@vibe-agent-toolkit/agent-skills 0.2.0-rc.7 → 0.2.0-rc.9

package/dist/skill-test/declared-env.js

@@ -11,23 +11,22 @@
  * agent-skills-specific concerns: token computation, interpolation, and gluing
  * the result onto the plugin-root env.
  */
-import { safePath } from '@vibe-agent-toolkit/utils';
+import { safePath, VatError } from '@vibe-agent-toolkit/utils';
 import { applyDeclaredEnv, formatForwardedEnvLine } from '@vibe-agent-toolkit/utils/skill-test';
 import { withPluginRootEnv } from './plugin-env.js';
 /**
  * A declared `env` value referenced an unknown `${token}`. Exit 2 (preflight):
  * fail loud naming the offending token rather than forwarding a literal `${x}`.
  */
-export class UnknownEnvTokenError extends Error {
+export class UnknownEnvTokenError extends VatError {
     token;
     key;
-    exitCode = 2;
+    reason = 'preflight';
     constructor(token, key) {
-        super(`Unknown interpolation token \${${token}} in env value for "${key}". ` +
+        super('UNKNOWN_ENV_TOKEN', `Unknown interpolation token \${${token}} in env value for "${key}". ` +
             `Known tokens: fixturesDir, stagedSkillDir, harnessRoot, resultsDir.`);
         this.token = token;
         this.key = key;
-        this.name = 'UnknownEnvTokenError';
     }
 }
 /**
@@ -40,18 +39,17 @@ export class UnknownEnvTokenError extends Error {
  * deleted directory produced a dead path the skill only discovered at runtime,
  * where it read as a skill bug rather than a harness one.
  */
-export class UnresolvableEnvTokenError extends Error {
+export class UnresolvableEnvTokenError extends VatError {
     token;
     key;
-    exitCode = 2;
+    reason = 'preflight';
     constructor(token, key) {
-        super(`Cannot resolve \${${token}} in env value for "${key}": this eval declares no input ` +
+        super('UNRESOLVABLE_ENV_TOKEN', `Cannot resolve \${${token}} in env value for "${key}": this eval declares no input ` +
             `\`files\`, so it has no staged fixtures directory. Declare the fixture under the ` +
             `eval's \`files\` list (it is staged into the eval's own workspace, which is also the ` +
             `executor's working directory), or drop \${${token}} from that env value.`);
         this.token = token;
         this.key = key;
-        this.name = 'UnresolvableEnvTokenError';
     }
 }
 /**

