niceeval 0.12.1 → 0.12.2-canary.31

package/dist/i18n/zh-CN.js

@@ -1,458 +0,0 @@
-export const zhCN = {
-    "agent.installFailed": "安装失败:{{key}}\n{{tail}}",
-    "agent.unknown": "未知 agent \"{{name}}\"。已注册:{{known}}。",
-    "agent.diagnose.exitCode": "agent 运行退出码 {{code}}",
-    "agent.diagnose.lastError": "最后错误:{{message}}",
-    "agent.diagnose.noTranscript": "transcript 未生成",
-    "agent.diagnose.outputTail": "输出末尾:\n{{tail}}",
-    "agent.diagnose.zeroEvents": "transcript 存在但 0 事件",
-    "agent.ensure.identityMissingAgent": "AgentProvisioner.identity.agent 不能为空。",
-    "agent.ensure.identityMissingRevision": "AgentProvisioner.identity.revision 不能为空(agent={{agent}})。",
-    "agent.ensure.unstableVersion": "AgentProvisioner.identity.version 必须是精确版本,不能是 \"{{version}}\"(agent={{agent}})。无钉死版本的安装不参与可携带结果。",
-    "agent.ensure.stagedNeedsPrepare": "staged 模式的 AgentProvisioner({{agent}})必须提供 prepare(),在题面网络外准备锁定制品。",
-    "agent.ensure.verifyOnlyHasPrepare": "verifyOnly 模式不安装,AgentProvisioner({{agent}})不应同时提供 prepare()。",
-    "agent.ensure.prepareWrongMode": "只有 staged 模式才调用 prepare(agent={{agent}}, mode={{mode}})。",
-    "agent.ensure.artifactMissingDigest": "staged artifact 缺少 digest(agent={{agent}})。",
-    "agent.ensure.stagedMissingArtifact": "staged 安装缺少已准备的 artifact(agent={{agent}})。Runner 应先 Run 级 prepare,或 ensure 经 coordinator 懒准备。",
-    "agent.ensure.failed": "Agent Ensure 失败(agent={{agent}}, phase={{phase}}):期望版本 {{expected}},实际 {{actual}}。{{detail}}\n下一步:{{next}}",
-    "agent.ensure.nextVerifyOnly": "预装匹配的 Agent,或改用 staged / sandbox-network provisioner。",
-    "agent.ensure.nextInstallerMissing": "换用预装了该精确 Agent identity 的环境,或在 Experiment layer 中用 installTool() 显式安装。",
-    "agent.ensure.nextRecheck": "核对安装目录、PATH 与精确版本后重跑;不要把坏环境交给做题。",
-    "agent.ensure.missingBin": "找不到命令 {{bin}}。{{tail}}",
-    "agent.ensure.versionUnparseable": "无法从 `{{bin}} --version` 解析版本:{{stdout}}",
-    "agent.ensure.versionMismatch": "{{bin}} 版本不匹配:期望 {{expected}},实际 {{actual}}。",
-    "agent.ensure.digestMismatch": "staged artifact digest 不匹配(agent={{agent}}):期望 {{expected}},实际 {{actual}}。",
-    "agent.ensure.npmPackFailed": "宿主 npm pack {{packageName}}@{{version}} 失败:\n{{tail}}",
-    "agent.ensure.npmPackEmpty": "宿主 npm pack {{packageName}}@{{version}} 未产出 .tgz(目录 {{dest}})。",
-    "agent.ensure.npmInstallFailed": "沙箱内从 staged tarball 安装 {{agent}} 失败:\n{{tail}}",
-    "agent.ensure.platformDetectFailed": "无法从主 Sandbox 探测目标平台(uname / ldd):{{tail}}",
-    "agent.ensure.selfContainedInstallFailed": "沙箱内解开 {{agent}} 原生包失败(需要 tar + gzip):\n{{tail}}",
-    "agent.ensure.npmMissingInSandbox": "沙箱里没有 npm,而 {{agent}} 这个平台只有依赖 Node 的 npm 包。任务镜像不带 Node 工具链时,请给该平台提供自带运行时的原生包,或换一个自定义 provisioner。",
-    "agent.ensure.homeDetectFailed": "无法探测沙箱 $HOME,staged 安装需要展开用户前缀路径。",
-    "bub.homeDetectFailed": "无法探测沙箱 $HOME(printf $HOME 输出为空)。不兜底到 provider 专属固定路径,请检查沙箱 provider。",
-    "bub.checkpointCaptureFailed": "bub checkpoint 缓存回填失败(本沙箱不受影响,后续沙箱会重新安装):{{error}}",
-    "bub.checkpointRestoreFailed": "bub checkpoint 还原失败,回退到全量安装:{{error}}",
-    "bub.installFailed": "bub 安装失败(重试 {{attempts}} 次):\n{{tail}}",
-    "bub.setupNotRun": "bub adapter 的 setup() 尚未在该沙箱运行(缺 home/workspace 信息);运行器应先调 setup 再 send。",
-    "checkpoint.emptyTar": "checkpoint: tar 为空(paths: {{paths}})",
-    "checkpoint.archiveFailed": "checkpoint 归档失败(exit {{exitCode}}): {{detail}}",
-    "checkpoint.restoreFailed": "checkpoint 恢复失败(exit {{exitCode}}): {{detail}}",
-    "skill.localMissing": "本地 skill 路径不存在:\"{{path}}\"(解析到 {{resolved}})。path 相对跑 niceeval 的项目根解析。",
-    "skill.localDirNoSkillFile": "本地 skill 目录 \"{{path}}\" 里没有 SKILL.md。目录形态的 skill 必须在根下带一个 SKILL.md。",
-    "skill.localUnsupportedShape": "本地 skill 路径 \"{{path}}\" 形态不支持。只接受:含 SKILL.md 的目录,或单个 .md 文件。",
-    "skill.repoCloneFailed": "repo skill 拉取失败:{{source}}(ref: {{ref}})\n{{tail}}",
-    "skill.repoNoSkills": "repo skill {{source}} 里没找到任何 SKILL.md。",
-    "skill.repoAmbiguous": "repo skill {{source}} 里有多个 skill,必须用 `skills: [...]` 明确选择要启用哪些。可选:{{available}}。",
-    "skill.repoUnknownSkill": "repo skill {{source}}(ref: {{ref}})里没有名为 \"{{skill}}\" 的 skill。可选:{{available}}。",
-    "skill.copyFailed": "skill \"{{name}}\" 装进 {{dest}} 失败:\n{{tail}}",
-    "mcp.ambiguousTransport": "MCP server \"{{name}}\" 同时给出了 command 和 url——二选一:本地 stdio 进程写 command,远程 Streamable HTTP 端点写 url。",
-    "plugin.marketplaceFailed": "{{agent}} marketplace \"{{name}}\" 连接失败(source: {{source}}, ref: {{ref}}):\n{{tail}}",
-    "plugin.marketplaceVerifyFailed": "{{agent}} marketplace \"{{name}}\" add 后回读注册列表失败({{command}}):\n{{tail}}",
-    "plugin.marketplaceNameMismatch": "{{agent}} marketplace 名不匹配:配置的 name \"{{expected}}\"(source: {{source}})不在 add 后回读的注册列表里,本次实际注册为:{{actual}}。" +
-        "marketplace.name 必须等于目标仓库 manifest 声明的 name,改成真实名字再跑。",
-    "plugin.installFailed": "{{agent}} plugin \"{{name}}\"(marketplace: {{marketplace}})安装失败:\n{{tail}}",
