niceeval 0.9.1 → 0.10.2

package/src/i18n/zh-CN.ts

@@ -58,9 +58,16 @@ export const zhCN = {
   "cli.flag.invalidOutput": "标志 --output 需要 auto|human|agent|ci 之一,收到 \"{{value}}\"。\n",
   "runner.budgetUnenforceable":
     "{{budgetKey}} 的 budget:连续多个 attempt 完成后都拿不到成本数据(agent 不上报用量且模型不在价格表)——该 agent 的 budget 无法执行,取消护栏继续跑。\n",
+  "runner.experimentTeardownFailed":
+    "实验 {{experimentId}} 的 teardown 执行失败:{{message}}。结果不受影响,但该实验起的宿主机资源可能没有回收,请手动检查。\n",
+  "runner.cleanupTimeout": "cleanup 执行超过 {{timeoutMs}}ms 清理超时\n",
+  "runner.setupReturnedCleanup":
+    "{{layer}} 返回了一个函数。setup 不承载收尾,返回值不会被执行——收尾写在同层成对的 teardown 里({{hint}});见 docs-site 的实验教程或 docs/runner.md「环境预置」。\n",
+  "runner.experimentTeardownLate":
+    "实验 {{experimentId}} 的 teardown 未被正常计数路径触发,已在运行收尾兜底执行。结果不受影响;这行出现说明命中了一个未定位的调度间歇问题,请把本次运行信息记入 memory 台账。\n",
   "judge.modelMissing":
     "judge 未配置模型:在 defineConfig({ judge: { model: \"...\" } })、eval 的 judge 配置或环境变量 NICEEVAL_JUDGE_MODEL 里指定裁判模型(没有内置默认模型)。\n" +
-    "  文档:node_modules/niceeval/docs-site/zh/how-to/scoring-guide.mdx",
+    "  文档:node_modules/niceeval/docs-site/zh/tutorials/scoring-guide.mdx",
   "loaders.yamlMissing":
     "loadYaml(\"{{path}}\") 需要 YAML 解析器:请先 `pnpm add yaml`(或改用 loadJson + JSON 数据集)。",
   "cli.flag.parseError": "{{message}}\n运行 `niceeval --help` 查看用法。\n",
@@ -68,7 +75,7 @@ export const zhCN = {
   "cli.help":
     "niceeval — agent-native evals\n\n" +
     "用法:\n" +
-    "  niceeval exp [组|实验] [eval-id 前缀…]   跑实验\n" +
+    "  niceeval exp [路径|实验] [eval-id 前缀…]   跑实验\n" +
     "  niceeval show [eval-id 前缀… | @<locator>]   终端读结果\n" +
     "      裸跑:每个 experiment 的现刻判定(跨 run 合成),每行带紧凑 attempt 索引\n" +
     "        (locator + 失败原因)\n" +
@@ -116,10 +123,14 @@ export const zhCN = {
   "cli.eval.noMatchKnown": "已发现 {{count}} 个 eval:{{evals}}\n",
   "cli.exp.agentModelFlagUnsupported": "experiment 运行不支持 --agent / --model。请新增或复制一个 experiment 文件并修改 model。\n",
   "cli.exp.viewerFlagUnsupported": "`{{flag}}` 只适用于 niceeval {{command}},不能用于 niceeval exp。\n",
-  "cli.experiment.noMatch": "没有匹配的实验:{{arg}}。已发现:{{experiments}}\n",
+  "cli.experiment.noMatch":
+    "没有匹配的实验:{{arg}}。可用路径:{{experiments}}。\n" +
+    "运行 `niceeval exp <路径> --dry` 预览计划。\n",
   "cli.experiment.viewerCommandHint": "你可能想运行:niceeval {{command}}{{args}}\n",
-  "cli.experiment.noEvalsSelected": "未选择任何 eval:{{selection}} 匹配到 0 个 eval。可用实验:{{experiments}}。\n",
-  "cli.experimentGroup": "组",
+  "cli.experiment.noEvalsSelected":
+    "未选择任何 eval:{{selection}} 匹配到 0 个 eval。可用的 eval 前缀:{{experiments}}。\n" +
+    "运行 `niceeval exp {{selection}} --dry` 查看它覆盖了什么,或去掉 eval 过滤跑这些实验选中的全部 eval。\n",
+  "cli.experimentGroup": "路径",
   "cli.fallbackCleanupTimeout": "\ngraceful 清理超时,强制清理沙箱…\n",
   "cli.forceCleanupExit": "\n强制清理沙箱并退出…\n",
   "cli.init.done": "已就绪:evals/、niceeval.config.ts,以及 AGENTS.md 里的 niceeval agent 指引区块(指向 node_modules/niceeval/docs-site/zh)。\n",
@@ -130,8 +141,8 @@ export const zhCN = {
   "cli.pressCtrlC": "按 Ctrl+C 退出。\n",
   "cli.resultsPath": "结构化结果:{{path}}(snapshot.json + 每 attempt 的 result.json / events.json / trace.json / diff.json)\n",
   "cli.run.experimentRequired":
-    "运行 eval 必须通过 experiment:用 `niceeval exp [实验组|配置] [eval id 前缀]`。\n" +
-    "  文档:node_modules/niceeval/docs-site/zh/how-to/write-experiment.mdx\n",
+    "运行 eval 必须通过 experiment:用 `niceeval exp [路径|配置] [eval id 前缀]`。\n" +
+    "  文档:node_modules/niceeval/docs-site/zh/tutorials/write-experiment.mdx\n",
   "cli.run.experimentRequiredHint": "提示:\"{{pattern}}\" 是实验{{kind}},你大概想跑:niceeval exp {{pattern}}\n",
   "cli.run.experimentRequiredKnown": "已发现实验:{{experiments}}\n",
   "cli.unimplemented": "命令 \"{{command}}\" 暂未实现(MVP)。\n",
@@ -142,7 +153,7 @@ export const zhCN = {
   "cli.view.url": "niceeval view: {{url}}\n",
   "context.capabilityMissing":