package/dist/skill-test/declared-env.js.map

@@ -1 +1 @@
-{"version":3,"file":"declared-env.js","sourceRoot":"","sources":["../../src/skill-test/declared-env.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AACrD,OAAO,EAAE,gBAAgB,EAAE,sBAAsB,EAAE,MAAM,sCAAsC,CAAC;AAEhG,OAAO,EAAE,iBAAiB,EAAE,MAAM,iBAAiB,CAAC;AAgBpD;;;GAGG;AACH,MAAM,OAAO,oBAAqB,SAAQ,KAAK;IAEjB;IAA+B;IADlD,QAAQ,GAAG,CAAU,CAAC;IAC/B,YAA4B,KAAa,EAAkB,GAAW;QACpE,KAAK,CACH,kCAAkC,KAAK,uBAAuB,GAAG,KAAK;YACpE,qEAAqE,CACxE,CAAC;QAJwB,UAAK,GAAL,KAAK,CAAQ;QAAkB,QAAG,GAAH,GAAG,CAAQ;QAKpE,IAAI,CAAC,IAAI,GAAG,sBAAsB,CAAC;IACrC,CAAC;CACF;AAED;;;;;;;;;GASG;AACH,MAAM,OAAO,yBAA0B,SAAQ,KAAK;IAEtB;IAA+B;IADlD,QAAQ,GAAG,CAAU,CAAC;IAC/B,YAA4B,KAAa,EAAkB,GAAW;QACpE,KAAK,CACH,qBAAqB,KAAK,uBAAuB,GAAG,iCAAiC;YACnF,mFAAmF;YACnF,uFAAuF;YACvF,6CAA6C,KAAK,wBAAwB,CAC7E,CAAC;QANwB,UAAK,GAAL,KAAK,CAAQ;QAAkB,QAAG,GAAH,GAAG,CAAQ;QAOpE,IAAI,CAAC,IAAI,GAAG,2BAA2B,CAAC;IAC1C,CAAC;CACF;AAeD;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,gBAAgB,CAAC,MAAsB;IACrD,OAAO;QACL,GAAG,CAAC,MAAM,CAAC,YAAY,KAAK,SAAS;YACnC,CAAC,CAAC,EAAE;YACJ,CAAC,CAAC,EAAE,WAAW,EAAE,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,YAAY,EAAE,UAAU,CAAC,EAAE,CAAC;QACpE,cAAc,EAAE,MAAM,CAAC,gBAAgB;QACvC,WAAW,EAAE,MAAM,CAAC,WAAW;QAC/B,UAAU,EAAE,MAAM,CAAC,UAAU;KAC9B,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,IAAY,EAAE,GAAW,EAAE,MAA8B;IAC5E,QAAQ,IAAI,EAAE,CAAC;QACb,KAAK,aAAa,CAAC,CAAC,CAAC;YACnB,IAAI,MAAM,CAAC,WAAW,KAAK,SAAS;gBAAE,MAAM,IAAI,yBAAyB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;YACrF,OAAO,MAAM,CAAC,WAAW,CAAC;QAC5B,CAAC;QACD,KAAK,gBAAgB,CAAC,CAAC,OAAO,MAAM,CAAC,cAAc,CAAC;QACpD,KAAK,aAAa,CAAC,CAAC,OAAO,MAAM,CAAC,WAAW,CAAC;QAC9C,KAAK,YAAY,CAAC,CAAC,OAAO,MAAM,CAAC,UAAU,CAAC;QAC5C,OAAO,CAAC,CAAC,MAAM,IAAI,oBAAoB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;IACrD,CAAC;AACH,CAAC;AAED,gGAAgG;AAChG,MAAM,UAAU,mBAAmB,CAAC,KAAa,EAAE,GAAW,EAAE,MAA8B;IAC5F,OAAO,KAAK,CAAC,UAAU,CAAC,gBAAgB,EAAE,CAAC,MAAM,EAAE,IAAY,EAAE,EAAE,CAAC,WAAW,CAAC,IAAI,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC,CAAC;AACtG,CAAC;AAED,oDAAoD;AACpD,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC,CAAC,aAAa,EAAE,gBAAgB,EAAE,aAAa,EAAE,YAAY,CAAC,CAAC,CAAC;AAE7F;;;;;;GAMG;AACH,MAAM,UAAU,oBAAoB,CAAC,SAA6C;IAChF,IAAI,SAAS,KAAK,SAAS;QAAE,OAAO;IACpC,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;QACrD,KAAK,MAAM,CAAC,EAAE,IAAI,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,gBAAgB,CAAC,EAAE,CAAC;YACxD,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC;gBAAE,MAAM,IAAI,oBAAoB,CAAC,IAAI,IAAI,EAAE,EAAE,GAAG,CAAC,CAAC;QACrF,CAAC;IACH,CAAC;AACH,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,gBAAgB,CAC9B,SAA6C,EAC7C,MAA8B;IAE9B,IAAI,SAAS,KAAK,SAAS;QAAE,OAAO,SAAS,CAAC;IAC9C,MAAM,GAAG,GAA2B,EAAE,CAAC;IACvC,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;QACrD,GAAG,CAAC,GAAG,CAAC,GAAG,mBAAmB,CAAC,KAAK,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;IACrD,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAsBD;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAAC,KAA4B;IAC3D,MAAM,QAAQ,GAAG,gBAAgB,CAAC,KAAK,CAAC,IAAI,EAAE;QAC5C,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACpD,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC3D,CAAC,CAAC;IACH,MAAM,GAAG,GAAG,iBAAiB,CAAC,QAAQ,CAAC,GAAG,EAAE,KAAK,CAAC,iBAAiB,CAAC,CAAC;IACrE,MAAM,IAAI,GAAG,sBAAsB,CAAC,GAAG,EAAE,QAAQ,CAAC,CAAC;IACnD,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,QAAQ,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC;AACpD,CAAC"}