-    "plugin.listFailed": "{{agent}} 读取已安装 plugin 列表失败({{command}}):\n{{tail}}",
-    "plugin.removeFailed": "{{agent}} 移除同名 plugin \"{{name}}\" 失败:\n{{tail}}\n" +
-        "安装前要把沙箱状态收敛到声明,同名的残留安装必须先移除再从声明的 marketplace 重装。",
-    "nativeConfig.pathNotProjectRelative": "{{agent}} {{field}} 只接受项目根内的相对路径,收到 \"{{path}}\"。绝对路径、包含 `..` 的路径和 `~` 路径都不行;项目根外的配置先复制进项目再引用。",
-    "nativeConfig.missing": "{{agent}} {{field}} 指向的文件不存在:\"{{path}}\"(解析到 {{resolved}})。路径相对运行 niceeval 的项目根(含 niceeval.config.ts 的目录)解析,不相对 eval / experiment 源码文件。",
-    "nativeConfig.escapesRoot": "{{agent}} {{field}} \"{{path}}\" 经符号链接解析到项目根之外({{resolved}})。配置文件必须真实位于项目根内。",
-    "nativeConfig.notFile": "{{agent}} {{field}} \"{{path}}\" 不是普通文件。指向一份完整的官方配置文件。",
-    "nativeConfig.invalidSyntax": "{{agent}} {{field}} \"{{path}}\" 不是合法的 {{format}}:{{detail}}",
-    "nativeConfig.reservedKeys": "{{agent}} {{field}} \"{{path}}\" 含保留键:{{keys}}。这些键由 experiment 与 Adapter 拥有(model、鉴权、MCP、OTel 经独立配置层叠加),从文件里删掉再跑。",
-    "nativeConfig.uploadFailed": "原生配置文件 \"{{path}}\" 上传沙箱失败({{dest}}):\n{{tail}}",
-    "cli.all": "(全部)",
-    "cli.browserOpenFailed": "无法自动打开浏览器,请手动访问:{{url}}\n",
-    "cli.clean.done": "已删除 .niceeval/ 历史运行 artifact。\n",
-    "cli.config.missing": "找不到 niceeval.config.ts。\n" +
-        "修法:\n" +
-        "  - [init] 运行 `npx niceeval init` 生成 niceeval.config.ts 和 evals/\n" +
-        "  - [cd] 切到包含 niceeval.config.ts 的项目根再运行\n" +
-        "  文档:node_modules/niceeval/docs-site/zh/tutorials/quickstart.mdx",
-    "cli.config.noDefault": "niceeval.config.ts 需要 default export(defineConfig(...))。",
-    "cli.dry.header": "plan: {{attempts}} · {{evals}} × {{configs}} · attempts {{attemptCount}}",
-    "cli.dry.unit.attempt": "个 attempt",
-    "cli.dry.unit.attempts": "个 attempt",
-    "cli.dry.unit.eval": "个 eval",
-    "cli.dry.unit.evals": "个 eval",
-    "cli.dry.unit.config": "个运行配置",
-    "cli.dry.unit.configs": "个运行配置",
-    "cli.dry.affects": "影响 {{evals}} · {{ids}}",
-    "cli.dry.acceptHint": "accept:  {{command}}",
-    "cli.accept.choiceHeader": "stale  {{selector}}{{change}}  ({{evals}} 条 eval)\n",
-    "cli.accept.prompt": "  复用这些结果吗? [y/N] ",
-    "cli.accept.nothingToAccept": "本次计划里没有可授权的差异(没有条目卡在指纹门上)。\n" +
-        "按原计划运行。\n",
-    "cli.accept.equivalent": "等价命令:  {{command}}\n",
-    "cli.accept.noneChosen": "没有授权任何差异,按原计划运行。\n",
-    "cli.accept.usage": "error: niceeval accept 必须接收一个或多个 @<locator>\n" +
-        "  fix: 从 `niceeval show --history` 复制明确的 locator,再运行 `niceeval accept @<locator>...`\n",
-    "cli.accept.flagUnsupported": "error: {{flag}} 不能用于 niceeval accept\n" +
-        "  fix: 只传 @<locator>(可选 `--record <目录>`)\n",
-    "cli.accept.failed": "error: 接受结果失败:{{error}}\n",
-    "cli.accept.done": "已接受 {{sourceLocator}}。新结果 locator: {{locator}}。当前指纹:{{fingerprint}}\n",
-    "cli.error": "niceeval 出错:{{error}}\n",
-    "cli.flag.acceptNeedsSelector": "error: --accept 必须带 selector,例如 --accept config:judge.model\n" +
-        "  fix: 先跑 `niceeval exp <选择> --dry`,每条 `stale` 行都打出了它可授权的 selector,原样复制一条\n" +
-        "  本次计划里可以授权的差异:{{available}}\n",
-    "cli.flag.acceptWithRerunAll": "--accept 不能与 --rerun all 同用:一边全不采信缓存,一边又要采信这条差异,方向自相矛盾。\n" +
-        "要授权这条差异就去掉 --rerun all;要全量重跑就去掉 --accept。\n",
-    "cli.flag.acceptNoSuchDifference": "本次计划里没有这条差异:{{selectors}}。\n" +
-        "这里可以授权的差异:{{available}}。\n" +
-        "跑 `niceeval exp <选择> --dry` 看每条差异拦下了哪些条目。\n",
-    "cli.flag.invalidNumber": "标志 --{{flag}} 需要数字,收到 \"{{value}}\"。\n",
-    "cli.flag.outputRemoved": "error: unknown option '--output'\n  fix: 不加 flag 运行得到人读文本;机器面用 --json\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",
-    "runner.teardownRegistrationWriteFailed": "实验 {{experimentId}} 的强杀恢复收尾登记写入失败:{{message}}。本次运行照常继续,但这次运行期间若被 SIGKILL,`niceeval exp --teardown` 与启动自愈都无法找到它——检查 .niceeval/teardowns/ 下的磁盘空间或权限。\n",
-    "runner.lockTakenOver": "接管了 {{experimentId}}/{{evalId}} 的过期用例锁(原持有者 pid {{pid}}@{{host}},心跳已过期)——那次运行大概率没能正常释放它;本次运行现在接手派发这条用例。\n",
-    "runner.gateLeaseTakenOver": "接管了实验 {{experimentId}} 的过期并发名额租约(槽位 {{slot}},原持有者 pid {{pid}}@{{host}},心跳已过期)——那次运行大概率没能正常释放它;本次运行现在接手这个名额。\n",
-    "runner.gateLeaseWaiting": "在等别的运行让出实验 {{experimentId}} 的并发名额:生效的 {{effectiveN}} 个位子全被占着({{holders}})。并行运行共用同一实验的名额,生效值取在场声明里最小的那个——本次运行声明的是 {{declaredN}}。名额腾出来之前不会派发任何 attempt;对方的名额会在它的 attempt 跑完时释放,它若已死则 30s 后过期被接管。\n",
-    "runner.dispatchHaltedExperiment": "实验已止损(dispatch-halted):{{message}}\n",
-    "runner.dispatchHaltedEval": "eval 已止损:{{message}}\n",
-    "judge.modelMissing": "judge 未配置模型:在 Experiment、Eval 或 defineConfig 的 judge 配置里指定裁判模型(没有内置默认模型,也没有对应的环境变量)。\n" +
-        "  文档:node_modules/niceeval/docs-site/zh/tutorials/evaluation-kinds.mdx",
-    "loaders.yamlMissing": "loadYaml(\"{{path}}\") 需要 YAML 解析器:请先 `pnpm add yaml`(或改用 loadJson + JSON 数据集)。",
