niceeval 0.6.1 → 0.6.2

package/src/i18n/en.ts

@@ -14,6 +14,8 @@ export const en = {
   "bub.installFailed": "bub install failed after {{attempts}} attempts:\n{{tail}}",
   "bub.setupNotRun": "bub adapter setup() has not run in this sandbox (missing home/workspace info). The runner must call setup before send.",
   "checkpoint.emptyTar": "checkpoint: tar is empty (paths: {{paths}})",
+  "checkpoint.archiveFailed": "checkpoint archive failed (exit {{exitCode}}): {{detail}}",
+  "checkpoint.restoreFailed": "checkpoint restore failed (exit {{exitCode}}): {{detail}}",
   "skill.localMissing": "Local skill path \"{{path}}\" does not exist (resolved to {{resolved}}). Paths are resolved from the project root you run niceeval in.",
   "skill.localDirNoSkillFile": "Local skill directory \"{{path}}\" has no SKILL.md. A directory-shaped skill must contain SKILL.md at its root.",
   "skill.localUnsupportedShape": "Local skill path \"{{path}}\" has an unsupported shape. Accepted: a directory containing SKILL.md, or a single .md file.",
@@ -23,7 +25,21 @@ export const en = {
   "skill.repoUnknownSkill": "Repo skill {{source}} (ref: {{ref}}) has no skill named \"{{skill}}\". Available: {{available}}.",
   "skill.copyFailed": "Could not install skill \"{{name}}\" into {{dest}}:\n{{tail}}",
   "plugin.marketplaceFailed": "Could not connect {{agent}} marketplace \"{{name}}\" (source: {{source}}, ref: {{ref}}):\n{{tail}}",
+  "plugin.marketplaceVerifyFailed": "Could not read back the registered marketplace list after adding {{agent}} marketplace \"{{name}}\" ({{command}}):\n{{tail}}",
+  "plugin.marketplaceNameMismatch":
+    "{{agent}} marketplace name mismatch: the configured name \"{{expected}}\" (source: {{source}}) is not in the registered list after add; actually registered: {{actual}}. " +
+    "marketplace.name must equal the name declared in the target repo's manifest — use the real name.",
   "plugin.installFailed": "Could not install {{agent}} plugin \"{{name}}\" (marketplace: {{marketplace}}):\n{{tail}}",
+  "nativeConfig.pathNotProjectRelative":
+    "{{agent}} {{field}} only accepts relative paths inside the project root, got \"{{path}}\". Absolute paths, `..` segments and `~` paths are rejected; copy configs from outside the project into it first.",
+  "nativeConfig.missing":
+    "{{agent}} {{field}} points to a missing file: \"{{path}}\" (resolved to {{resolved}}). Paths resolve from the project root you run niceeval in (the directory containing niceeval.config.ts), not from eval / experiment source files.",
+  "nativeConfig.escapesRoot": "{{agent}} {{field}} \"{{path}}\" resolves through a symlink to outside the project root ({{resolved}}). The config file must physically live inside the project root.",
+  "nativeConfig.notFile": "{{agent}} {{field}} \"{{path}}\" is not a regular file. Point it at a complete official config file.",
+  "nativeConfig.invalidSyntax": "{{agent}} {{field}} \"{{path}}\" is not valid {{format}}: {{detail}}",
+  "nativeConfig.reservedKeys":
+    "{{agent}} {{field}} \"{{path}}\" contains reserved keys: {{keys}}. These keys are owned by the experiment and the Adapter (model, auth, MCP and OTel are layered separately) — remove them from the file.",
+  "nativeConfig.uploadFailed": "Could not upload native config file \"{{path}}\" into the sandbox ({{dest}}):\n{{tail}}",
   "cli.all": "(all)",
   "cli.browserOpenFailed": "Could not open the browser automatically. Open manually: {{url}}\n",
   "cli.clean.done": "Deleted .niceeval/ historical run artifacts.\n",
@@ -39,6 +55,7 @@ export const en = {
   "cli.dry.row": "  {{who}}{{experiment}}: {{evals}}  ×{{runs}}\n",
   "cli.error": "niceeval error: {{error}}\n",
   "cli.flag.invalidNumber": "Flag --{{flag}} expects a number, got \"{{value}}\".\n",
+  "cli.flag.invalidOutput": "Flag --output expects one of auto|human|agent|ci, got \"{{value}}\".\n",
   "runner.budgetUnenforceable":
     "budget for {{budgetKey}}: several attempts completed without any cost data (agent reports no usage and the model is not in the price table) — the budget cannot be enforced for this agent; continuing without the guard.\n",