+{"version":3,"file":"declared-env.js","sourceRoot":"","sources":["../../src/skill-test/declared-env.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAC/D,OAAO,EAAE,gBAAgB,EAAE,sBAAsB,EAAE,MAAM,sCAAsC,CAAC;AAEhG,OAAO,EAAE,iBAAiB,EAAE,MAAM,iBAAiB,CAAC;AAgBpD;;;GAGG;AACH,MAAM,OAAO,oBAAqB,SAAQ,QAAQ;IAEpB;IAA+B;IADlD,MAAM,GAAG,WAAoB,CAAC;IACvC,YAA4B,KAAa,EAAkB,GAAW;QACpE,KAAK,CACH,mBAAmB,EACnB,kCAAkC,KAAK,uBAAuB,GAAG,KAAK;YACpE,qEAAqE,CACxE,CAAC;QALwB,UAAK,GAAL,KAAK,CAAQ;QAAkB,QAAG,GAAH,GAAG,CAAQ;IAMtE,CAAC;CACF;AAED;;;;;;;;;GASG;AACH,MAAM,OAAO,yBAA0B,SAAQ,QAAQ;IAEzB;IAA+B;IADlD,MAAM,GAAG,WAAoB,CAAC;IACvC,YAA4B,KAAa,EAAkB,GAAW;QACpE,KAAK,CACH,wBAAwB,EACxB,qBAAqB,KAAK,uBAAuB,GAAG,iCAAiC;YACnF,mFAAmF;YACnF,uFAAuF;YACvF,6CAA6C,KAAK,wBAAwB,CAC7E,CAAC;QAPwB,UAAK,GAAL,KAAK,CAAQ;QAAkB,QAAG,GAAH,GAAG,CAAQ;IAQtE,CAAC;CACF;AAeD;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,gBAAgB,CAAC,MAAsB;IACrD,OAAO;QACL,GAAG,CAAC,MAAM,CAAC,YAAY,KAAK,SAAS;YACnC,CAAC,CAAC,EAAE;YACJ,CAAC,CAAC,EAAE,WAAW,EAAE,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,YAAY,EAAE,UAAU,CAAC,EAAE,CAAC;QACpE,cAAc,EAAE,MAAM,CAAC,gBAAgB;QACvC,WAAW,EAAE,MAAM,CAAC,WAAW;QAC/B,UAAU,EAAE,MAAM,CAAC,UAAU;KAC9B,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,IAAY,EAAE,GAAW,EAAE,MAA8B;IAC5E,QAAQ,IAAI,EAAE,CAAC;QACb,KAAK,aAAa,CAAC,CAAC,CAAC;YACnB,IAAI,MAAM,CAAC,WAAW,KAAK,SAAS;gBAAE,MAAM,IAAI,yBAAyB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;YACrF,OAAO,MAAM,CAAC,WAAW,CAAC;QAC5B,CAAC;QACD,KAAK,gBAAgB,CAAC,CAAC,OAAO,MAAM,CAAC,cAAc,CAAC;QACpD,KAAK,aAAa,CAAC,CAAC,OAAO,MAAM,CAAC,WAAW,CAAC;QAC9C,KAAK,YAAY,CAAC,CAAC,OAAO,MAAM,CAAC,UAAU,CAAC;QAC5C,OAAO,CAAC,CAAC,MAAM,IAAI,oBAAoB,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;IACrD,CAAC;AACH,CAAC;AAED,gGAAgG;AAChG,MAAM,UAAU,mBAAmB,CAAC,KAAa,EAAE,GAAW,EAAE,MAA8B;IAC5F,OAAO,KAAK,CAAC,UAAU,CAAC,gBAAgB,EAAE,CAAC,MAAM,EAAE,IAAY,EAAE,EAAE,CAAC,WAAW,CAAC,IAAI,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC,CAAC;AACtG,CAAC;AAED,oDAAoD;AACpD,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC,CAAC,aAAa,EAAE,gBAAgB,EAAE,aAAa,EAAE,YAAY,CAAC,CAAC,CAAC;AAE7F;;;;;;GAMG;AACH,MAAM,UAAU,oBAAoB,CAAC,SAA6C;IAChF,IAAI,SAAS,KAAK,SAAS;QAAE,OAAO;IACpC,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;QACrD,KAAK,MAAM,CAAC,EAAE,IAAI,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,gBAAgB,CAAC,EAAE,CAAC;YACxD,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC;gBAAE,MAAM,IAAI,oBAAoB,CAAC,IAAI,IAAI,EAAE,EAAE,GAAG,CAAC,CAAC;QACrF,CAAC;IACH,CAAC;AACH,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,gBAAgB,CAC9B,SAA6C,EAC7C,MAA8B;IAE9B,IAAI,SAAS,KAAK,SAAS;QAAE,OAAO,SAAS,CAAC;IAC9C,MAAM,GAAG,GAA2B,EAAE,CAAC;IACvC,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;QACrD,GAAG,CAAC,GAAG,CAAC,GAAG,mBAAmB,CAAC,KAAK,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;IACrD,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAsBD;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAAC,KAA4B;IAC3D,MAAM,QAAQ,GAAG,gBAAgB,CAAC,KAAK,CAAC,IAAI,EAAE;QAC5C,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACpD,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC3D,CAAC,CAAC;IACH,MAAM,GAAG,GAAG,iBAAiB,CAAC,QAAQ,CAAC,GAAG,EAAE,KAAK,CAAC,iBAAiB,CAAC,CAAC;IACrE,MAAM,IAAI,GAAG,sBAAsB,CAAC,GAAG,EAAE,QAAQ,CAAC,CAAC;IACnD,OAAO,EAAE,GAAG,EAAE,QAAQ,EAAE,QAAQ,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC;AACpD,CAAC"}