-    "loaders.criteriaNoMatch": "loadCriteria 的这些 pattern 一个文件都没匹配到(或命中的文件全被后写的 `!` 排除了):{{patterns}}。多半是写错了或判据文件搬走了——pattern 按项目根 {{root}} 展开,不是按 eval 文件所在目录。请对着磁盘上的真实路径核对这几条,确认不再需要就把它删掉;别的 pattern 有命中也不放过,放过等于判据悄悄变窄,而改窄了的判据会让本该重跑的 eval 照常携带旧判定。",
-    "loaders.criteriaOutsideRoot": "loadCriteria 匹配到的 \"{{path}}\" 落在项目根外(实际指向 {{resolved}}):判据文件的项目根相对路径就是它进指纹的键,穿出根之后这个键不再稳定。请把这棵树(或符号链接的目标)搬进项目根 {{root}} 内,或用 `!` 把这条链接从 pattern 里排除。",
-    "loaders.privateNoMatch": "loadPrivate 的这些 pattern 一个文件都没匹配到(或命中的文件全被后写的 `!` 排除了):{{patterns}}。多半是写错了或 private 文件搬走了——pattern 按项目根 {{root}} 展开,不是按 eval 文件所在目录。请对着磁盘上的真实路径核对;静默放过等于泄题门与指纹悄悄丢掉隐藏输入。",
-    "loaders.outsideDiscovery": "在发现阶段之外读了 \"{{path}}\":loader 只能在 eval 文件的模块顶层调用(发现阶段求值 eval 模块时),文件内容才来得及进这条 eval 的指纹。请把这次读取挪到模块顶层、把结果存进一个常量再在 test(t) 里用;test(t) 与各生命周期 hook 的运行期不能调 loader。",
-    "loaders.nonFileUrl": "loader 的 URL 入参只支持 file: 协议,收到的是 {{protocol}}({{url}})。请改传 new URL(相对路径, import.meta.url),或直接写项目根相对的字符串路径。",
-    "cli.flag.parseError": "{{message}}\n运行 `niceeval --help` 查看用法。\n",
-    "cli.command.unknown": "未知命令 \"{{command}}\"。\n运行 `niceeval --help` 查看用法。\n",
-    "cli.help": "niceeval — agent-native evals\n\n" +
-        "用法:\n" +
-        "  niceeval exp [路径|实验] [eval-id 前缀…]   跑实验\n" +
-        "  niceeval exp list [实验前缀]              列出可运行的实验配置(不派发)\n" +
-        "      --teardown   强杀后补收尾:只对选中的实验各执行一次 teardown(不派发\n" +
-        "        attempt、不跑 setup);与 eval id 前缀组合是用法错误\n" +
-        "  niceeval accept @<locator>...           接受明确列出的历史结果\n" +
-        "  niceeval show [eval-id 前缀… | @<locator>]   终端读结果\n" +
-        "      不带证据 flag:命中范围的默认报告(裸跑、eval id 前缀、单个 --exp 都落在这里);\n" +
-        "        两个以上 --exp 改为逐条件对照\n" +
-        "      @<locator>  精确一个 attempt:无 flag → 紧凑全景;带 flag → 对应证据切面\n" +
-        "      --source      该 attempt 运行时保存的 Eval 源码,断言标回源码行\n" +
-        "      --execution   该 attempt 的执行事件流(消息/thinking/Skill/工具调用),\n" +
-        "        有 OTel 时同一节点补时间\n" +
-        "      --execution --grep <pattern>   只输出命中卡片,末尾附跨 attempt 汇总;\n" +
-        "        --execution --expand <t<n>.c<n>|cmd<n>>   展开一张卡片完整内容\n" +
-        "        (两者互斥;--expand 要求范围恰好一个 attempt)\n" +
-        "      --timing      整个 attempt 的统一时间树(阶段 + hook/命令/turn + 轮内 OTel)\n" +
-        "      --diff[=文件] agent 归因的文件改动摘要;=文件 按窗口展开单个文件\n" +
-        "      证据 flag 接受任意范围:范围含多个 attempt 时逐 attempt 分节\n" +
-        "        (按 experimentId、evalId、attempt 序)\n" +
-        "      --history   逐 experiment × eval 的执行时间轴(与 --report 互斥)\n" +
-        "      --usage     范围内逐 attempt 的用量表,按 experiment 分节各自合计\n" +
-        "      --stats     eval × experiment 的历史全执行稳定性矩阵\n" +
-        "        (与 @<locator>、--report 互斥)\n" +
-        "      --json      任何切片的结构化形态:一个 JSON 文档到 stdout,选择与 text 面同一批\n" +
-        "        (与 --report、--expand 互斥)\n" +
-        "      --record <目录>  钉死记录根   --exp <id> 可重复,两个以上进入对照\n" +
-        "      --report <文件> 自定义报告   --page <id> 定初始页(多页报告渲染该页,\n" +
-        "        尾部再附其余页索引)\n" +
-        "      --fresh   只统计新执行的 attempt(排除携带条目与跨快照拼入的历史执行);\n" +
-        "        被排除的题转成占位行\n" +
-        "  niceeval list                            列出发现到的 eval\n" +
-        "  niceeval session list [--all] [实验前缀]  查询 Session(只读)\n" +
-        "  niceeval session show <sessionId>         查看一个 Session(只读)\n" +
-        "  niceeval view [eval-id 前缀…] [--out 目录] [--port n] [--no-open]\n" +
-        "      报告页 + 证据室;--report <文件> 整槽换成自定义报告(与 show 同一文件)\n" +
-        "      --page <id> 定初始页   --record <目录> 钉死记录根\n" +
-        "      --run <文件> 只打开这一份快照   --exp <id>(可重复)收窄到这些实验;\n" +
-        "      --fresh 只看新执行\n" +
-        "      --out <目录> 静态导出:index.html 连同查看器 artifact,可直接静态托管\n" +
-        "  niceeval sandbox list|enter|history|diff|stop  查看与销毁 --keep-sandbox 留下的现场\n" +
-        "  niceeval sandbox list --orphans / prune         核对并收回被强杀留下的无主实例\n" +
-        "  niceeval clean                           删除 .niceeval/ 历史 artifact\n" +
-        "  niceeval init                            脚手架 config + evals/\n\n" +
-        "标志:\n" +
-        "  --attempts n  --max-concurrency n  --max-build-concurrency n  --timeout ms\n" +
-        "  --budget usd  --tag t\n" +
-        "  --early-exit / --no-early-exit  --strict  --rerun[=failed|all]  --dry  --keep-sandbox[=failed|all]\n" +
-        "  --json  (机器面:stdout 上的 NDJSON 事件流;默认是人读文本)\n" +
-        "  --junit path  --out dir  --port n  --open / --no-open  -h, --help  -v, --version\n\n" +
-        "位置参数只选「跑哪些 eval」(id 前缀);对着哪个 agent、怎么跑来自 experiments/ 与\n" +
-        "标志。取值优先级:标志 > experiment > eval(只有 timeoutMs / judge 有这一层)>\n" +
-        "niceeval.config.ts > 内置默认;config 是缺省底不是覆盖层。--timeout 没有内置默认:\n" +
-        "四层都没写就是无上限,attempt 不设 deadline。配置项没有\n" +
-        "环境变量层,环境变量只放 API key 这类凭据。\n",
-    // show 的错误文案保持英文(错误文案英文的仓库约定);noResults 是提示,翻译。
-    "cli.show.noResults": "{{root}} 下没有结果。先 `niceeval exp` 跑一轮,再 `niceeval show`。\n",