   "judge.modelMissing":
@@ -62,6 +79,7 @@ export const en = {
     "        assertions mapped back to source lines\n" +
     "      --execution   this attempt's execution event stream (messages/thinking/\n" +
     "        Skill loads/tool calls); OTel adds timing to the same node when present\n" +
+    "      --timing      unified timing tree for the attempt (phases + hooks/commands/turns + per-turn OTel)\n" +
     "      --diff[=file] sandbox workspace file-change summary; =file expands one file\n" +
     "      --history   cross-run timeline (mutually exclusive with --report)\n" +
     "      --run <dir>   pin a results dir    --experiment <id>   one experiment\n" +
@@ -73,11 +91,13 @@ export const en = {
     "      --experiment <id> one experiment\n" +
     "      --out <dir> exports a static site: index.html plus the viewer\n" +
     "      artifacts, ready for any static host\n" +
+    "  niceeval sandbox list|enter|history|diff|stop  inspect & destroy sandboxes kept by --keep-sandbox\n" +
     "  niceeval clean                                      delete .niceeval/ artifacts\n" +
     "  niceeval init                                       scaffold config + evals/\n\n" +
     "Flags:\n" +
     "  --runs n  --max-concurrency n  --timeout ms  --budget usd  --tag t\n" +
-    "  --early-exit / --no-early-exit  --strict  --force  --dry  --quiet\n" +
+    "  --early-exit / --no-early-exit  --strict  --force  --dry\n" +
+    "  --output auto|human|agent|ci\n" +
     "  --junit path  --json path  --out dir  --port n  --open / --no-open  -h, --help  -v, --version\n\n" +
     "Positional args only select which evals to run (id prefixes); which agent and\n" +
     "how to run come from experiments/ + flags. Env overrides (flag > env > config):\n" +
@@ -95,8 +115,11 @@ export const en = {
   "cli.eval.noMatch": "No eval matched: {{patterns}}.\n",
   "cli.eval.noMatchHintExperiment": "Hint: \"{{pattern}}\" is an experiment{{kind}}; you probably meant: niceeval exp {{pattern}}\n",
   "cli.eval.noMatchKnown": "Discovered {{count}} evals: {{evals}}\n",
-  "cli.exp.agentModelFlagUnsupported": "`--agent` / `--model` cannot override an experiment. Add or copy a config file under experiments/ instead.\n",
+  "cli.exp.agentModelFlagUnsupported": "experiment runs do not support --agent / --model. Add or copy an experiment file and change its model instead.\n",
+  "cli.exp.viewerFlagUnsupported": "`{{flag}}` only applies to niceeval {{command}}, not niceeval exp.\n",
   "cli.experiment.noMatch": "No experiment matched: {{arg}}. Discovered: {{experiments}}\n",
+  "cli.experiment.viewerCommandHint": "Did you mean: niceeval {{command}}{{args}}\n",
+  "cli.experiment.noEvalsSelected": "No evals selected: {{selection}} matched 0 evals. Available experiments: {{experiments}}.\n",
   "cli.experimentGroup": " group",
   "cli.fallbackCleanupTimeout": "\ngraceful cleanup timed out; force-cleaning sandboxes...\n",
   "cli.forceCleanupExit": "\nForce-cleaning sandboxes and exiting...\n",
@@ -128,6 +151,7 @@ export const en = {
   "define.evalIdRejected": "defineEval does not accept id; ids are derived from file paths.",
   "define.evalTestRequired": "defineEval requires an async test(t) function.",
   "define.experimentAgentRequired": "defineExperiment requires agent.",
+  "define.experimentFlagNotJson": "experiment.flags.{{key}} is not JSON-serializable (functions / undefined / cycles / bigint are not allowed); flags are persisted verbatim into result snapshots and must be plain JSON.",
   "define.experimentIdRejected": "defineExperiment does not accept id; ids are derived from file paths.",
   "define.sandboxAgentNameRequired": "defineSandboxAgent requires name.",
   "define.sandboxCreateRequired": "defineSandbox requires a create() function.",
@@ -138,6 +162,41 @@ export const en = {
   "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": "budget exhausted for {{experimentId}} (spent {{spent}}, unstarted {{unstarted}})",
+  "feedback.human.compare": "Compare: niceeval view {{group}}",
