reeval 0.3.0 → 0.5.0

package/dist/bridge/package.json

@@ -4,8 +4,8 @@
   "private": true,
   "type": "module",
   "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "0.3.276",
-    "@anthropic-ai/claude-code": "2.1.276",
+    "@anthropic-ai/claude-agent-sdk": "0.3.281",
+    "@anthropic-ai/claude-code": "2.1.281",
     "@modelcontextprotocol/sdk": "1.30.0",
     "ws": "8.21.0",
     "zod": "4.4.3"

package/dist/bridge/pnpm-lock.yaml

@@ -9,11 +9,11 @@ importers:
   .:
     dependencies:
       '@anthropic-ai/claude-agent-sdk':
-        specifier: 0.3.276
-        version: 0.3.276(@anthropic-ai/sdk@0.97.1(zod@4.4.3))(@modelcontextprotocol/sdk@1.30.0(zod@4.4.3))(zod@4.4.3)
+        specifier: 0.3.281
+        version: 0.3.281(@anthropic-ai/sdk@0.97.1(zod@4.4.3))(@modelcontextprotocol/sdk@1.30.0(zod@4.4.3))(zod@4.4.3)
       '@anthropic-ai/claude-code':
-        specifier: 2.1.276
-        version: 2.1.276
+        specifier: 2.1.281
+        version: 2.1.281
       '@modelcontextprotocol/sdk':
         specifier: 1.30.0
         version: 1.30.0(zod@4.4.3)
@@ -26,104 +26,104 @@ importers:
 
 packages:
 
-  '@anthropic-ai/claude-agent-sdk-darwin-arm64@0.3.276':
-    resolution: {integrity: sha512-WdpZ7UXoIDGYeb97OWRFADqkMDLv/bB3YrRxvomqdxfRZC2RLAPLUxsBat9zVL9B5dsiJmZN8h1D0NlMe92Uzw==}
+  '@anthropic-ai/claude-agent-sdk-darwin-arm64@0.3.281':
+    resolution: {integrity: sha512-wydRJU0CX7c/w1qEsgqGUJFzmlOMTvUpBkWHsj1pJFbrRFQUnIW0EWor+pDd4WiKWWihVqUhYEA+3LTmUiLZIQ==}
     cpu: [arm64]
     os: [darwin]
 
-  '@anthropic-ai/claude-agent-sdk-darwin-x64@0.3.276':
-    resolution: {integrity: sha512-y5J+K1q7llwYkf85OhjG71R5AqsulHzC+HpgUOrg6klO+jBKsesGGS7tsEGzsM4YJtDntr6Rylw/sH3ZmXvjhg==}
+  '@anthropic-ai/claude-agent-sdk-darwin-x64@0.3.281':
+    resolution: {integrity: sha512-yvtdCwCimbI7dcFcQ204gEX4quzZBcyW2orCh8ESlkZY/Mn480qKCjyKWGuR17vGKGePynoiS1W895rxWNfHZQ==}
     cpu: [x64]
     os: [darwin]
 
-  '@anthropic-ai/claude-agent-sdk-linux-arm64-musl@0.3.276':
-    resolution: {integrity: sha512-Pb7PVWLktFCgD1cUtF61lx9riT2bh3ZSE6rq94EjaesyCnpzddlQ59/LLwLVUwGunBq9N2WayXaElOsk8+pK/g==}
+  '@anthropic-ai/claude-agent-sdk-linux-arm64-musl@0.3.281':
+    resolution: {integrity: sha512-B5l5tnO0ymQjGSq+uZBnwLb8IdCj+gr/niKoxiUDqzfh7OPwuGyceq/g7bpzZ6VC/U7+Zsjv+Gce/VzcKVT1pA==}
     cpu: [arm64]
     os: [linux]
     libc: [musl]
 
-  '@anthropic-ai/claude-agent-sdk-linux-arm64@0.3.276':
-    resolution: {integrity: sha512-IaBni1kRiUKEsSRTHocVW8jqzl4KyC6ms6fr6KAJPPcjJTu2Ptpb1ohj/OfDixTc/AErrNVL0ye8ICpZ+88k7w==}
+  '@anthropic-ai/claude-agent-sdk-linux-arm64@0.3.281':
+    resolution: {integrity: sha512-PrxuObUGayjtUS6yq5Sz+dMNXn8KaK1BeIXG/PRE+85Wt9DzchoIduycIoFaXWrG0xbbzYz0AKwttNqSAeDdSA==}
     cpu: [arm64]
     os: [linux]
     libc: [glibc]
 
-  '@anthropic-ai/claude-agent-sdk-linux-x64-musl@0.3.276':
-    resolution: {integrity: sha512-OaZAlVBhAUFkivWgb5ZsJYCT08qTo9LyomaN/2jNwsvVkpurr8kXEGncsTk8sM5JyIzSSiWuR+Sfxphq071XPA==}
+  '@anthropic-ai/claude-agent-sdk-linux-x64-musl@0.3.281':
+    resolution: {integrity: sha512-H+TWmbIhlEvZcOGJ6S9BQBBQuizaXfDhdSjgWbP6JGTrfW+U8dm49/0I9pdjHB2VT+1zgFzIt1YkkfkvSGtOVQ==}
     cpu: [x64]
     os: [linux]
     libc: [musl]
 