-    "cli.show.runDirMissing": "Record directory not found: {{dir}}\n",
-    "cli.show.noEvalMatch": "No results matched: {{pattern}}. Evals with results: {{evals}}\n",
-    "cli.show.noExperimentMatch": "No experiment matched --exp {{arg}}. Experiments with results: {{experiments}}\n",
-    "cli.show.expAmbiguous": "error: --exp {{arg}} matched {{matched}} experiments: {{candidates}}\n  fix: use one of the exact ids above, or a longer prefix — each --exp in a compare must resolve to exactly one experiment\n",
-    "cli.show.locatorExpConflict": "error: {{locator}} cannot combine with repeated --exp ({{exp}})\n  fix: drop the extra --exp flags — a locator already pins one attempt to one experiment; for a multi-condition comparison, drop the locator and use eval id prefixes with --exp instead\n",
-    "cli.show.statsLocatorConflict": "error: --stats cannot combine with a locator ({{locator}}) — a single attempt has no stability to measure\n  fix: drop the locator and use eval id prefixes / --exp to select a range for --stats\n",
-    "cli.show.statsReportConflict": "error: --stats cannot combine with --report ({{report}}) — --stats is a zero-config slice, it does not render a user report tree\n  fix: drop --report to use --stats, or drop --stats and put a StabilityMatrix in your own report file\n",
-    "cli.show.grepExpandConflict": "error: --grep and --expand cannot combine — --grep scans for matching cards, --expand prints one card in full\n  fix: drop one of the two flags\n",
-    "cli.show.grepExecutionOnly": "error: --grep only combines with --execution — it narrows that block's text rendering, not a slice of its own\n  fix: add --execution, or drop --grep\n",
-    "cli.show.expandExecutionOnly": "error: --expand only combines with --execution — it narrows that block's text rendering, not a slice of its own\n  fix: add --execution, or drop --expand\n",
-    "cli.show.grepInvalidPattern": "error: --grep pattern is not a valid JS regular expression: \"{{pattern}}\" ({{message}})\n  fix: fix the pattern syntax (it is passed to `new RegExp(...)`)\n",
-    "cli.show.expandMultiAttempt": "error: --expand requires the range to resolve to exactly one attempt, got {{count}}\n  fix: narrow the range to a single attempt — an eval id prefix matching one eval, or @<locator>\n",
-    "cli.show.expandNotFound": "error: {{message}}\n  fix: use a handle from a truncated card's own hint (t<turn>.c<card> or cmd<n>), or drop --expand to see the whole attempt\n",
-    "cli.show.historyReportConflict": "`--history` and `--report` are mutually exclusive: both take over the main output. --history is the host's per-attempt execution timeline; for run-level trends, compose exp.runs inside your report file instead.\n",
-    "cli.show.jsonReportConflict": "error: --json cannot combine with --report ({{report}}) — a report tree says how to look at the data, --json says what the data is\n  fix: drop --report to use --json, or drop --json and read the report tree as text/HTML\n",
-    "cli.show.jsonExpandConflict": "error: --json cannot combine with --expand — JSON never truncates cards, there is nothing to expand\n  fix: drop --expand; --json already returns the full untruncated value\n",
-    "cli.show.jsonMultiEvidenceConflict": "error: --json requires exactly one of --source/--execution/--timing/--diff at a time — the envelope's \"view\" is a single value, there is no combined shape for more than one\n  fix: drop the extra evidence flags, or make one --json call per flag\n",
-    "cli.show.locatorMalformed": "{{message}}\n",
-    "cli.show.locatorNotFound": "{{message}}\n",
-    "cli.show.locatorAmbiguous": "{{message}}\n",
-    "cli.eval.noMatch": "没有匹配的 eval:{{patterns}}。\n",
-    "cli.eval.noMatchHintExperiment": "提示:\"{{pattern}}\" 是实验{{kind}},你大概想跑:niceeval exp {{pattern}}\n",
-    "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.exp.teardownNoEvalPatterns": "--teardown 只选择实验,不跑任何 eval,所以不能再带 eval id 前缀。用法:niceeval exp <experiment 路径> --teardown\n",
-    "cli.exp.teardownDone": "teardown 完成:{{experimentId}}\n",
-    "cli.exp.teardownFailed": "teardown 失败:{{experimentId}}:{{message}}\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.noEvalPrefixMatch": "没有 eval 匹配前缀:{{pattern}}(在 {{selection}} 选中的实验里)。\n" +
-        "第一个之后的位置参数选的是 eval id 前缀。要跑另一个实验,\n" +
-        "把它作为独立命令跑:niceeval exp {{pattern}}\n",
-    "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.experiment.strictOnPoints": "实验 \"{{experimentId}}\" 选中的 {{count}} 个 eval 全是计分制(defineScoreEval),而 `--strict` 对它们一件事都做不了:\n" +
-        "这个开关只把「带通过线的 soft 断言」翻成 gate,而计分制的判定只认前置 `.gate()` 中止——丢分从不改 verdict。\n" +
-        "去掉 `--strict` 重跑;要收紧计分制的判定,把该硬的检查写成前置 `.gate()`。\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(要求 coding agent 写 eval 前先读随包文档)。\n",
