprs.length})按序对齐;`
+                + `某个 treatment 不需要 cwd 就在该位留空(如 /a,,/c)。`);
+        }
         variantSpecs = [];
         if (controlExpr) {
-            variantSpecs.push({ name: parseVariantCwd(controlExpr).name, role: 'control', expr: controlExpr });
-        }
-        for (const expr of treatmentExprs) {
-            variantSpecs.push({ name: parseVariantCwd(expr).name, role: 'treatment', expr });
+            variantSpecs.push(cliVariantSpec(controlExpr, 'control', controlCwd));
         }
+        treatmentExprs.forEach((expr, i) => {
+            variantSpecs.push(cliVariantSpec(expr, 'treatment', treatmentCwds[i]));
+        });
+    }
+    else if (controlCwd !== undefined || treatmentCwds.length > 0) {
+        throw new Error('--control-cwd / --treatment-cwd 需要配 --control / --treatment 一起用(eval.yaml 用 variant 的 cwd: 字段)。');
     }
     else if (evalConfig) {
         variantSpecs = configVariantsToSpecs(evalConfig.variants);
@@ -45,15 +77,15 @@ export function resolveVariantSpecs(values, evalConfig, skillDir) {
         throw new Error(`请通过 --control / --treatment 或 --config eval.yaml 声明 variant 角色。\n`
             + `  示例：omk eval --control baseline --treatment my-skill${hint}\n`
             + `  --batch 模式下自动用 baseline vs 每个 skill,无需显式声明\n`
-            + `  术语见 docs/specs/terminology-spec.md（v0.16 起废除 --variants，改用 experiment role 显式声明）`);
+            + `  术语见 docs/zh/specs/terminology-spec.md（v0.16 起废除 --variants，改用 experiment role 显式声明）`);
     }
-    const seenNames = new Set();
+    const seenIdentities = new Set();
     for (const spec of variantSpecs) {
-        if (seenNames.has(spec.name)) {
-            throw new Error(`variant "${spec.name}" 重复出现——同一 variant 不能同时属于 --control 与 --treatment，也不能在 --treatment 中重复。`);
+        const key = variantIdentity(spec.expr, skillDir, spec.cwd);
+        if (seenIdentities.has(key)) {
+            throw new Error(`variant "${spec.expr}" 重复出现——同一 variant 不能同时属于 --control 与 --treatment，也不能在 --treatment 中重复。`);
         }
-        seenNames.add(spec.name);
+        seenIdentities.add(key);
     }
     return variantSpecs;
 }
-//# sourceMappingURL=variant-resolution.js.map

package/dist/cli/lib/parse-run-config.d.ts

@@ -90,4 +90,3 @@ export declare const DEFAULT_REPORTS_DIR: string;
  * 把 oclif flags 当 values 喂进来。
  */
 export declare function parseRunConfig(values: Record<string, unknown>): ParseRunConfigResult;
-//# sourceMappingURL=parse-run-config.d.ts.map

package/dist/cli/lib/parse-run-config.js

@@ -35,7 +35,7 @@ export function parseRunConfig(values) {
     if (values.variants !== undefined) {
         throw new Error(`--variants 已在 v0.16 废除，请改用 --control <expr> 与 --treatment <v1,v2,...>\n`
             + `  迁移示例：--variants baseline,my-skill  →  --control baseline --treatment my-skill\n`
-            + `  复杂场景可用 --config eval.yaml（参见 docs/specs/terminology-spec.md）`);
+            + `  复杂场景可用 --config eval.yaml（参见 docs/zh/specs/terminology-spec.md）`);
     }
     // 1) Load --config (if provided). All subsequent fields fall back to it when CLI is silent.
     const evalConfig = values.config
@@ -154,4 +154,3 @@ export function parseRunConfig(values) {
         evalConfig,
     };
 }
-//# sourceMappingURL=parse-run-config.js.map

package/dist/cli/lib/progress.d.ts

@@ -22,4 +22,3 @@ export interface ProgressInfo {
     error?: string;
 }
 export declare function makeOnProgress(lang: CliLang): (info: ProgressInfo) => void;
-//# sourceMappingURL=progress.d.ts.map

package/dist/cli/lib/progress.js

@@ -59,4 +59,3 @@ export function makeOnProgress(lang) {
         }
     };
 }
-//# sourceMappingURL=progress.js.map

package/dist/cli/lib/resolve-skill-input.d.ts

@@ -5,4 +5,3 @@ export interface ResolvedSkillInput {
     samplesPath: string;
 }
 export declare function resolveSkillInput(input: string, lang: CliLang): ResolvedSkillInput;
-//# sourceMappingURL=resolve-skill-input.d.ts.map

package/dist/cli/lib/resolve-skill-input.js

@@ -2,9 +2,9 @@ import { resolve, join, dirname } from 'node:path';
 import { existsSync, statSync } from 'node:fs';
 import { tCli } from './i18n.js';
 // 统一 skill 入参解析:既接受 SKILL.md 文件(老式 + flat skill),也接受 directory-skill
-// 目录(skills/foo/ 自动找 foo/SKILL.md)。samples 发现优先级:.omk/samples.json >
-// eval-samples.json > .yaml > .yml,fallback 是第一个候选(给上游 "samples 不存在"
-// 的提示一个稳定路径)。
+// 目录(skills/foo/ 自动找 foo/SKILL.md)。samples 发现优先级:.omk/ 目录(loadSamples
+// 支持目录模式，自动 glob 多文件) > eval-samples.json > .yaml > .yml,fallback 是
+// .omk/ 目录路径(给上游 "samples 不存在" 的提示一个稳定路径)。
 //
 // 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。
 export function resolveSkillInput(input, lang) {
@@ -26,13 +26,14 @@ export function resolveSkillInput(input, lang) {
         skillPath = resolved;
         skillDir = dirname(resolved);
     }
+    const omkDir = join(skillDir, '.omk');
     const candidates = [
-        join(skillDir, '.omk', 'samples.json'),
+        ...(existsSync(omkDir) ? [omkDir] : []),
         join(skillDir, 'eval-samples.json'),
         join(skillDir, 'eval-samples.yaml'),
         join(skillDir, 'eval-samples.yml'),
     ];
-    const samplesPath = candidates.find(existsSync) ?? candidates[0];
+    // fallback 到 .omk/ 目录：上游会报 "no sample files found in directory" 引导用户创建
+    const samplesPath = candidates.find(existsSync) ?? omkDir;
     return { skillPath, skillDir, samplesPath };
 }
-//# sourceMappingURL=resolve-skill-input.js.map

package/dist/cli/lib/run-tally.d.ts

@@ -15,4 +15,3 @@ export declare function computeRunTally(report: ReportDocument): {
     passed: number;
     failed: number;
 };
-//# sourceMappingURL=run-tally.d.ts.map

package/dist/cli/lib/run-tally.js

@@ -19,7 +19,7 @@ export function computeRunTally(report) {
             failed += s.errorCount;
         }
     };
-    if (report.kind === 'batch-evaluation') {
+    if (report.reportKind === 'batch-evaluation') {
         for (const item of report.items)
             accumulate(item.summary);
     }
@@ -28,4 +28,3 @@ export function computeRunTally(report) {
     }
     return { passed, failed };
 }
-//# sourceMappingURL=run-tally.js.map

package/dist/cli/lib/shared.d.ts

@@ -9,4 +9,3 @@ export interface ReportServer {
 }
 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