-  '@anthropic-ai/claude-agent-sdk-linux-x64@0.3.276':
-    resolution: {integrity: sha512-MR/HQn8xi3RGujnzgJhyVolKR7JO+BiPhBBJ0GqbIlM4huso1aIDFhYhetdGIPb4shr7R1OZzSj1QFJoJQQ8/A==}
+  '@anthropic-ai/claude-agent-sdk-linux-x64@0.3.281':
+    resolution: {integrity: sha512-3kxVGQrn9wNeFZV0MJEXI0XOj7U8e/1wL41W2MInYVIVWJnTanuX+pPqQH2DzyVNRcw6CMP5k3eockewGptWig==}
     cpu: [x64]
     os: [linux]
     libc: [glibc]
 
-  '@anthropic-ai/claude-agent-sdk-win32-arm64@0.3.276':
-    resolution: {integrity: sha512-dk7lxCdvRwXVf5LAh9nQ2/MLLedfT3K1HIQsv5KBgexUE1F7JYYd5/R3eYs/qT9UQOeKrMmKq7GG9RxDDXCgjA==}
+  '@anthropic-ai/claude-agent-sdk-win32-arm64@0.3.281':
+    resolution: {integrity: sha512-ySM49+TS94VrMkIzk65RHcNKfvXKhx6YuR2mplw8BbaYpcnKL4RTAr5WVbH1JhMLjPEBaDGSP1uCLZy/YkA9gg==}
     cpu: [arm64]
     os: [win32]
 
-  '@anthropic-ai/claude-agent-sdk-win32-x64@0.3.276':
-    resolution: {integrity: sha512-CKPz5K94qpVHdP0hmSugntpiTshHrPFs4RtQYcFvP4PVltZv/wtth090Rff21XJna3Lxt4Lb5ce3tMqTnDRu9Q==}
+  '@anthropic-ai/claude-agent-sdk-win32-x64@0.3.281':
+    resolution: {integrity: sha512-hbnymWj9O31K0VMd/VOpMTZ0CNT3BkwgGGe84zHdAP6MDxmoZvqyS94UEaLQit3xmozCxfkfGns7/WgWH1Q6Sg==}
     cpu: [x64]
     os: [win32]
 
-  '@anthropic-ai/claude-agent-sdk@0.3.276':
-    resolution: {integrity: sha512-Dic43v4uuGLhibPAArWy3RSfq3zHdn+4Oh2AI2/18a5tlbNKkvChEpazyp8c+grUJvzxkzt0jETCa1wrADbkyw==}
+  '@anthropic-ai/claude-agent-sdk@0.3.281':
+    resolution: {integrity: sha512-JTbluTKmY3wKQPIXzeCTvz/K0FElTuQU5Y7ueLgUOBrHcks8cHlGugEgIRezffhc13WzBAqYtaY15tJ+Bfunzg==}
     engines: {node: '>=18.0.0'}
     peerDependencies:
       '@anthropic-ai/sdk': '>=0.93.0'
       '@modelcontextprotocol/sdk': ^1.29.0
       zod: ^4.0.0
 
-  '@anthropic-ai/claude-code-darwin-arm64@2.1.276':
-    resolution: {integrity: sha512-JchRlXfYFrHGjm7fTZpWzlDvaP67SSO2yx2yFQSxt/T57mC6ickpApiuvS5jHCia6qjwzgjuzWSkemfLYaxB7w==}
+  '@anthropic-ai/claude-code-darwin-arm64@2.1.281':
+    resolution: {integrity: sha512-rEI/YGBDX4YTfdq5w1B86NicoLgpFHGp4IrKM6sDrmUemruePSXF7ybICthHClIsRY8a/Kp7PQ6UJah1VWTbSA==}
     cpu: [arm64]
     os: [darwin]
 
-  '@anthropic-ai/claude-code-darwin-x64@2.1.276':
-    resolution: {integrity: sha512-7LBOcz1A5cqbQ9z363R5YjnsNvZCDAztV8/zOXM3i0PM7nHDrG3blYoZPDal54uPiovGCPVsa+bEuj+kbSZPYA==}
+  '@anthropic-ai/claude-code-darwin-x64@2.1.281':
+    resolution: {integrity: sha512-nGJBmWAMlyHAlf0i/vwBzjvTfEC86KcGZT4VNlpInL03jufnBztZr8+ILcXcwg5h94McpmvI/tu8lgHJtHKLVQ==}
     cpu: [x64]
     os: [darwin]
 
-  '@anthropic-ai/claude-code-linux-arm64-musl@2.1.276':
-    resolution: {integrity: sha512-eKm/iIwLNdc8VtY00ETzDaLQN8tt1u7cf38b+Y/tCCfZWz8vT5EnQkJatpkm94Yq2m5iRXjcILmN2hHWZ53s0Q==}
+  '@anthropic-ai/claude-code-linux-arm64-musl@2.1.281':
+    resolution: {integrity: sha512-prPxIRpyi75pnUOVugK4hSW6i1FONp9XMUuVSOmt/OSEZuurVqySj8UBjGM8RdDwZGuaK4F0dd3h8iylZIuE0g==}
     cpu: [arm64]
     os: [linux]
     libc: [musl]
 
-  '@anthropic-ai/claude-code-linux-arm64@2.1.276':
-    resolution: {integrity: sha512-/wzWD8U5mGmSoeiB99gGfvVuj/7/p7nh/zZACTuuPEwvo/CfMMvW/ZxBB6LG8w0nz6B//UOIx3ryKdGCENAPOA==}
+  '@anthropic-ai/claude-code-linux-arm64@2.1.281':
+    resolution: {integrity: sha512-xY37iXJ+v6izc2yArEUD8T3d7S45GIztzsSW+fWv8KB63buuO1qhN6Eqzq2yR3RqWBMjwiRL/BbkLB8FZWBsXw==}
     cpu: [arm64]
     os: [linux]
     libc: [glibc]
 