-    "cli.init.esmHint": "提示:项目 package.json 没有 \"type\": \"module\"。niceeval 两种形态都能跑,但 CommonJS 形态下 config / eval 文件用不了顶层 await——建议在 package.json 加上 \"type\": \"module\"。\n",
-    "cli.interruptCleanup": "\n收到中断,正在清理沙箱容器…(再按一次强制清理并退出)\n",
-    "cli.list.header": "发现 {{count}} 个 eval:\n",
-    "cli.noAgent": "未指定 agent(用 --agent <name>)。\n",
-    "cli.none": "(无)",
-    "cli.pressCtrlC": "按 Ctrl+C 退出。\n",
-    "cli.resultsPath": "结构化结果:{{path}}(run.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 [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.view.exportedDir": "已导出静态查看站:{{out}}(整个目录可直接静态托管;本地打开 {{out}}/index.html 需经 http 服务,file:// 下 artifact fetch 不可用)\n",
-    "cli.view.incompatible": "{{dir}}: 由 niceeval {{producer}} 写入(schemaVersion {{schemaVersion}}),当前 CLI 只读 schemaVersion {{supported}}。\n运行 `{{command}}` 查看这份报告。\n",
-    "cli.view.noResults": "{{root}} 下没有结果。先 `niceeval exp` 跑一轮,再 `niceeval view`。\n",
-    "cli.view.incompatibleForeign": "{{dir}}: 由 {{name}} {{version}} 写入(schemaVersion {{schemaVersion}}),当前 CLI 只读 schemaVersion {{supported}}。\n请用写出它的那个工具查看这份报告。\n",
-    "cli.view.urls": "niceeval view 可在浏览器打开：\n{{urls}}\n",
-    "codex.envPathManaged": "codexAgent config.env.PATH 不支持:PATH 是 Sandbox 受管变量,静默丢弃或覆盖它会让 hooks 与子进程读到错误的可执行文件,且零报错。" +
-        "改用 Sandbox factory 的 pathPrepend 选项前置目录(见 docs/feature/sandbox/library.md)。",
-    "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/tutorials/sandbox-agent.mdx",
-    "context.skipEmpty": "skip() 需要一个非空理由。",
-    "define.agentNameRequired": "defineDirectAgent 需要 name。",
-    "define.evalIdRejected": "defineEval 不接受 id —— id 由文件路径推导。",
-    "define.evalEnvironmentEmpty": "defineEval 的 environment 如有提供，必须是非空的 profile id。",
-    "define.evalTestRequired": "defineEval 需要一个 async test(t) 函数。",
-    "define.evalEvaluationKindRejected": "defineEval 不接受 evaluationKind —— 恒定为 \"pass\"(通过制)。计分制请用 defineScoreEval。",
-    "define.evalConfigHashRejected": "defineEval 不接受 configHash —— configHash 由运行规划计算。",
-    "define.scoreEvalIdRejected": "defineScoreEval 不接受 id —— id 由文件路径推导。",
-    "define.scoreEvalEnvironmentEmpty": "defineScoreEval 的 environment 如有提供，必须是非空的 profile id。",
-    "define.scoreEvalTestRequired": "defineScoreEval 需要一个 async test(t) 函数。",
-    "define.scoreEvalEvaluationKindRejected": "defineScoreEval 不接受 evaluationKind —— 恒定为 \"points\"(计分制)。通过制请用 defineEval。",
-    "define.scoreEvalConfigHashRejected": "defineScoreEval 不接受 configHash —— configHash 由运行规划计算。",
-    "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.experimentClassifyFailureNotFunction": "experiment.classifyFailure 必须是函数((failure) => FailureClass | undefined):它识别以第三方错误形态浮出的失败,认不出的一律返回 undefined 交给后续链路。",
-    "define.experimentIdRejected": "defineExperiment 不接受 id —— id 由文件路径推导。",
-    "define.sandboxAgentNameRequired": "defineSandboxAgent 需要 name。",
-    "define.sandboxAgentEnsureRequired": "defineSandboxAgent 需要 ensure 声明。",
-    "define.sandboxCreateRequired": "defineSandbox 需要一个 create() 函数。",
-    "define.sandboxNameRequired": "defineSandbox 需要 name。",
-    "docker.commandTimeout": "Command timed out after {{timeoutMs}}ms",
-    "docker.containerNotInitialized": "Container not initialized",
-    "docker.imagePullDone": "Docker image ready: {{image}}",
-    "docker.imagePullStart": "Pulling Docker image: {{image}}...",
-    "docker.readFileFailed": "Failed to read file {{path}}: {{stderr}}",
-    "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",
-    "feedback.human.counts": "共 {{total}} · 复用 {{reused}} · 运行中 {{running}} · 排队 {{queued}} · 通过 {{passed}} · 失败 {{failed}} · 出错 {{errored}} · 跳过 {{skipped}}",
-    "feedback.human.diffHint": "Diff:    niceeval show {{locator}} --diff",
-    "feedback.human.evalHint": "Eval:    niceeval show {{locator}} --source",
-    "feedback.human.exampleLocator": "例如 {{locator}}",
-    "feedback.human.failureFacts": "facts ×{{count}}",
-    "feedback.human.failuresHeader": "FAILURES",
-    "feedback.human.failuresSoFar": "已发生 {{count}} 条",
-    "feedback.human.failuresTotalKinds": "共 {{total}} 条 · {{kinds}} 种形态",
-    "feedback.human.heartbeat": "已运行 {{elapsed}} · {{counts}}",
-    "feedback.human.inspect": "Inspect: niceeval show {{locator}}",
-    "feedback.human.keptSandboxesHeader": "KEPT SANDBOXES",
-    "feedback.human.moreActive": "… 还有 {{count}} 项运行中",
-    "feedback.human.moreFailureKinds": "还有 {{count}} 种形态 — niceeval view",
-    "feedback.human.nextHeader": "NEXT",
-    "feedback.human.plan": "{{total}} 个 attempt · {{evals}} 个 eval × {{configs}} 个配置 · 并发 {{concurrency}}",
-    "feedback.human.planHeader": "PLAN",
-    "feedback.human.resultFailed": "FAILED",
-    "feedback.human.resultIncomplete": "INCOMPLETE",
-    "feedback.human.resultInterrupted": "INTERRUPTED",
-    "feedback.human.resultPassed": "PASSED",
-    "feedback.human.resultsHeader": "RESULTS",
-    "feedback.human.resultsMore": "… 还有 {{count}} 个",