+  "feedback.human.counts": "{{total}} total · {{reused}} reused · {{running}} running · {{queued}} queued · {{completed}} completed",
+  "feedback.human.diffHint": "Diff:    niceeval show {{locator}} --diff",
+  "feedback.human.evalHint": "Eval:    niceeval show {{locator}} --eval",
+  "feedback.human.failuresHeader": "FAILURES",
+  "feedback.human.heartbeat": "{{elapsed}} elapsed · {{counts}}",
+  "feedback.human.inspect": "Inspect: niceeval show {{locator}}",
+  "feedback.human.moreActive": "… {{count}} more active",
+  "feedback.human.plan": "Plan: {{total}} attempts · {{evals}} evals × {{configs}} configs · concurrency {{concurrency}}",
+  "feedback.human.resultFailed": "FAILED",
+  "feedback.human.resultIncomplete": "INCOMPLETE",
+  "feedback.human.resultInterrupted": "INTERRUPTED",
+  "feedback.human.resultPassed": "PASSED",
+  "feedback.human.resultsHeader": "Results:",
+  "feedback.human.resultsMore": "… {{count}} more",
+  "feedback.human.reuse": "Reuse: {{reused}} of {{total}} carried in from cache · {{toRun}} to run",
+  "feedback.human.summaryLine": "{{passed}} passed · {{failed}} failed · {{errored}} errored  ({{reused}} reused)",
+  "feedback.human.summaryAllReusedLine": "{{passed}} passed · {{failed}} failed · {{errored}} errored  (all {{reused}} reused)",
+  "feedback.human.suppressedFailures": "… {{count}} more failures suppressed",
+  "feedback.human.trace": "Trace:   niceeval show {{locator}} --execution",
+  "feedback.phase.agentSetup": "agent setup",
+  "feedback.phase.evalRun": "running eval",
+  "feedback.phase.evalSetup": "eval setup",
+  "feedback.phase.sandboxCreate": "creating sandbox",
+  "feedback.phase.sandboxQueue": "queued for sandbox",
+  "feedback.phase.sandboxSetup": "sandbox setup",
+  "feedback.phase.scoring": "scoring",
+  "feedback.phase.teardown": "cleaning up",
+  "feedback.phase.telemetryCollect": "collecting trace",
+  "feedback.phase.telemetryConfigure": "configuring telemetry",
+  "feedback.phase.workspaceBaseline": "preparing workspace",
+  "feedback.phase.workspaceDiff": "capturing diff",
+  "feedback.rendererError": "  · [feedback] renderer failed while handling {{context}} (ignored): {{message}}\n",
   "hitl.answerNeedsOptionOrText": "The object form of t.respond needs either optionId or text (neither was given).",
   "hitl.invalidOption": "Answer \"{{optionId}}\" is not an option of request {{requestId}} ({{options}}).",
   "hitl.noOptions": "this request has no options",
@@ -189,6 +248,7 @@ export const en = {
   "otel.portInUse": "OTLP receiver port {{port}} is already in use (another process is bound to it). Pick a free port in defineConfig({ telemetry: { port } }), or stop whatever is using {{port}} and retry.",
   "otel.windowAttribution": "otel: spans missing our traceparent, attributing by time window (turns for this agent serialized; concurrency resumes once W3C propagation is confirmed)",
   "runner.diffProgress": "captured diff: {{changed}} changed / {{deleted}} deleted",
+  "runner.failFast": "error {{code}} recurred consecutively on {{evalId}}; treating it as deterministic and skipping the remaining attempts for this config (fail-fast).",
   "runner.driveAgent": "driving agent...",
   "runner.evalSetup": "eval setup (installing dependencies)...",
   "runner.interrupted": "  · interrupted: sandbox containers cleaned up; printing partial results completed so far.\n",
@@ -219,6 +279,8 @@ export const en = {
   "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] force-cleaning {{count}} sandboxes...\n",
+  "sandbox.provisionReconcileFailed": "  · [sandbox] provision reconcile failed, aborting retry (a possibly-created instance could not be verified/killed): {{error}}\n",
+  "sandbox.provisionRetry": "  · [sandbox] provisioning rate-limited, retrying in {{delayMs}}ms (attempt {{attempt}}/{{maxAttempts}})...\n",
   "sandbox.stopFailed": "  · [sandbox] failed to stop sandbox {{id}} (ignored; provider TTL should clean it up): {{message}}\n",
   "sandbox.stopTimeout": "stop timed out ({{timeoutMs}}ms)",
   "scoring.evalError": "evaluation error: {{error}}",