-  '@anthropic-ai/claude-code-linux-x64-musl@2.1.276':
-    resolution: {integrity: sha512-1y66zgPpIWV14fn75veSXCCNpJB92+dlM2jSt5hvlB15LH6wk2JHTD8H7ogRCYPrMtBOkdz3yfH/fxSeD4UP1w==}
+  '@anthropic-ai/claude-code-linux-x64-musl@2.1.281':
+    resolution: {integrity: sha512-4eiBWD03AyGuuvMN7xyMFKhrToSaVi79Vi0R+K79gS7+XeSn2O8Oc7U44I3gX9ziqHsN82Puk2Ju2QAhLxKoYQ==}
     cpu: [x64]
     os: [linux]
     libc: [musl]
 
-  '@anthropic-ai/claude-code-linux-x64@2.1.276':
-    resolution: {integrity: sha512-xFhD9UyIVKWLEQjcFg1ynSmYPpiXSdyN9+9rtrM0aIVnSqFP15Hd5/owEf7N6JrHtAXijfxDC0usXFxAOJQ1vA==}
+  '@anthropic-ai/claude-code-linux-x64@2.1.281':
+    resolution: {integrity: sha512-Z7ld7AE2AtvghO0qlNFBGD8S2w+nPVUk8uq4UprAUKJr6HHavXm605iGhbzSOTgXu3F/y540JHVUW9ON87w4IQ==}
     cpu: [x64]
     os: [linux]
     libc: [glibc]
 
-  '@anthropic-ai/claude-code-win32-arm64@2.1.276':
-    resolution: {integrity: sha512-QGbJPUb6uhngwPCKociJUArJfqSV6gW55qcZLiu7RJQIUK4LHApntAIO6NdJ/du4YiuHsty/5Hy2zAzekXk+mQ==}
+  '@anthropic-ai/claude-code-win32-arm64@2.1.281':
+    resolution: {integrity: sha512-mPY+B6CSUVIFPvJld+BEaMYHE1/qdReoy73+gyXitI+rTmTUvnL5Aa21eXbox714naQZct8Ym2vgQS/yLQdq0Q==}
     cpu: [arm64]
     os: [win32]
 
-  '@anthropic-ai/claude-code-win32-x64@2.1.276':
-    resolution: {integrity: sha512-5LtGQOG70zw7svIs7rRN8hutS8vXmn6FrXZk5hmt1Tsr18ycbesxD1YgJydgyzy99HRqOTkFMXQCWZOUsY8xpg==}
+  '@anthropic-ai/claude-code-win32-x64@2.1.281':
+    resolution: {integrity: sha512-IP2nJlr02d9BRQCTf2ErZD78pemjcQ3h0tg16AY+e8i5+42FGZELUdB1wzf4066QJQvpU7BgFpSt0KiHq3SQZA==}
     cpu: [x64]
     os: [win32]
 
-  '@anthropic-ai/claude-code@2.1.276':
-    resolution: {integrity: sha512-xgVXCbFdqOhSAcMTSC61gjL1jCQuoAA4TNU1I7Dmf/yuYcDxHf796r6LaB3Jswm1floKnP1V+/zfphtDlPdlkg==}
+  '@anthropic-ai/claude-code@2.1.281':
+    resolution: {integrity: sha512-sldhFsGqOnGA8pfOMLjyc+Sc331VkZ1TsP1sdpvMeVFYBikC5x53miTCsnTiE6Bio74IYPn7BIzTNbHcRp5fXQ==}
     engines: {node: '>=22.0.0'}
     hasBin: true
 
@@ -540,79 +540,79 @@ packages:
 
 snapshots:
 
-  '@anthropic-ai/claude-agent-sdk-darwin-arm64@0.3.276':
+  '@anthropic-ai/claude-agent-sdk-darwin-arm64@0.3.281':
     optional: true
 
-  '@anthropic-ai/claude-agent-sdk-darwin-x64@0.3.276':
+  '@anthropic-ai/claude-agent-sdk-darwin-x64@0.3.281':
     optional: true
 
-  '@anthropic-ai/claude-agent-sdk-linux-arm64-musl@0.3.276':
+  '@anthropic-ai/claude-agent-sdk-linux-arm64-musl@0.3.281':
     optional: true
 
-  '@anthropic-ai/claude-agent-sdk-linux-arm64@0.3.276':
+  '@anthropic-ai/claude-agent-sdk-linux-arm64@0.3.281':
     optional: true
 
-  '@anthropic-ai/claude-agent-sdk-linux-x64-musl@0.3.276':
+  '@anthropic-ai/claude-agent-sdk-linux-x64-musl@0.3.281':
     optional: true
 
-  '@anthropic-ai/claude-agent-sdk-linux-x64@0.3.276':
+  '@anthropic-ai/claude-agent-sdk-linux-x64@0.3.281':
     optional: true
 
-  '@anthropic-ai/claude-agent-sdk-win32-arm64@0.3.276':
+  '@anthropic-ai/claude-agent-sdk-win32-arm64@0.3.281':
     optional: true
 
-  '@anthropic-ai/claude-agent-sdk-win32-x64@0.3.276':
+  '@anthropic-ai/claude-agent-sdk-win32-x64@0.3.281':
     optional: true
 