     "agent \"{{agent}}\" 不是沙箱型(defineSandboxAgent 构造),t.{{method}} 这类断言只有沙箱型 agent 可用。换用 defineSandboxAgent 构造的 agent,或去掉这条断言。\n" +
-    "  文档:node_modules/niceeval/docs-site/zh/how-to/sandbox-agent.mdx",
+    "  文档:node_modules/niceeval/docs-site/zh/tutorials/sandbox-agent.mdx",
   "context.skipEmpty": "skip() 需要一个非空理由。",
   "context.turnFailed": "本轮 send 返回 failed(turn status = failed):{{message}}",
   "context.turnFailedDefault": "本轮 send 返回 failed(turn status = failed)",
@@ -152,6 +163,8 @@ export const zhCN = {
   "define.evalTestRequired": "defineEval 需要一个 async test(t) 函数。",
   "define.experimentAgentRequired": "defineExperiment 需要 agent。",
   "define.experimentFlagNotJson": "experiment.flags.{{key}} 不是可 JSON 序列化的值(函数 / undefined / 循环引用 / bigint 不允许);flags 会原样进入结果快照,必须是纯 JSON。",
+  "define.experimentLabelInvalid": "experiment.labels.{{key}} 必须是字符串或有限数字;labels 是报告侧的归类坐标,会原样进入结果快照。",
+  "define.experimentSetupNotFunction": "experiment.setup 必须是函数((ctx) => void);要清理请挂 experiment.teardown;要按实验准备沙箱内环境请挂 sandbox spec 的 .setup() 钩子链。",
   "define.experimentIdRejected": "defineExperiment 不接受 id —— id 由文件路径推导。",
   "define.sandboxAgentNameRequired": "defineSandboxAgent 需要 name。",
   "define.sandboxCreateRequired": "defineSandbox 需要一个 create() 函数。",
@@ -164,7 +177,7 @@ export const zhCN = {
   "docker.unsupportedRuntime": "Unsupported runtime: {{runtime}}",
   "feedback.human.active": "ACTIVE",
   "feedback.human.budgetExhausted": "{{experimentId}} 预算已耗尽(已花 {{spent}},未跑 {{unstarted}})",
-  "feedback.human.compare": "Compare: niceeval view {{group}}",
+  "feedback.human.compare": "Compare: niceeval view",
   "feedback.human.counts": "共 {{total}} · 复用 {{reused}} · 运行中 {{running}} · 排队 {{queued}} · 已完成 {{completed}}",
   "feedback.human.diffHint": "Diff:    niceeval show {{locator}} --diff",
   "feedback.human.evalHint": "Eval:    niceeval show {{locator}} --source",
@@ -189,6 +202,10 @@ export const zhCN = {
   "feedback.phase.evalSetup": "eval 预置",
   "feedback.phase.sandboxCreate": "创建沙箱",
   "feedback.phase.sandboxQueue": "排队等沙箱",
+  "feedback.phase.experimentSetup": "实验预置",
+  "feedback.phase.experimentTeardown": "实验收尾",
+  "feedback.human.hookDone": "完成",
+  "feedback.human.hookFailed": "失败",
   "feedback.phase.sandboxSetup": "沙箱预置",
   "feedback.phase.scoring": "评分",
   "feedback.phase.teardown": "清理中",
@@ -268,13 +285,11 @@ export const zhCN = {
   "runner.startSandboxTeardown": "sandbox teardown(环境预置钩子)…",
   "runner.timeout": "attempt 超时({{timeoutMs}}ms)\n最近进度:\n{{recentLogs}}",
   "runner.traceSelected": " → 留 {{count}}(按语义)",
-  "runner.resumeCarry": "  · 复用上次 {{carried}} 个已判定的结果,重跑 {{retry}} 个 eval\n",
-  "runner.resumeCarryDetail": "      复用 [{{experiment}}] {{evals}}\n",
   "runner.useRemoteAgent": "使用 remote agent(不创建沙箱)…",
   "sandbox.providerNotImplemented": "{{provider}} sandbox provider not implemented; use docker, vercel, or e2b",
   "sandbox.missingSpec":
     "沙箱型 agent 需要一个 sandbox,但没有提供。niceeval 不再自动选默认 provider——请在 defineExperiment()/defineConfig() 里把 sandbox 设成 dockerSandbox() / vercelSandbox() / e2bSandbox()(从 \"niceeval/sandbox\" 导入)。\n" +
-    "  文档:node_modules/niceeval/docs-site/zh/how-to/sandbox-providers.mdx",
+    "  文档:node_modules/niceeval/docs-site/zh/tutorials/sandbox-providers.mdx",
   "sandbox.dependencyMissing.docker": "Docker sandbox requires 'dockerode'. Install it with: pnpm add dockerode @types/dockerode",
   "sandbox.dependencyMissing.e2b": "E2B sandbox requires 'e2b'. Install it with: pnpm add e2b",
   "sandbox.dependencyMissing.vercel": "Vercel sandbox requires '@vercel/sandbox'. Install it with: pnpm add @vercel/sandbox",

package/src/index.ts

@@ -24,7 +24,9 @@ export type {
   Severity,
   Verdict,
   EvalDef,
+  EvalDescriptor,
   ExperimentDef,
+  ExperimentHookContext,
   Config,
   LocalizedText,
   JudgeConfig,