package/dist/skill-test/eval-executor.d.ts

@@ -43,7 +43,7 @@ export interface ExecutorOutcome {
 }
 /**
  * Run ONE eval's executor: a blind `claude -p` spawn that performs `task`
- * against the staged subject (issue #145). The executor's entire stream-json
+ * against the staged subject. The executor's entire stream-json
  * stdout is accumulated in memory (never written to a skill-writable disk
  * path — the "stop discarding stdout" fix) and echoed to `onProgress` chunk
  * by chunk for visibility.

package/dist/skill-test/eval-executor.js

@@ -1,10 +1,10 @@
 import { parseStreamJsonTranscript, spawnHeadlessClaude, } from '@vibe-agent-toolkit/utils/skill-test';
 import { assertExecutorPromptInvariants, buildExecutorPrompt } from './executor-prompt.js';
-import { InternalHarnessError } from './exit-codes.js';
+import { InternalHarnessError } from './failure-reason.js';
 import { RateLimitSignal } from './pipeline.js';
 /**
  * Run ONE eval's executor: a blind `claude -p` spawn that performs `task`
- * against the staged subject (issue #145). The executor's entire stream-json
+ * against the staged subject. The executor's entire stream-json
  * stdout is accumulated in memory (never written to a skill-writable disk
  * path — the "stop discarding stdout" fix) and echoed to `onProgress` chunk
  * by chunk for visibility.