-  '@anthropic-ai/claude-agent-sdk@0.3.276(@anthropic-ai/sdk@0.97.1(zod@4.4.3))(@modelcontextprotocol/sdk@1.30.0(zod@4.4.3))(zod@4.4.3)':
+  '@anthropic-ai/claude-agent-sdk@0.3.281(@anthropic-ai/sdk@0.97.1(zod@4.4.3))(@modelcontextprotocol/sdk@1.30.0(zod@4.4.3))(zod@4.4.3)':
     dependencies:
       '@anthropic-ai/sdk': 0.97.1(zod@4.4.3)
       '@modelcontextprotocol/sdk': 1.30.0(zod@4.4.3)
       zod: 4.4.3
     optionalDependencies:
-      '@anthropic-ai/claude-agent-sdk-darwin-arm64': 0.3.276
-      '@anthropic-ai/claude-agent-sdk-darwin-x64': 0.3.276
-      '@anthropic-ai/claude-agent-sdk-linux-arm64': 0.3.276
-      '@anthropic-ai/claude-agent-sdk-linux-arm64-musl': 0.3.276
-      '@anthropic-ai/claude-agent-sdk-linux-x64': 0.3.276
-      '@anthropic-ai/claude-agent-sdk-linux-x64-musl': 0.3.276
-      '@anthropic-ai/claude-agent-sdk-win32-arm64': 0.3.276
-      '@anthropic-ai/claude-agent-sdk-win32-x64': 0.3.276
-
-  '@anthropic-ai/claude-code-darwin-arm64@2.1.276':
+      '@anthropic-ai/claude-agent-sdk-darwin-arm64': 0.3.281
+      '@anthropic-ai/claude-agent-sdk-darwin-x64': 0.3.281
+      '@anthropic-ai/claude-agent-sdk-linux-arm64': 0.3.281
+      '@anthropic-ai/claude-agent-sdk-linux-arm64-musl': 0.3.281
+      '@anthropic-ai/claude-agent-sdk-linux-x64': 0.3.281
+      '@anthropic-ai/claude-agent-sdk-linux-x64-musl': 0.3.281
+      '@anthropic-ai/claude-agent-sdk-win32-arm64': 0.3.281
+      '@anthropic-ai/claude-agent-sdk-win32-x64': 0.3.281
+
+  '@anthropic-ai/claude-code-darwin-arm64@2.1.281':
     optional: true
 
-  '@anthropic-ai/claude-code-darwin-x64@2.1.276':
+  '@anthropic-ai/claude-code-darwin-x64@2.1.281':
     optional: true
 
-  '@anthropic-ai/claude-code-linux-arm64-musl@2.1.276':
+  '@anthropic-ai/claude-code-linux-arm64-musl@2.1.281':
     optional: true
 
-  '@anthropic-ai/claude-code-linux-arm64@2.1.276':
+  '@anthropic-ai/claude-code-linux-arm64@2.1.281':
     optional: true
 
-  '@anthropic-ai/claude-code-linux-x64-musl@2.1.276':
+  '@anthropic-ai/claude-code-linux-x64-musl@2.1.281':
     optional: true
 
-  '@anthropic-ai/claude-code-linux-x64@2.1.276':
+  '@anthropic-ai/claude-code-linux-x64@2.1.281':
     optional: true
 
-  '@anthropic-ai/claude-code-win32-arm64@2.1.276':
+  '@anthropic-ai/claude-code-win32-arm64@2.1.281':
     optional: true
 
-  '@anthropic-ai/claude-code-win32-x64@2.1.276':
+  '@anthropic-ai/claude-code-win32-x64@2.1.281':
     optional: true
 
-  '@anthropic-ai/claude-code@2.1.276':
+  '@anthropic-ai/claude-code@2.1.281':
     optionalDependencies:
-      '@anthropic-ai/claude-code-darwin-arm64': 2.1.276
-      '@anthropic-ai/claude-code-darwin-x64': 2.1.276
-      '@anthropic-ai/claude-code-linux-arm64': 2.1.276
-      '@anthropic-ai/claude-code-linux-arm64-musl': 2.1.276
-      '@anthropic-ai/claude-code-linux-x64': 2.1.276
-      '@anthropic-ai/claude-code-linux-x64-musl': 2.1.276
-      '@anthropic-ai/claude-code-win32-arm64': 2.1.276
-      '@anthropic-ai/claude-code-win32-x64': 2.1.276
+      '@anthropic-ai/claude-code-darwin-arm64': 2.1.281
+      '@anthropic-ai/claude-code-darwin-x64': 2.1.281
+      '@anthropic-ai/claude-code-linux-arm64': 2.1.281
+      '@anthropic-ai/claude-code-linux-arm64-musl': 2.1.281
+      '@anthropic-ai/claude-code-linux-x64': 2.1.281
+      '@anthropic-ai/claude-code-linux-x64-musl': 2.1.281
+      '@anthropic-ai/claude-code-win32-arm64': 2.1.281
+      '@anthropic-ai/claude-code-win32-x64': 2.1.281
 
   '@anthropic-ai/sdk@0.97.1(zod@4.4.3)':
     dependencies:

package/dist/bridge/pnpm-workspace.yaml

@@ -1,2 +1,2 @@
 allowBuilds:
-  '@anthropic-ai/claude-code@2.1.276': true
+  '@anthropic-ai/claude-code@2.1.281': true