-    "feedback.human.reuse": "{{total}} 中 {{reused}} 条来自缓存 · {{toRun}} 待跑",
-    "feedback.human.summaryLine": "{{passed}} 通过 · {{failed}} 失败 · {{errored}} 出错  (复用 {{reused}})",
-    "feedback.human.summaryIncompleteLine": "{{passed}} 通过 · {{failed}} 失败 · {{errored}} 出错 · {{unstarted}} 未派发  (复用 {{reused}})",
-    "feedback.human.summaryAllReusedLine": "{{passed}} 通过 · {{failed}} 失败 · {{errored}} 出错  (全部 {{reused}} 条复用)",
-    "feedback.human.suppressedFailures": "… 还有 {{count}} 条失败被折叠",
-    "feedback.human.trace": "Trace:   niceeval show {{locator}} --execution",
-    "feedback.human.warningsHeader": "WARNINGS",
-    "feedback.phase.agentSetup": "agent 预置",
-    "feedback.phase.agentEnsure": "正在准备 Agent",
-    "feedback.phase.stateLoad": "载入实验状态",
-    "feedback.phase.stateSave": "回存实验状态",
-    "feedback.phase.evalRun": "运行 eval",
-    "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.phase.judgePrecheck": "judge 预检",
-    "feedback.human.hookDone": "完成",
-    "feedback.human.hookFailed": "失败",
-    "feedback.human.precheckJudge": "预检 judge 配置",
-    "feedback.human.precheckJudgeDone": "judge 配置就绪",
-    "feedback.human.precheckJudgeFailed": "judge 预检失败",
-    "feedback.human.planExperimentConcurrency": "{{experimentId}} ≤{{limit}}",
-    "feedback.human.countsWithElsewhere": "共 {{total}} · 复用 {{reused}} · 运行中 {{running}} · 等待中 {{elsewhere}} · 排队 {{queued}} · 通过 {{passed}} · 失败 {{failed}} · 出错 {{errored}} · 跳过 {{skipped}}",
-    "feedback.human.waitingOnAnotherRun": "等待另一个并行 run",
-    "feedback.human.lockWaitDetail": "{{count}} 条用例 · pid {{pid}}",
-    "feedback.human.lockWaitStarted": "等待另一个并行 run · {{experimentId}}({{count}} 条用例,pid {{pid}})",
-    "feedback.human.lockWaitResolved": "等待结束 · {{experimentId}}({{summary}},{{elapsed}})",
-    "feedback.human.lockWaitCarried": "{{count}} 条携入",
-    "feedback.human.lockWaitDispatched": "{{count}} 条自跑",
-    "feedback.human.lockedRowSuffix": "locked",
-    "feedback.phase.sandboxPrepare": "准备沙箱",
-    "feedback.phase.assertions": "评估断言",
-    "feedback.phase.teardown": "清理中",
-    "feedback.phase.telemetryCollect": "收集 trace",
-    "feedback.phase.telemetryConfigure": "配置 telemetry",
-    "feedback.phase.workspaceBaseline": "准备工作区",
-    "feedback.phase.workspaceDiff": "采集 diff",
-    "feedback.rendererError": "  · [feedback] renderer 处理 {{context}} 失败(已忽略):{{message}}\n",
-    "hitl.answerNeedsOptionOrText": "t.respond 的对象形式必须在 optionId 与 text 中恰好选择一项。",
-    "hitl.invalidOption": "回答 \"{{optionId}}\" 不是请求 {{requestId}} 的可选项({{options}})。",
-    "hitl.noOptions": "该请求没有可选项",
-    "hitl.requestMissingId": "该 input.requested 请求没有稳定的 id,无法生成 responses——adapter 侧要给每条待回答请求一个稳定 id。",
-    "hitl.respondAllEmpty": "没有待回答的 input.requested 请求,respond() / respondAll() 无法工作;先用 t.parked() 确认停轮,再用 t.requireInputRequest() 或 t.respond() 回答。",
-    "hitl.respondEmpty": "t.respond(...) 至少需要一个回答。",
-    "hitl.stringAmbiguous": "有 {{count}} 条待回答请求,字符串回答无法对位,请用 { request, optionId } 或 { request, text } 对象形式显式指名。",
-    "judge.apiKeyMissing": "judge 缺少 API key:设置 NICEEVAL_JUDGE_KEY,或用 judge.apiKeyEnv 指向别的环境变量。",
-    "judge.httpError": "judge HTTP {{status}}: {{body}}",
-    "judge.probeFailed": "judge 预检失败:{{endpoint}}({{model}}): {{error}}",
-    "judge.probeTimeout": "judge 预检失败:{{endpoint}}({{model}})连续 {{attempts}} 次各 {{seconds}}s 超时 —— 端点接受连接但不回,先查同一账号的其它流量是否占满了网关并发,再核对 judge 的 baseUrl(NICEEVAL_JUDGE_BASE)",
-    "judge.probeMissingKey": "judge 模型 {{model}} 缺少 API key —— 请配置 {{envHint}}",
-    "live.more": "… 其余 {{hidden}} 项({{running}} 运行中 · {{waiting}} 等待 · {{done}} 已完成)",
-    "live.running": "  正在运行 {{totalRuns}} 次 ({{evals}} eval × {{configs}} 配置, 并发 {{concurrency}})       {{completed}}/{{total}} 完成",
-    "live.runningUnknown": "  正在运行…  {{completed}}/{{total}} 完成",
-    "live.waiting": "排队等待中…",
-    "local.commandTimeout": "Command timed out after {{timeoutMs}}ms",
-    "local.dirMissing": "本地沙箱目录不存在:{{dir}}。请先创建它,或让 localSandbox({ dir }) 指向一个已存在的目录。",
-    "local.dirNotWritable": "本地沙箱目录不可写:{{dir}}({{message}})",
-    "local.notARepo": "当前目录(及其上级目录)都不在任何 git 仓库内,localSandbox() 没有确定的 workdir 可用。cd 进要跑的仓库再执行,或显式传 localSandbox({ dir: \"/path/to/repo\" })。",
-    "local.userUnsupported": "本地 provider 不支持切换执行身份,收到 { user: \"{{user}}\" }——niceeval 不会在你的机器上提权或换身份。需要别的身份请改用容器 provider(docker / e2b / vercel)。",
-    "report.assertionThreshold": " (得分 {{score}} < {{threshold}})",
-    "report.error": "错误",
-    "report.errored": "错误",
-    "report.failed": "失败",
-    "report.gate": "gate",
-    "report.passed": "通过",
-    "report.result": "\n结果:{{parts}}  ({{duration}} · {{tokens}}{{cost}})\n\n",
-    "report.runStart": "\n本次运行 {{count}} 个 eval{{extra}}(并发 {{concurrency}})\n\n",
-    "report.runStartExtra": " × {{configs}} 配置 = {{totalRuns}} 次运行",
-    "report.viewHint": "运行 `pnpm exec niceeval view` 以图形化查看结果。\n",
-    "report.skipped": "跳过",
-    "report.soft": "soft",