package/src/i18n/zh-CN.ts

@@ -12,6 +12,8 @@ export const zhCN = {
   "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。",
@@ -21,7 +23,21 @@ export const zhCN = {
   "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}}",
   "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}}",
+  "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",
@@ -37,6 +53,7 @@ export const zhCN = {
   "cli.dry.row": "  {{who}}{{experiment}}: {{evals}}  ×{{runs}}\n",
   "cli.error": "niceeval 出错:{{error}}\n",
   "cli.flag.invalidNumber": "标志 --{{flag}} 需要数字,收到 \"{{value}}\"。\n",
+  "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",
   "judge.modelMissing":
@@ -58,7 +75,8 @@ export const zhCN = {
     "      --eval        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" +
-    "      --diff[=文件] sandbox 工作区文件改动摘要;=文件 展开单个文件\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" +
     "      --history   跨 run 时间轴(与 --report 互斥)\n" +
     "      --run <目录> 钉死结果目录   --experiment <id> 只看该实验\n" +
     "      --report <文件> 自定义报告\n" +
@@ -67,11 +85,13 @@ export const zhCN = {
     "      报告槽 + 证据室;--report <文件> 整槽换成自定义报告(与 show 同一文件)\n" +
     "      --run <目录> 钉死结果目录   --experiment <id> 只看该实验\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 clean                           删除 .niceeval/ 历史 artifact\n" +
     "  niceeval init                            脚手架 config + evals/\n\n" +
     "标志:\n" +
     "  --runs n  --max-concurrency n  --timeout ms  --budget usd  --tag t\n" +
-    "  --early-exit / --no-early-exit  --strict  --force  --dry  --quiet\n" +
+    "  --early-exit / --no-early-exit  --strict  --force  --dry  --keep-sandbox[=failed|all]\n" +
+    "  --output auto|human|agent|ci\n" +
     "  --junit path  --json 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。环境变量覆盖(标志 > 环境变量 > config):\n" +
@@ -90,8 +110,11 @@ export const zhCN = {
   "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": "`--agent` / `--model` 不能覆盖 experiment。请在 experiments/ 里新增或复制一个配置文件。\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.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.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",
@@ -123,6 +146,7 @@ export const zhCN = {
   "define.evalIdRejected": "defineEval 不接受 id —— id 由文件路径推导。",
   "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.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。",
@@ -133,6 +157,41 @@ export const zhCN = {
   "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 {{group}}",
+  "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}} --eval",
+  "feedback.human.failuresHeader": "FAILURES",
+  "feedback.human.heartbeat": "已运行 {{elapsed}} · {{counts}}",
+  "feedback.human.inspect": "Inspect: niceeval show {{locator}}",
+  "feedback.human.moreActive": "… 还有 {{count}} 项运行中",
+  "feedback.human.plan": "计划:{{total}} 个 attempt · {{evals}} 个 eval × {{configs}} 个配置 · 并发 {{concurrency}}",
+  "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": "复用:{{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.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.phase.agentSetup": "agent 预置",
+  "feedback.phase.evalRun": "运行 eval",
+  "feedback.phase.evalSetup": "eval 预置",
+  "feedback.phase.sandboxCreate": "创建沙箱",
+  "feedback.phase.sandboxQueue": "排队等沙箱",
+  "feedback.phase.sandboxSetup": "沙箱预置",
+  "feedback.phase.scoring": "评分",
+  "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 的对象形式需要 optionId 或 text 二选一(两者都没给)。",
   "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",
@@ -184,6 +243,7 @@ export const zhCN = {
   "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",
@@ -214,6 +274,8 @@ export const zhCN = {
   "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,靠 provider 过期兜底):{{message}}\n",
   "sandbox.stopTimeout": "stop 超时({{timeoutMs}}ms)",
   "scoring.evalError": "评估出错: {{error}}",