package/dist/skill-test/eval-executor.js.map

@@ -1 +1 @@
-{"version":3,"file":"eval-executor.js","sourceRoot":"","sources":["../../src/skill-test/eval-executor.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,yBAAyB,EACzB,mBAAmB,GAGpB,MAAM,sCAAsC,CAAC;AAE9C,OAAO,EAAE,8BAA8B,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAC;AAC3F,OAAO,EAAE,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACvD,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AA8ChD;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,KAAuB;IAC9D,MAAM,OAAO,GAAG,KAAK,CAAC,YAAY,CAAC;IACnC,MAAM,MAAM,GAAG,mBAAmB,CAAC;QACjC,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,GAAG,CAAC,KAAK,CAAC,gBAAgB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,KAAK,CAAC,gBAAgB,EAAE,CAAC;QACxF,YAAY,EAAE,KAAK,CAAC,YAAY;KACjC,CAAC,CAAC;IACH,yEAAyE;IACzE,0EAA0E;IAC1E,yEAAyE;IACzE,0EAA0E;IAC1E,8BAA8B,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IAEnD,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,IAAI,mBAAmB,CAAC;IAEjD,IAAI,UAAU,GAAG,EAAE,CAAC;IACpB,MAAM,QAAQ,GAAG,CAAC,KAAa,EAAQ,EAAE;QACvC,UAAU,IAAI,KAAK,CAAC;QACpB,KAAK,CAAC,UAAU,EAAE,CAAC,KAAK,CAAC,CAAC;IAC5B,CAAC,CAAC;IAEF,IAAI,WAAwB,CAAC;IAC7B,IAAI,CAAC;QACH,WAAW,GAAG,MAAM,KAAK,CAAC;YACxB,MAAM;YACN,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,UAAU,EAAE,OAAO;YACnB,GAAG,EAAE,OAAO;YACZ,GAAG,EAAE,KAAK,CAAC,GAAG;YACd,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,QAAQ;YACR,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,YAAY,EAAE,KAAK,CAAC,YAAY;YAChC,GAAG,CAAC,KAAK,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC;YAC5D,GAAG,CAAC,KAAK,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC;SACnE,CAAC,CAAC;IACL,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,IAAI,oBAAoB,CAC5B,mCAAmC,KAAK,CAAC,MAAM,MAAM,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CACxG,CAAC;IACJ,CAAC;IAED,sEAAsE;IACtE,yEAAyE;IACzE,IAAI,WAAW,CAAC,OAAO,EAAE,CAAC;QACxB,MAAM,IAAI,oBAAoB,CAC5B,8BAA8B,KAAK,CAAC,MAAM,oBAAoB,KAAK,CAAC,OAAO,IAAI,CAAC,MAAM,CACvF,CAAC;IACJ,CAAC;IACD,IAAI,WAAW,CAAC,QAAQ,EAAE,CAAC;QACzB,MAAM,IAAI,oBAAoB,CAC5B,gCAAgC,KAAK,CAAC,MAAM,YAAY,KAAK,CAAC,SAAS,MAAM,CAC9E,CAAC;IACJ,CAAC;IAED,MAAM,MAAM,GAAG,yBAAyB,CAAC,UAAU,CAAC,CAAC;IAErD,2EAA2E;IAC3E,4EAA4E;IAC5E,6DAA6D;IAC7D,IAAI,MAAM,CAAC,WAAW,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACnD,MAAM,IAAI,eAAe,CAAC,mCAAmC,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC;IACjF,CAAC;IAED,OAAO;QACL,UAAU;QACV,MAAM;QACN,WAAW;QACX,YAAY,EAAE,WAAW,CAAC,MAAM,KAAK,CAAC;KACvC,CAAC;AACJ,CAAC"}
+{"version":3,"file":"eval-executor.js","sourceRoot":"","sources":["../../src/skill-test/eval-executor.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,yBAAyB,EACzB,mBAAmB,GAGpB,MAAM,sCAAsC,CAAC;AAE9C,OAAO,EAAE,8BAA8B,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAC;AAC3F,OAAO,EAAE,oBAAoB,EAAE,MAAM,qBAAqB,CAAC;AAC3D,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AA8ChD;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,KAAuB;IAC9D,MAAM,OAAO,GAAG,KAAK,CAAC,YAAY,CAAC;IACnC,MAAM,MAAM,GAAG,mBAAmB,CAAC;QACjC,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,GAAG,CAAC,KAAK,CAAC,gBAAgB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,KAAK,CAAC,gBAAgB,EAAE,CAAC;QACxF,YAAY,EAAE,KAAK,CAAC,YAAY;KACjC,CAAC,CAAC;IACH,yEAAyE;IACzE,0EAA0E;IAC1E,yEAAyE;IACzE,0EAA0E;IAC1E,8BAA8B,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IAEnD,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,IAAI,mBAAmB,CAAC;IAEjD,IAAI,UAAU,GAAG,EAAE,CAAC;IACpB,MAAM,QAAQ,GAAG,CAAC,KAAa,EAAQ,EAAE;QACvC,UAAU,IAAI,KAAK,CAAC;QACpB,KAAK,CAAC,UAAU,EAAE,CAAC,KAAK,CAAC,CAAC;IAC5B,CAAC,CAAC;IAEF,IAAI,WAAwB,CAAC;IAC7B,IAAI,CAAC;QACH,WAAW,GAAG,MAAM,KAAK,CAAC;YACxB,MAAM;YACN,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,UAAU,EAAE,OAAO;YACnB,GAAG,EAAE,OAAO;YACZ,GAAG,EAAE,KAAK,CAAC,GAAG;YACd,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,QAAQ;YACR,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,YAAY,EAAE,KAAK,CAAC,YAAY;YAChC,GAAG,CAAC,KAAK,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC;YAC5D,GAAG,CAAC,KAAK,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC;SACnE,CAAC,CAAC;IACL,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,IAAI,oBAAoB,CAC5B,mCAAmC,KAAK,CAAC,MAAM,MAAM,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CACxG,CAAC;IACJ,CAAC;IAED,sEAAsE;IACtE,yEAAyE;IACzE,IAAI,WAAW,CAAC,OAAO,EAAE,CAAC;QACxB,MAAM,IAAI,oBAAoB,CAC5B,8BAA8B,KAAK,CAAC,MAAM,oBAAoB,KAAK,CAAC,OAAO,IAAI,CAAC,MAAM,CACvF,CAAC;IACJ,CAAC;IACD,IAAI,WAAW,CAAC,QAAQ,EAAE,CAAC;QACzB,MAAM,IAAI,oBAAoB,CAC5B,gCAAgC,KAAK,CAAC,MAAM,YAAY,KAAK,CAAC,SAAS,MAAM,CAC9E,CAAC;IACJ,CAAC;IAED,MAAM,MAAM,GAAG,yBAAyB,CAAC,UAAU,CAAC,CAAC;IAErD,2EAA2E;IAC3E,4EAA4E;IAC5E,6DAA6D;IAC7D,IAAI,MAAM,CAAC,WAAW,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACnD,MAAM,IAAI,eAAe,CAAC,mCAAmC,KAAK,CAAC,MAAM,IAAI,CAAC,CAAC;IACjF,CAAC;IAED,OAAO;QACL,UAAU;QACV,MAAM;QACN,WAAW;QACX,YAAY,EAAE,WAAW,CAAC,MAAM,KAAK,CAAC;KACvC,CAAC;AACJ,CAAC"}