package/dist/{cli-DemDdg8J.js → cli-Ben0kbJt.js}

@@ -1,25 +1,25 @@
-import { t as UsageError } from "./errors-CSxfJ4qu.js";
-import { b as parseDuration, v as workspaceLabel } from "./environment-ByCF9s3n.js";
+import { n as UsageError } from "./errors-yDAEwmZo.js";
+import { D as parseDuration, T as workspaceLabel, u as environmentSandbox } from "./environment-D3sZUybY.js";
 import { n as loadEnvFiles, t as inCI } from "./env-BQqj93gu.js";
-import { a as registerSecret, i as redactSecrets } from "./redact-C4fL_bZP.js";
+import { o as redactSecrets, s as registerSecret } from "./redact-DJ9JCShu.js";
 import { o as PACKAGE_VERSION } from "./root-Zeokqt3-.js";
 import { a as fromCwd, i as evalSlug, n as attemptKey, o as newRunId } from "./ids-Bp6ut7ZD.js";
-import { I as LABEL, X as imageInfo } from "./src-D3eW1e7Y.js";
-import { d as sandboxId, s as harnessId } from "./options-CIutEeuy.js";
-import { c as hostCapacity, d as missingCredentials, f as resolveConcurrency, i as configPathFor, l as loadConfig, p as selectProfile, r as configLabel, s as findConfig, u as malformedCredentials } from "./config-CqSLZkwx.js";
+import { I as LABEL, X as imageInfo } from "./src-CQXA5zkO.js";
+import { d as sandboxId, s as harnessId } from "./options-BBLwgaSz.js";
+import { c as hostCapacity, d as missingCredentials, f as resolveConcurrency, i as configPathFor, l as loadConfig, p as selectProfile, r as configLabel, s as findConfig, u as malformedCredentials } from "./config-kbc7kZz3.js";
 import { s as debugTime } from "./debug-D24NOQH1.js";
 import { i as sandboxWarningLine, n as clip, r as formatDuration } from "./text-gEcKjAyp.js";
 import { t as PrepareError } from "./provider-CmAYTQQs.js";
-import { I as effortLabel, S as runOutcome, g as hasUngatedCells, j as cellContrast, m as formatUsage, y as isUngatedMiss } from "./model-ORtLm2UB.js";
+import { A as cellJudgeRates, F as judgeRatesLabel, M as collapsedJudgeLabel, R as cellContrast, S as runOutcome, U as effortLabel, g as hasUngatedCells, j as cellUsage, m as formatUsage, y as isUngatedMiss } from "./model-DGafiJdy.js";
 import { n as LIVE_FILE, o as liveCounts } from "./live-DZkCtmNP.js";
-import { a as liftLabel, i as gateLabel, n as answersLabel } from "./summary-gawvEpNu.js";
-import { n as FLAGS, o as helpText, r as GLOBAL, t as COMMANDS } from "./cli-spec-BxLPgNN-.js";
-import { i as runtimeInfo, n as gitInfo, r as runnerInfo, t as author } from "./provenance-D4vo7RF6.js";
+import { a as liftLabel, i as gateLabel, n as answersLabel } from "./summary-78yLjokt.js";
+import { n as FLAGS, o as helpText, r as GLOBAL, t as COMMANDS } from "./cli-spec-DUeyuSIQ.js";
+import { i as runtimeInfo, n as gitInfo, r as runnerInfo, t as author } from "./provenance-rP7Unzb4.js";
 import { t as budgetMarker } from "./budget-DBUrH4jX.js";
-import { n as prepareAll, o as writePrepared, r as prepareTargets, t as RunStopError } from "./preflight-vISQIiF1.js";
-import { r as imageHarness } from "./prepare-BTeAiW9Q.js";
-import { n as createProviders } from "./providers-Bjgky1u8.js";
-import { a as redactArgv, r as commandOf, t as logInvocation } from "./invocations-DSMfs3Dx.js";
+import { n as prepareAll, o as writePrepared, r as prepareTargets, t as RunStopError } from "./preflight-Cl__EL9Y.js";
+import { r as imageHarness } from "./prepare-jmZNuhe5.js";
+import { n as createProviders } from "./providers-Cs0GAhfO.js";
+import { a as redactArgv, r as commandOf, t as logInvocation } from "./invocations-DHC946b-.js";
 import { n as stripAnsi, t as colors } from "./colors-QV9dFVhN.js";
 import { existsSync, mkdirSync, readdirSync, renameSync, rmSync, writeFileSync } from "node:fs";
 import { hostname } from "node:os";
@@ -515,10 +515,12 @@ function evalColumns(c, e) {
 	}
 	if (attemptOutcome(attempt) === "cancelled") return c.yellow("cancelled");
 	const cols = [formatDuration(attempt.durationMs).padStart(7)];
-	const scores = attempt.judgements.map((j) => j.score);
-	if (scores.length) cols.push(`judge ${Math.min(...scores).toFixed(2)}`);
+	const judges = collapsedJudgeLabel(attempt.judgements);
+	if (judges) cols.push(judges);
 	const turns = attempt.turns.length;
 	cols.push(`${turns} turn${turns === 1 ? "" : "s"}`);
+	const tokens = formatUsage(attempt.usage);
+	if (tokens) cols.push(tokens);
 	return c.dim(cols.join("  "));
 }
 /** Pass rate per cell for evals with several attempts; with compare, each cell's role and lift. */
@@ -538,7 +540,11 @@ function cellRows(c, e, indent) {
 			note = `   ${below ? c.red(`${text} ✗`) : text}`;
 		} else if (compare && cell.gate === "none" && cell.fail > 0) note = `   ${c.dim("not gated")}`;
 		if (cell.gate !== "all" && cell.gate !== "none") note += `   ${c.dim(`gate ${gateLabel(cell.gate)}`)}`;
-		const rows = [`${indent}${c.dim(cell.key.padEnd(width))}${role}  ${counts}  ${pct(cell.passRate)}${note}${infra}`];
+		const tokens = formatUsage(cellUsage(e, cell.key));
+		const usage = tokens ? `   ${c.dim(tokens)}` : "";