package/src/o11y/cost.test.ts

@@ -1,3 +1,4 @@
+// cases: docs/engineering/unit-tests/adapters/cases.md
 import { describe, expect, it } from "vitest";
 
 import { estimateCost } from "./cost.ts";

package/src/o11y/execution-tree.test.ts

@@ -1,3 +1,4 @@
+// cases: docs/engineering/unit-tests/adapters/cases.md
 // buildExecutionTree 的单测(定稿见 docs/observability.md「OTLP traces → 统一瀑布图」、
 // docs/concepts.md「执行树」词条)。覆盖:无 OTel 时骨架完整、有 OTel 时按 callId 精确合并、
 // 唯一关联不上时降级成 telemetry-only 节点(不猜)、同 callId 撞多条 span 时不強行择一、
@@ -147,18 +148,6 @@ describe("buildExecutionTree", () => {
     expect(Object.keys(tree.nodes[0].span?.attributes ?? {})).toEqual(["call_id"]);
   });
 
-  it("correlates a subagent node the same way as an action node", () => {
-    const events: StreamEvent[] = [
-      { type: "subagent.called", callId: "s1", name: "Task" },
-      { type: "subagent.completed", callId: "s1", output: "done", status: "completed" },
-    ];
-    const agentSpan = span({ spanId: "sp1", attributes: { call_id: "s1" } });
-
-    const tree = buildExecutionTree(events, [agentSpan]);
-
-    expect(tree.nodes[0]).toMatchObject({ kind: "subagent", callId: "s1", span: agentSpan });
-  });
-
   it("never guesses by name/text: a span whose call_id matches no node becomes a telemetry-only node, and the skeleton node's span stays absent", () => {
     const events: StreamEvent[] = [
       { type: "action.called", callId: "c1", name: "Bash", input: {}, tool: "shell" },
@@ -252,14 +241,6 @@ describe("buildExecutionTree", () => {
     expect(action.span?.status).toBe("error");
   });
 