package/dist/skill-test/eval-fragment.d.ts

@@ -1,3 +1,4 @@
+import { VatError } from '@vibe-agent-toolkit/utils';
 import { z } from 'zod';
 /** One graded expectation inside a per-eval grader fragment. */
 export declare const EvalFragmentExpectationSchema: z.ZodObject<{
@@ -15,7 +16,7 @@ export declare const EvalFragmentExpectationSchema: z.ZodObject<{
 }>;
 export type EvalFragmentExpectation = z.infer<typeof EvalFragmentExpectationSchema>;
 /**
- * Strict shape of a single grader subagent's output (issue #145 Task 4). One
+ * Strict shape of a single grader subagent's output. One
  * GRADER spawn grades ONE eval's expectations from that eval's captured
  * transcript and writes exactly one fragment matching this schema; vat reads
  * every eval's fragment back and merges them into the run's aggregate grading
@@ -28,7 +29,7 @@ export type EvalFragmentExpectation = z.infer<typeof EvalFragmentExpectationSche
  * `arm` distinguishes a WITH/WITHOUT baseline run's fragment; absent means the
  * default 'with' (skill present) arm.
  *
- * `tool` (issue #145 Phase T) carries this eval's tool-expectation verdict —
+ * `tool` carries this eval's tool-expectation verdict —
  * a SEPARATE channel from `expectations[]` and from `friction` (see
  * tool-eval-schema.ts). It is {@link ToolVerdictBodySchema}, i.e. a
  * {@link import('./tool-eval-schema.js').ToolVerdict} WITHOUT `evalId`: the
@@ -38,7 +39,7 @@ export type EvalFragmentExpectation = z.infer<typeof EvalFragmentExpectationSche
  */
 export declare const EvalFragmentSchema: z.ZodObject<{
     runNonce: z.ZodString;
-    evalId: z.ZodString;
+    evalId: z.ZodEffects<z.ZodUnion<[z.ZodString, z.ZodNumber]>, string, string | number>;
     arm: z.ZodOptional<z.ZodEnum<["with", "without"]>>;
     expectations: z.ZodArray<z.ZodObject<{
         text: z.ZodString;
@@ -273,7 +274,7 @@ export declare const EvalFragmentSchema: z.ZodObject<{
         passed: boolean;
         evidence?: string | undefined;
     }[];
-    evalId: string;
+    evalId: string | number;
     runNonce: string;
     tool?: {
         passed: boolean;
@@ -328,7 +329,8 @@ export type EvalFragment = z.infer<typeof EvalFragmentSchema>;
  * should point at the offending eval, not at "Re-sync the vendored
  * skill-creator".
  */
-export declare class EvalFragmentError extends Error {
+export declare class EvalFragmentError extends VatError {
+    readonly reason: "internal";
     constructor(message: string);
 }
 /**