+		const rows = [`${indent}${c.dim(cell.key.padEnd(width))}${role}  ${counts}  ${pct(cell.passRate)}${note}${infra}${usage}`];
+		const rates = judgeRatesLabel(cellJudgeRates(e, cell.key));
+		if (rates) rows.push(`${indent}  ${c.dim(rates)}`);
 		const effort = cellContrast(e, cell.key);
 		if (effort) rows.push(`${indent}  ${c.dim(`effort: ${effortLabel(effort)}`)}`);
 		if (cell.overrides.length) rows.push(`${indent}  ${c.dim(`overrides: ${cell.overrides.join(", ")}`)}`);
@@ -1188,7 +1194,7 @@ async function run(argv) {
 		return values.help ? EXIT.ok : EXIT.usage;
 	}
 	if (command === "init") {
-		const { initCommand } = await import("./init-DPJv7OmX.js");
+		const { initCommand } = await import("./init-C5Kp2CVo.js");
 		return initCommand(values, filters, {
 			out: (s) => console.log(s),
 			input: process.stdin,
@@ -1197,7 +1203,7 @@ async function run(argv) {
 		});
 	}
 	if (command === "doctor") {
-		const { doctorCommand } = await import("./doctor-DhvVbNKv.js");
+		const { doctorCommand } = await import("./doctor-X2ba-e1Y.js");
 		return doctorCommand(values, {
 			out: (s) => console.log(s),
 			color: !!process.stdout.isTTY && !process.env.NO_COLOR,
@@ -1205,7 +1211,7 @@ async function run(argv) {
 		});
 	}
 	if (command === "report") {
-		const { reportCommand } = await import("./command-CxwnRrTs.js");
+		const { reportCommand } = await import("./command-DpR1iz_b.js");
 		return reportCommand(values, filters, {
 			out: (s) => console.log(s),
 			err: (s) => console.error(dim(s)),
@@ -1220,7 +1226,7 @@ async function run(argv) {
 	}
 	if (command === "studio") {
 		const config = await loadConfig(configPath, "studio");
-		const { studioCommand } = await import("./command-pHfluVCF.js");
+		const { studioCommand } = await import("./command-CmFpqrTW.js");
 		return studioCommand({
 			config,
 			flags: values,
@@ -1231,7 +1237,7 @@ async function run(argv) {
 	if (command === "mcp") {
 		if (!values.http) console.log = console.info = console.debug = console.error;
 		const config = await loadConfig(configPath, "mcp");
-		const { mcpCommand } = await import("./command-PEbIzJj3.js");
+		const { mcpCommand } = await import("./command-zD7IIwQF.js");
 		return mcpCommand({
 			config,
 			flags: values,
@@ -1240,7 +1246,7 @@ async function run(argv) {
 	}
 	if (command in HISTORY_COMMANDS) {
 		const config = await loadConfig(configPath, command);
-		const commands = await import("./commands-De-SxIgY.js");
+		const commands = await import("./commands-8xIhRt-7.js");
 		const ctx = {
 			config,
 			flags: values,
@@ -1253,7 +1259,7 @@ async function run(argv) {
 			case "compare": return commands.compareCommand(ctx, filters);
 			case "checkout": return commands.checkoutCommand(ctx, filters);
 			case "query": {
-				const { queryCommand } = await import("./query-CWkZnV62.js");
+				const { queryCommand } = await import("./query-BjqPXDPV.js");
 				return queryCommand({
 					...ctx,
 					err: (s) => console.error(dim(s))
@@ -1261,7 +1267,7 @@ async function run(argv) {
 			}
 			case "analyze": {
 				loadEnvFiles(configPath);
-				const { analyzeCommand } = await import("./command-DhgZPhsY.js");
+				const { analyzeCommand } = await import("./command-pIntI3Iq.js");
 				return analyzeCommand({
 					...ctx,
 					err: (s) => console.error(dim(s))
@@ -1300,8 +1306,8 @@ async function cleanCommand(flags, configPath) {
 	const where = fromCwd(debugRoot(config.root)) || ".";
 	console.log(runs.length ? `Removed debug data for ${runs.length} run${runs.length === 1 ? "" : "s"} (${formatBytes(bytes)}) from ${where}` : `No debug data to remove in ${where}`);
 	if (flags.images) {
-		const { pruneImages } = await import("./src-C91RndLw.js");
-		const { requireDocker } = await import("./providers-DfVMglqP.js");
+		const { pruneImages } = await import("./src-CQzUdIZj.js");
+		const { requireDocker } = await import("./providers-Dh6hQDI6.js");
 		await requireDocker();
 		const removed = await pruneImages();
 		console.log(removed ? `Removed ${removed} image${removed === 1 ? "" : "s"} reeval built that nothing uses` : "No leftover images to remove");
@@ -1319,7 +1325,7 @@ function verbosityOf(flags) {
 * does. Without a config, just reeval's image.
 */
 async function imageBuildCommand(flags, configPath, filters) {
-	const { buildImages, imagePlanLines, imageTargets } = await import("./images-C9Y4ijbG.js");
+	const { buildImages, imagePlanLines, imageTargets } = await import("./images-CPfNRSuK.js");
 	const { docker } = await import("./sandbox.js");
 	let root = process.cwd();
 	const { fakeHarness } = await import("./harness.js");
@@ -1335,7 +1341,7 @@ async function imageBuildCommand(flags, configPath, filters) {
 			"dry-run": true
 		}, configPath));
 		root = config.root;
-		const { runEngine } = await import("./node-OXPxBD2N.js");
+		const { runEngine } = await import("./node-B72WWfgg.js");
 		let planned = [];
 		await runEngine({
 			run: JSON.parse(JSON.stringify({