-  it("skill.loaded is a first-class node carrying the skill name (and optional callId), never a generic action", () => {
-    const events: StreamEvent[] = [{ type: "skill.loaded", skill: "pdf-processing", callId: "tool_1" }];
-
-    const tree = buildExecutionTree(events, []);
-
-    expect(tree.nodes).toEqual([{ id: "skill-0", kind: "skill.loaded", skill: "pdf-processing", callId: "tool_1" }]);
-  });
-
   it("a skill.loaded node passes through immediately as its own node — it has no result event to pair with, so it never gets stuck 'pending' the way action/subagent nodes do while awaiting a result", () => {
     const events: StreamEvent[] = [
       { type: "action.called", callId: "c1", name: "Bash", input: {}, tool: "shell" },

package/src/o11y/otlp/mappers/claude-code.test.ts

@@ -1,3 +1,4 @@
+// cases: docs/engineering/unit-tests/adapters/cases.md
 // mapClaudeCodeSpans:span 形状取自真实运行(claude CLI 2.1.202 beta 遥测,E2B 实测)。
 
 import { describe, expect, it } from "vitest";

package/src/o11y/otlp/parse.test.ts

@@ -1,3 +1,4 @@
+// cases: docs/engineering/unit-tests/adapters/cases.md
 // parseOtlpTraces 两种线编码的回归测试。protobuf 路径此前零覆盖——bub(Python OTLP
 // 出口只有 protobuf)的 trace 全靠它,解析静默失败的表现就是"span 收到了但 trace 是空的"。
 // 这里不引 opentelemetry 依赖,用一个最小 protobuf writer 按 opentelemetry-proto 字段号手工编码。

package/src/o11y/otlp/turn-otel.test.ts

@@ -1,3 +1,4 @@
+// cases: docs/engineering/unit-tests/adapters/cases.md
 import { describe, it, expect } from "vitest";
 import type { TraceSpan } from "../../types.ts";
 import type { TraceReceiver } from "./receiver.ts";

package/src/o11y/parsers/bub.test.ts

@@ -1,3 +1,4 @@
+// cases: docs/engineering/unit-tests/adapters/cases.md
 // bub tape 解析回归锁。
 //
 // 背景(2026-07 实测):模型在同一步里「先说话再调工具」时,bub 的 tape 把伴随文本

package/src/o11y/parsers/claude-code.test.ts

@@ -1,3 +1,4 @@
+// cases: docs/engineering/unit-tests/adapters/cases.md
 // claude-code transcript 解析器单测。
 //
 // 重点覆盖 Skill 加载识别(定稿见 docs/observability.md「OTLP traces → 统一瀑布图」、
@@ -40,14 +41,6 @@ describe("parseClaudeCodeTranscript — Skill 加载", () => {
     expect(events).toEqual([{ type: "skill.loaded", skill: "ms-office-suite:pdf", callId: "toolu_02" }]);
   });
 
-  it("工具名大小写不定也能识别(防御性小写比较)", () => {
-    const raw = toolUseLine({ id: "toolu_03", name: "skill", input: { skill: "xlsx" } });
-
-    const { events } = parseClaudeCodeTranscript(raw);
-
-    expect(events).toEqual([{ type: "skill.loaded", skill: "xlsx", callId: "toolu_03" }]);
-  });
-
   it("Skill 加载的 tool_result 被吃掉,不补发孤儿 action.result", () => {
     const raw = [
       toolUseLine({ id: "toolu_01", name: "Skill", input: { skill: "pdf" } }),
@@ -75,20 +68,6 @@ describe("parseClaudeCodeTranscript — Skill 加载", () => {
     ]);
   });
 
-  it("普通(非 Skill)工具调用完全不受影响:Bash 仍产出 action.called + action.result,顺序不变", () => {
-    const raw = [
-      toolUseLine({ id: "toolu_06", name: "Bash", input: { command: "ls" } }),
-      toolResultLine("toolu_06", "a.ts\nb.ts", false),
-    ].join("\n");
-
-    const { events } = parseClaudeCodeTranscript(raw);
-
-    expect(events).toEqual([
-      { type: "action.called", callId: "toolu_06", name: "Bash", input: { command: "ls" }, tool: "shell" },
-      { type: "action.result", callId: "toolu_06", output: "a.ts\nb.ts", status: "completed" },
-    ]);
-  });
-
   it("Skill 加载与普通工具调用混合出现时,各自独立、顺序保持事件出现顺序", () => {
     const raw = [
       toolUseLine({ id: "t1", name: "Skill", input: { skill: "pdf" } }),
@@ -106,18 +85,6 @@ describe("parseClaudeCodeTranscript — Skill 加载", () => {
     ]);
   });
 