-    "report.summary.errored": "{{count}} 错误",
-    "report.summary.failed": "{{count}} 失败",
-    "report.summary.passed": "{{count}} 通过",
-    "report.summary.skipped": "{{count}} 跳过",
-    "report.table.agent": "Agent",
-    "report.table.avgDuration": "平均耗时",
-    "report.table.cost": "预估成本",
-    "report.table.default": "默认",
-    "report.table.duration": "耗时",
-    "report.table.eval": "Eval",
-    "report.table.evalTitle": "各 Eval:",
-    "report.table.experiment": "实验",
-    "report.table.experimentsTitle": "实验",
-    "report.table.model": "模型",
-    "report.table.reason": "原因",
-    "report.table.result": "结果",
-    "report.table.runs": "轮次",
-    "report.table.status": "状态",
-    "report.table.successRate": "成功率",
-    "report.table.tokens": "Tokens",
-    "otel.noSpans": "otel:本轮 0 span —— 端点没接上?(env 没注入 / 服务没重启 / 没 flush)",
-    "otel.portInUse": "OTLP 接收端口 {{port}} 已被占用(另一个进程占着这个端口)。在 defineConfig({ telemetry: { port } }) 里换一个空闲端口,或者停掉占用 {{port}} 的进程后重试。",
-    "otel.windowAttribution": "otel:span 未带本轮 traceparent,按时间窗口归属(该 agent 的轮次已串行;应用支持 W3C 传播后自动并发)",
-    "runner.diffProgress": "采 diff:{{changed}} 改 / {{deleted}} 删",
-    "runner.failFast": "{{evalId}} 的错误 {{code}} 连续复现,判定为确定性错误;停止派发该配置剩余的 attempt(fail-fast)。",
-    "runner.driveAgent": "驱动 agent…",
-    "runner.evalSetup": "eval setup(装依赖)…",
-    "runner.interrupted": "  · 已中断:沙箱容器已清理,输出本次已完成的部分结果。\n",
-    "runner.otlpInSandbox": "OTLP in-sandbox collector → {{endpoint}}{{proto}}",
-    "runner.otlpOverride": "OTLP 接收器(覆盖 host) → {{endpoint}}",
-    "runner.otlpReceiver": "OTLP 接收器 → {{endpoint}}{{proto}}",
-    "runner.otlpShared": "OTLP 共享接收器(run 级) → {{endpoint}}",
-    "runner.providerExclusiveSerial": "  · [sandbox] “{{provider}}” provider 强制 attempt 一次只跑一个(独占串行);不论 --max-concurrency 写了 {{concurrency}},它的实际并发恒为 1\n",
-    "runner.directAgentSandboxUnavailable": "Direct Agent 没有 sandbox.{{method}}；请改用 Sandbox Agent 或移除 workspace 断言。",
-    "runner.reporterDiagnostic": "  · [diagnostic] {{stage}} 失败(已忽略):{{message}}\n",
-    "runner.skip": "skip:{{reason}}",
-    "runner.startAgentSetup": "agent setup(装 CLI / 写配置)…",
-    "runner.startAgentEnsure": "正在准备 Agent…",
-    "runner.startAgentTracing": "agent tracing(写 otel 导出配置)…",
-    "runner.startSandbox": "起沙箱…",
-    "runner.startSandboxSetup": "sandbox setup(环境预置钩子)…",
-    "runner.startSandboxTeardown": "sandbox teardown(环境预置钩子)…",
-    // 超时消息必带「上限是哪一层给的」:from flag / experiment / eval / config 四值,对应
-    // runner/timeout.ts 的解析链(链末端没有内置默认,四层都没声明就不会撞线)。
-    // 撞线时用户要的下一步是「去哪儿把它调大」。
-    "runner.timeout": "attempt 超时({{timeoutMs}}ms, from {{source}})\n最近进度:\n{{recentLogs}}",
-    "runner.traceSelected": " → 留 {{count}}(按语义)",
-    "runner.useDirectAgent": "使用 Direct Agent（不创建 Sandbox）…",
-    "sandbox.deadlineExceedsSession": "error: 这条 attempt 的超时上限({{timeoutMs}}ms)超过 {{provider}} 单个会话的寿命({{limitMs}}ms),沙箱会在 attempt 跑到一半时被回收。\n" +
-        "  fix: 把 timeoutMs 调到 {{limitMs}}ms 以内,或在 sandbox spec 上声明更长的 lifetimeMs(账号档位允许的话)。\n",
-    "sandbox.providerNotImplemented": "{{provider}} sandbox provider not implemented; use docker, vercel, e2b, or local",
-    "sandbox.missingSpec": "沙箱型 agent 需要一条带 template 的 SandboxLayer，但 Eval 与 Experiment 都没有声明——请从 \"niceeval/sandbox\" 使用 dockerImageSandbox({ image })、dockerComposeSandbox({ file, workspaceService })、dockerfileSandbox({ context })、vercelSandbox({ snapshotId })、e2bSandbox({ template }) 或 localSandbox()。\n" +
-        "  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",
-    "sandbox.forceCleanup": "  · [sandbox] 强制清理 {{count}} 个沙箱…\n",
-    "sandbox.provisionReconcileFailed": "  · [sandbox] 创建重试前对账失败,放弃重试(可能已创建的实例无法核实/销毁):{{error}}\n",
-    "sandbox.provisionRetry": "  · [sandbox] 创建被限流,{{delayMs}}ms 后重试(第 {{attempt}}/{{maxAttempts}} 次)…\n",
-    "sandbox.stopFailed": "  · [sandbox] 停沙箱 {{id}} 失败(已忽略;它会继续运行并计费,直到该 provider 自己的超时——如果有——回收它):{{message}}\n",
-    "sandbox.stopTimeout": "stop 超时({{timeoutMs}}ms)",
-    // 三要素齐备的传输超时(见 docs/error-feedback.md「超时报错的三要素」):哪个操作、
-    // 对什么对象、这条线是谁给的。最后一句是关键——它把人从「去调 --timeout」引开。
-    "sandbox.transferTimeout": "{{provider}} {{operation}} 传输超时:{{object}}。这是 provider SDK / HTTP 往返的超时,不是 attempt 的 timeoutMs 预算——调大 --timeout 不会让它变好。" +
-        "修法:把这一次传输拆成多批小的、把大 fixture 预置进镜像 / 模板,或改成在沙箱内直接下载。",
-    // 沙箱内 OTLP 采集器的落点是系统临时目录(见 docs/feature/sandbox/architecture.md
-    // 「provider 的可写保证不止 workdir」):写不进去是镜像环境缺陷,报错点名路径与修法,
-    // provider 的原始 500 串留在 cause 里,不进这条消息。
-    "o11y.sandboxTempNotWritable": "沙箱内 OTLP 采集器写不进 {{path}}:系统临时目录对沙箱的运行用户不可写。这是镜像环境缺陷,不是 eval 或 niceeval 配置问题——" +
-        "provider 的可写保证不止 workdir,runner 要在 workdir 之外放采集器与变更分类账。" +
-        "修法:让 /tmp 对运行用户可写(镜像里 `chmod 1777 /tmp`,或换一个不把 /tmp 挂成只读的镜像 / 用户),修好后重跑即续上。",
-    "assertions.evaluationError": "断言评估出错: {{error}}",
-    "assertions.pointsInvalid": ".points({{n}}) 非法;给分必须是正有限数(n > 0)。",