@@ -1406,7 +1412,7 @@ async function runCommand(flags, configPath, filters) {
 	const setup = await prepareRun(flags, configPath);
 	const ctx = await startRun(setup);
 	invocation.runId = ctx.runId;
-	const { engineVersions } = await import("./node-OXPxBD2N.js");
+	const { engineVersions } = await import("./node-B72WWfgg.js");
 	const renderer = createRenderer({
 		out: setup.json ? { write: () => true } : process.stdout,
 		output: setup.output,
@@ -1428,14 +1434,14 @@ async function runCommand(flags, configPath, filters) {
 	const { engineRun, plan, sandboxVersions, budgetReached, live, notes, questions } = executed;
 	invocation.unhandledErrors = engineRun.unhandledErrors;
 	if ((interrupted || engineRun.cancelled) && plan?.sandboxes.some((s) => s.kind === "docker")) {
-		const { removeRunContainers } = await import("./docker-jf52ARP7.js");
+		const { removeRunContainers } = await import("./docker-CJkJ4-_n.js");
 		await removeRunContainers(ctx.runId);
 	}
 	if (notes) {
 		const cancelled = interrupted || engineRun.cancelled;
 		const written = await notes.finish(cancelled);
 		if (!cancelled && written.length) {
-			const { notesLine } = await import("./agent-BYy0Tc6P.js");
+			const { notesLine } = await import("./agent-Dcnbp_qS.js");
 			const text = notesLine(ctx.config.root, ctx.runId, written);
 			renderer.onEvent({
 				type: "notice",
@@ -1452,8 +1458,8 @@ async function runCommand(flags, configPath, filters) {
 	}
 	const stopped = interrupted || engineRun.cancelled;
 	const answers = questions && !stopped ? await questions.settle() : void 0;
-	const { lockfiles } = await import("./state-Bx-pH2ON.js");
-	const { writeScratchRun } = await import("./results-BI-zTqQy.js");
+	const { lockfiles } = await import("./state-BOeZS-Wa.js");
+	const { writeScratchRun } = await import("./results-EaWTGQtw.js");
 	const run = await writeScratchRun({
 		ctx,
 		author: ctx.author,
@@ -1476,8 +1482,8 @@ async function runCommand(flags, configPath, filters) {
 	}, () => live?.stop());
 	const { record } = run;
 	if (notes && !stopped) {
-		const { summarizeRun } = await import("./agent-BYy0Tc6P.js");
-		const { resolveAnalysisModel } = await import("./model-DQQQ3q_n.js");
+		const { summarizeRun } = await import("./agent-Dcnbp_qS.js");
+		const { resolveAnalysisModel } = await import("./model-BAB6ScG7.js");
 		const model = resolveAnalysisModel(ctx.config, "agent");
 		const { level, text } = await summarizeRun(ctx.config, record.runId, model);
 		renderer.onEvent({
@@ -1504,7 +1510,7 @@ async function runCommand(flags, configPath, filters) {
 	const { config } = ctx;
 	const mode = setup.record;
 	if ((flags.record || mode === "always" || mode === "ci" && inCI()) && record.status !== "interrupted") {
-		const { record: recordCmd } = await import("./commands-De-SxIgY.js");
+		const { record: recordCmd } = await import("./commands-8xIhRt-7.js");
 		const allow = record.status === "budget" ? { "allow-incomplete": true } : {};
 		console.log(await recordCmd(config, record.runId, {
 			...flags,
@@ -1539,10 +1545,10 @@ async function prepareRun(flags, configPath) {
 		runs = Number(flags.runs);
 		if (!(Number.isInteger(runs) && runs >= 1)) throw new UsageError(`--runs must be a whole number of at least 1 (got ${flags.runs})`);
 	}
-	const sandboxes = [config.sandbox, ...profile?.defaults.sandbox ? [profile.defaults.sandbox] : []];
+	const sandboxes = [environmentSandbox(config.environment, config.sandbox), ...profile?.defaults.environment ? [environmentSandbox(profile.defaults.environment, config.sandbox)] : []];
 	const concurrency = await resolveConcurrency(config, flags.concurrency, async () => {
 		if (!sandboxes.some((s) => s.kind === "docker")) return hostCapacity();
-		const { dockerCapacity } = await import("./src-C91RndLw.js");
+		const { dockerCapacity } = await import("./src-CQzUdIZj.js");
 		return await dockerCapacity() ?? hostCapacity();
 	});
 	return {
@@ -1567,11 +1573,11 @@ async function prepareRun(flags, configPath) {
 async function startRun({ config, profile, runs, tags, concurrency, verbosity, configFile }) {
 	const startedAt = /* @__PURE__ */ new Date();
 	const git = await gitInfo(config.root, [config.history.dir]);
-	const { worktreePatch } = await import("./state-Bx-pH2ON.js");
+	const { worktreePatch } = await import("./state-BOeZS-Wa.js");
 	const patch = git?.dirty ? await worktreePatch(config.root, [config.history.dir]) : null;
 	const who = await author(config.root);
-	const { stagingDir, sweepStaging } = await import("./results-BI-zTqQy.js");
-	const { scratchDir } = await import("./files-BC5Yqir-.js");
+	const { stagingDir, sweepStaging } = await import("./results-EaWTGQtw.js");
+	const { scratchDir } = await import("./files-BTF2wg0i.js");
 	const scratch = scratchDir(config.root);
 	const stamps = new Set([scratch, config.history.dir].flatMap((dir) => existsSync(dir) ? readdirSync(dir).map((name) => name.slice(0, 16)) : []));
 	let runId = newRunId(startedAt, git?.sha ?? null, who);