-  it("Read/Write/Edit/WebFetch/WebSearch/Task 等既有别名归一不受 Skill 识别改动影响(回归锁)", () => {
-    const raw = [
-      toolUseLine({ id: "a", name: "Write", input: { file_path: "/x", content: "y" } }),
-      toolUseLine({ id: "b", name: "Task", input: { prompt: "go" } }),
-      toolUseLine({ id: "c", name: "WebSearch", input: { query: "q" } }),
-    ].join("\n");
-
-    const { events } = parseClaudeCodeTranscript(raw);
-    const tools = events.filter((e) => e.type === "action.called").map((e) => (e as { tool?: string }).tool);
-
-    expect(tools).toEqual(["file_write", "agent_task", "web_search"]);
-  });
 });
 
 describe("parseClaudeCodeTranscript — 既有行为回归", () => {

package/src/o11y/parsers/openclaw.test.ts

@@ -0,0 +1,154 @@
+// cases: docs/engineering/unit-tests/adapters/cases.md
+import { describe, expect, it } from "vitest";
+
+import { parseOpenClawTranscript, parseOpenClaw, parseOpenClawRunJson } from "./openclaw.ts";
+
+function jsonl(lines: unknown[]): string {
+  return lines.map((l) => JSON.stringify(l)).join("\n");
+}
+
+describe("parseOpenClawTranscript", () => {
+  it("空 / 无内容 → 空事件流,parseSuccess true", () => {
+    expect(parseOpenClawTranscript(undefined)).toEqual({
+      events: [],
+      usage: { inputTokens: 0, outputTokens: 0 },
+      compactions: 0,
+      parseSuccess: true,
+    });
+    expect(parseOpenClawTranscript("  \n ").events).toEqual([]);
+  });
+
+  it("assistant 的 text/thinking/toolCall parts + toolResult 按 call ID 配对,usage 逐消息累加", () => {
+    const raw = jsonl([
+      { role: "user", content: "改一下配置" },
+      {
+        role: "assistant",
+        content: [
+          { type: "thinking", thinking: "先看文件" },
+          { type: "text", text: "我来改" },
+          { type: "toolCall", id: "tc_1", name: "read", arguments: { path: "a.ts" } },
+        ],
+        usage: { input: 100, output: 20, cacheRead: 5, cacheWrite: 3, cost: { total: 0.01 } },
+      },
+      { role: "toolResult", toolCallId: "tc_1", content: [{ type: "text", text: "file body" }], isError: false },
+      {
+        role: "assistant",
+        content: [{ type: "text", text: "改好了" }],
+        usage: { input: 50, output: 10, cacheRead: 0, cacheWrite: 0, cost: { total: 0.002 } },
+      },
+    ]);
+
+    const parsed = parseOpenClawTranscript(raw);
+    expect(parsed.events).toEqual([
+      { type: "message", role: "user", text: "改一下配置" },
+      { type: "thinking", text: "先看文件" },
+      { type: "message", role: "assistant", text: "我来改" },
+      { type: "action.called", callId: "tc_1", name: "read", input: { path: "a.ts" }, tool: "file_read" },
+      { type: "action.result", callId: "tc_1", output: "file body", status: "completed" },
+      { type: "message", role: "assistant", text: "改好了" },
+    ]);
+    expect(parsed.usage).toEqual({
+      inputTokens: 150,
+      outputTokens: 30,
+      cacheReadTokens: 5,
+      cacheWriteTokens: 3,
+      requests: 2,
+      costUSD: 0.012,
+    });
+    expect(parsed.parseSuccess).toBe(true);
+  });
+
+  it("{ type: 'message', message: {...} } 包装条目与 snake_case 变体也认", () => {
+    const raw = jsonl([
+      {
+        type: "message",
+        message: {