-    "assertions.scoreInvalid": "t.score({{label}}, {{n}}) 非法;给分必须是非负有限数(n >= 0)。",
-    "session.fileFallback": "[file]",
-    "session.tools": "{{count}} 工具",
-    "session.turn.primary": "第{{turn}}轮",
-    "session.turn.secondary": "会话{{session}}·第{{turn}}轮",
-    "session.turnRetry": "turn 重试 {{attempt}}/{{maxAttempts}}({{reason}})——等待 {{seconds}}s",
-    "session.turnRetryBudgetExhausted": " · attempt 重试预算已耗尽({{maxRetries}} 次重试,{{reason}})",
-    "session.turnRetrySendExhausted": " · 重试已耗尽({{maxAttempts}} 次尝试,{{reason}})",
-    "util.requiredEnv": "缺少必需的环境变量 {{name}}(请在 .env 里配置)。",
-    "vercel.fileNotFound": "File not found: {{path}}",
-    "vercel.rotateFailed": "[VercelSandbox] session rotate failed ({{seconds}}s): {{error}}",
-    "vercel.rotated": "[VercelSandbox] session rotated after {{seconds}}s → {{sessionId}}",
-    "vercel.userUnsupported": "Vercel Sandbox provider 的命令级 user 只认 { user: \"root\" }(映射 sudo: true),收到 { user: \"{{user}}\" }。需要别的身份请改用容器 provider(docker / e2b)。",
-};

package/dist/record/annotated-source.d.ts

@@ -1,126 +0,0 @@
-import type { AssertionResult, PhaseTiming, ScoreEntry, SourceArtifact, SourceLoc, StreamEvent } from "../types.ts";
-/**
- * 标回 `t.send(...)` 调用行的一轮 turn 头行事实(契约见 docs/feature/reports/show.md
- * 「--eval:把断言放回源码」)。身份标签与 --execution / --timing / diff windows 同一套;
- * 回复全文与轮内卡片不进这个模型——源码页只回答「这行代码对应哪一轮、这一轮成了没成」。
- */
-export interface SendAnnotation {
-    /** `turn<N>` 或 `session<K>/turn<N>`；已有 artifact 标签按不透明字符串保留。 */
-    label: string;
-    /** 轮的终态;时间树只记 failed 位,waiting 需要事件流佐证时由派生方给。 */
-    status: "completed" | "failed" | "waiting";
-    /** 该轮墙钟;时间树缺这一轮的节点时省略。 */
-    durationMs?: number;
-    /** send 调用位置(用户消息事件的 loc)。 */
-    loc: SourceLoc;
-    /** 与断言、直接给分共用的 attempt 级发生顺序；历史事件可省略。 */
-    sourceOrder?: number;
-}
-/**
- * 从标准事件流与阶段时间树派生 send 标注:第 i 条用户消息开第 i 轮(与 --execution 的
- * 分轮边界同一条规则,见 show/render.ts::executionText),头行事实取 `eval.run` 下第 i 个
- * turn 节点;用户消息没有 loc 的轮不产出标注。纯函数,无 IO。
- */
-export declare function deriveSendAnnotations(events: readonly StreamEvent[] | null, phases: readonly PhaseTiming[] | undefined): SendAnnotation[];
-/** 调用树的面无关完整证据。 */
-export type LineAnnotation = {
-    kind: "assertion";
-    assertion: AssertionResult;
-} | {
-    kind: "score";
-    score: ScoreEntry;
-} | {
-    kind: "send";
-    send: SendAnnotation;
-};
-export interface SourceTreeLine {
-    line: number;
-    text: string;
-    annotations: LineAnnotation[];
-    calls: SourceCall[];
-    aborted?: true;
-}
-export interface SourceNode {
-    file: string;
-    sha256: string;
-    lines: SourceTreeLine[];
-}
-export interface SourceCall {
-    summary: SourceCallSummary;
-    target: {
-        kind: "source";
-        node: SourceNode;
-    } | {
-        kind: "package";
-        package: string;
-        calls: SourceCall[];
-    } | {
-        kind: "unavailable";
-        file: string;
-        line?: number;
-        annotations: LineAnnotation[];
-        calls: SourceCall[];
-    };
-}
-export interface SourceCallSummary {
-    checks: number;
-    passed: number;
-    failed: number;
-    unavailable: number;
-    points?: {
-        earned: number;
-        available: number;
-    };
-    aborted: boolean;
-}
-export interface AnnotatedSourceTree {
-    spine: SourceNode;
-    detached: SourceNode[];
-    unmapped: {
-        assertions: AssertionResult[];
-        scores: ScoreEntry[];
-    };
-    summary: SourceCallSummary;
-}
-export interface ProjectedSourceLine extends Omit<SourceTreeLine, "calls"> {
-    calls: ProjectedSourceCall[];
-}
-export interface SourceContentNode extends Omit<SourceNode, "lines"> {
-    lines: ProjectedSourceLine[];
-}
-export interface ProjectedSourceCall extends Omit<SourceCall, "target"> {
-    open: boolean;
-    target: {
-        kind: "source";
-        node: SourceContentNode;
-    } | {
-        kind: "package";
-        package: string;
-        calls: ProjectedSourceCall[];
-    } | {
-        kind: "unavailable";
-        file: string;
-        line?: number;
-        annotations: LineAnnotation[];
-        calls: ProjectedSourceCall[];
-    };
-}
-export interface SourceContent extends Omit<AnnotatedSourceTree, "spine" | "detached"> {
-    spine: SourceContentNode;
-    detached: SourceContentNode[];
-}
-/** 用 entry role 决定主干，绝不按命中数猜入口。 */
-export declare function assembleSourceTree(input: {
-    entry: SourceArtifact;
-    sources: readonly SourceArtifact[];
-    assertions: readonly AssertionResult[];
-    scoreEntries: readonly ScoreEntry[];
-    sends: readonly SendAnnotation[];
-    abort?: SourceLoc;
-}): AnnotatedSourceTree;
-/** 唯一的裁行入口；full/file/default/web 不会改变完整树或事实顺序。 */
-export declare function projectSourceView(source: AnnotatedSourceTree, options: {
-    mode: "default" | "full" | "file" | "web";
-    file?: string;
-    budgetLines?: number;
-}): SourceContent;