@@ -1608,8 +1614,8 @@ function unmatchedFilter(config, parts) {
 */
 async function executeRun(ctx, setup, filters, renderer) {
 	const { runId, stagingDir, config, runs, tags, concurrency, profile, verbosity, configFile } = ctx;
-	const { runEngine } = await import("./node-OXPxBD2N.js");
-	const { buildPlan, checkAttemptBudget, planDocument, planLine, planTable } = await import("./plan-DcHTdiJN.js");
+	const { runEngine } = await import("./node-B72WWfgg.js");
+	const { buildPlan, checkAttemptBudget, planDocument, planLine, planTable } = await import("./plan-4AO_VH20.js");
 	let plan;
 	let sandboxVersions = { images: [] };
 	let preflightStop;
@@ -1624,7 +1630,7 @@ async function executeRun(ctx, setup, filters, renderer) {
 		if (matched === 0) throw noMatch(config, filters, tags);
 		toRun = planned;
 		for (const p of planned) for (const name of p.attempt.secretVars) registerSecret(process.env[name]);
-		const models = planned.some((p) => !p.attempt.plan.skip && p.attempt.plan.questions?.length) ? await import("./model-DQQQ3q_n.js") : void 0;
+		const models = planned.some((p) => !p.attempt.plan.skip && p.attempt.plan.questions?.length) ? await import("./model-BAB6ScG7.js") : void 0;
 		plan = await buildPlan({
 			planned,
 			profile,
@@ -1634,7 +1640,7 @@ async function executeRun(ctx, setup, filters, renderer) {
 			historyDir: config.history.dir,
 			questionModel: models?.questionModelLabel(config) ?? null
 		});
-		const { imageStages, imageTargets } = await import("./images-C9Y4ijbG.js");
+		const { imageStages, imageTargets } = await import("./images-CPfNRSuK.js");
 		const targets = imageTargets(planned.filter((p) => !p.attempt.plan.skip).map((p) => ({
 			id: p.id,
 			harness: p.attempt.harness,
@@ -1655,8 +1661,8 @@ async function executeRun(ctx, setup, filters, renderer) {
 		if (setup.dryRun) {
 			const todos = excluded.filter((e) => e.mode === "todo").map((e) => e.id);
 			lines.push(...planTable(plan, todos));
-			const { imageStatusLines } = await import("./images-C9Y4ijbG.js");
-			const { dockerAvailable } = await import("./src-C91RndLw.js");
+			const { imageStatusLines } = await import("./images-CPfNRSuK.js");
+			const { dockerAvailable } = await import("./src-CQzUdIZj.js");
 			if (targets.length && await dockerAvailable()) lines.push(...await imageStatusLines(targets, config.root));
 			else if (targets.length) lines.push("images  not checked: Docker is not running");
 		}
@@ -1670,7 +1676,7 @@ async function executeRun(ctx, setup, filters, renderer) {
 		});
 		if (setup.dryRun) return false;
 		checkAttemptBudget(plan, setup.force, config.file);
-		const analysis = config.analysis.agent ? await import("./model-DQQQ3q_n.js") : null;
+		const analysis = config.analysis.agent ? await import("./model-BAB6ScG7.js") : null;
 		analysis?.analysisModelName(config, "agent");
 		const questionModels = /* @__PURE__ */ new Set();
 		const asked = [];
@@ -1699,7 +1705,7 @@ async function executeRun(ctx, setup, filters, renderer) {
 		});
 		if (malformed.length) throw new UsageError(`Credentials with whitespace or a line break in them: ${malformed.join(", ")}. A key copied from a wrapped terminal line picks these up; set it again as one line.`);
 		if (analysis) {
-			const { startNotes } = await import("./agent-BYy0Tc6P.js");
+			const { startNotes } = await import("./agent-Dcnbp_qS.js");
 			notes = startNotes({
 				config,
 				model: analysis.resolveAnalysisModel(config, "agent"),
@@ -1715,7 +1721,7 @@ async function executeRun(ctx, setup, filters, renderer) {
 			});
 		}
 		if (asked.length) {
-			const { questionRunner } = await import("./questions-CzXdVVsh.js");
+			const { questionRunner } = await import("./questions-nUB6aKz_.js");
 			questions = questionRunner({
 				evals: asked,
 				stagingDir,
@@ -1732,7 +1738,7 @@ async function executeRun(ctx, setup, filters, renderer) {
 				}
 			});
 		}
-		const { prepareWorkspaces, workspaceKey } = await import("./workspace-DXV77ZqW.js");
+		const { prepareWorkspaces, workspaceKey } = await import("./workspace-BwKzynEj.js");
 		const live = planned.filter((p) => !p.attempt.plan.skip);
 		const outcomes = await prepareWorkspaces(live.flatMap(({ attempt: a }) => a.workspace ? [{
 			workspace: a.workspace,
@@ -1888,9 +1894,9 @@ async function executeRun(ctx, setup, filters, renderer) {
 		throw error;
 	} finally {
 		clearTimeout(budgetTimer);
-		const { removePrepared } = await import("./preflight-BQyUr_hW.js");
+		const { removePrepared } = await import("./preflight-B8XgaXdX.js");
 		removePrepared(config.root, runId);
-		const { removePreparedWorkspaces } = await import("./workspace-DXV77ZqW.js");
+		const { removePreparedWorkspaces } = await import("./workspace-BwKzynEj.js");
 		removePreparedWorkspaces(config.root, runId);
 	}
 	if (preflightStop) throw preflightStop;
@@ -1973,4 +1979,4 @@ run(process.argv.slice(2)).finally(() => {
 //#endregion
 export {};
 
-//# sourceMappingURL=cli-DemDdg8J.js.map
+//# sourceMappingURL=cli-Ben0kbJt.js.map