+          role: "assistant",
+          content: [{ type: "tool_call", tool_call_id: "c9", toolName: "exec", input: { cmd: "ls" } }],
+        },
+      },
+      { type: "message", message: { role: "tool", tool_call_id: "c9", output: "a.txt", is_error: true } },
+    ]);
+    expect(parseOpenClaw(raw)).toEqual([
+      { type: "action.called", callId: "c9", name: "exec", input: { cmd: "ls" }, tool: "shell" },
+      { type: "action.result", callId: "c9", output: "a.txt", status: "failed" },
+    ]);
+  });
+
+  it("compaction 条目计数,error 条目落 error 事件", () => {
+    const parsed = parseOpenClawTranscript(
+      jsonl([
+        { type: "compaction", reason: "context-limit" },
+        { type: "error", message: "boom" },
+      ]),
+    );
+    expect(parsed.events).toEqual([
+      { type: "compaction", reason: "context-limit" },
+      { type: "error", message: "boom" },
+    ]);
+    expect(parsed.compactions).toBe(1);
+  });
+
+  it("坏 JSON 行不中断解析,标 parseSuccess: false 但保留其余行的事件", () => {
+    const raw = `{"role":"assistant","content":[{"type":"text","text":"hi"}]}\nnot-json\n`;
+    const parsed = parseOpenClawTranscript(raw);
+    expect(parsed.parseSuccess).toBe(false);
+    expect(parsed.events).toEqual([{ type: "message", role: "assistant", text: "hi" }]);
+  });
+
+  it("字符串形态的 arguments 先按 JSON 解;解不开原样保留", () => {
+    const raw = jsonl([
+      {
+        role: "assistant",
+        content: [{ type: "toolCall", id: "c1", name: "write", arguments: '{"path":"b.ts"}' }],
+      },
+    ]);
+    expect(parseOpenClaw(raw)).toEqual([
+      { type: "action.called", callId: "c1", name: "write", input: { path: "b.ts" }, tool: "file_write" },
+    ]);
+  });
+});
+
+describe("parseOpenClawRunJson", () => {
+  it("整段 pretty-print JSON 封包:text / sessionId / usage / failed", () => {
+    const stdout = JSON.stringify(
+      {
+        sessionId: "sess-42",
+        result: { text: "done" },
+        usage: { input_tokens: 10, output_tokens: 3 },
+        status: "ok",
+      },
+      null,
+      2,
+    );
+    expect(parseOpenClawRunJson(stdout)).toEqual({
+      text: "done",
+      sessionId: "sess-42",
+      usage: { inputTokens: 10, outputTokens: 3 },
+      failed: false,
+    });
+  });
+
+  it("混日志行的 stdout:取最后一个完整 JSON 对象;payloads[].text 拼接", () => {
+    const stdout = [
+      "starting agent...",
+      JSON.stringify({ payloads: [{ text: "第一段" }, { text: "第二段" }], sessionKey: "k1" }),
+    ].join("\n");
+    expect(parseOpenClawRunJson(stdout)).toEqual({
+      text: "第一段\n第二段",
+      sessionId: "k1",
+      failed: false,
+    });
+  });
+
+  it("error 字段 / status=error / success=false → failed", () => {
+    expect(parseOpenClawRunJson(JSON.stringify({ error: "rate limited" })).failed).toBe(true);
+    expect(parseOpenClawRunJson(JSON.stringify({ status: "error" })).failed).toBe(true);
+    expect(parseOpenClawRunJson(JSON.stringify({ success: false })).failed).toBe(true);
+  });
+
+  it("解析不出 JSON → 空摘要,failed 交给调用方 exitCode", () => {
+    expect(parseOpenClawRunJson("plain text output")).toEqual({ failed: false });
+    expect(parseOpenClawRunJson(undefined)).toEqual({ failed: false });
+  });
+});