package/dist/skill-test/eval-fragment.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"eval-fragment.d.ts","sourceRoot":"","sources":["../../src/skill-test/eval-fragment.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAOxB,gEAAgE;AAChE,eAAO,MAAM,6BAA6B;;;;;;;;;;;;EAI/B,CAAC;AAEZ,MAAM,MAAM,uBAAuB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,6BAA6B,CAAC,CAAC;AAEpF;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,eAAO,MAAM,kBAAkB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAO7B;;;;;OAKG;;;;;;;;;;;;;;IAEH;;;;;;;;;;;;;;OAcG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAEM,CAAC;AAEZ,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAE9D;;;;;;;;;GASG;AACH,qBAAa,iBAAkB,SAAQ,KAAK;gBAC9B,OAAO,EAAE,MAAM;CAI5B;AA+DD;;;;;;;;;;;;;;;;;;GAkBG;AACH,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,GAAG,YAAY,CAyB/F"}
+{"version":3,"file":"eval-fragment.d.ts","sourceRoot":"","sources":["../../src/skill-test/eval-fragment.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AACrD,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAOxB,gEAAgE;AAChE,eAAO,MAAM,6BAA6B;;;;;;;;;;;;EAI/B,CAAC;AAEZ,MAAM,MAAM,uBAAuB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,6BAA6B,CAAC,CAAC;AAEpF;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,eAAO,MAAM,kBAAkB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAU7B;;;;;OAKG;;;;;;;;;;;;;;IAEH;;;;;;;;;;;;;;OAcG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAEM,CAAC;AAEZ,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAE9D;;;;;;;;;GASG;AACH,qBAAa,iBAAkB,SAAQ,QAAQ;IAC7C,QAAQ,CAAC,MAAM,EAAG,UAAU,CAAU;gBAC1B,OAAO,EAAE,MAAM;CAG5B;AAgED;;;;;;;;;;;;;;;;;;GAkBG;AACH,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,GAAG,YAAY,CAyB/F"}

package/dist/skill-test/eval-fragment.js

@@ -1,3 +1,4 @@
+import { VatError } from '@vibe-agent-toolkit/utils';
 import { z } from 'zod';
 import { BaselineContaminationHitSchema, BaselineScanDegradationSchema } from './baseline-integrity.js';
 import { FrictionItemSchema } from './friction-schema.js';
@@ -10,7 +11,7 @@ export const EvalFragmentExpectationSchema = z.object({
     evidence: z.string().optional(),
 }).strict();
 /**
- * Strict shape of a single grader subagent's output (issue #145 Task 4). One
+ * Strict shape of a single grader subagent's output. One
  * GRADER spawn grades ONE eval's expectations from that eval's captured
  * transcript and writes exactly one fragment matching this schema; vat reads
  * every eval's fragment back and merges them into the run's aggregate grading
@@ -23,7 +24,7 @@ export const EvalFragmentExpectationSchema = z.object({
  * `arm` distinguishes a WITH/WITHOUT baseline run's fragment; absent means the
  * default 'with' (skill present) arm.
  *
- * `tool` (issue #145 Phase T) carries this eval's tool-expectation verdict —
+ * `tool` carries this eval's tool-expectation verdict —
  * a SEPARATE channel from `expectations[]` and from `friction` (see
  * tool-eval-schema.ts). It is {@link ToolVerdictBodySchema}, i.e. a
  * {@link import('./tool-eval-schema.js').ToolVerdict} WITHOUT `evalId`: the
@@ -33,7 +34,10 @@ export const EvalFragmentExpectationSchema = z.object({
  */
 export const EvalFragmentSchema = z.object({
     runNonce: z.string().min(1),
-    evalId: z.string().min(1),
+    // evals.json allows an integer id and every other layer carries it as a
+    // string; a grader that echoes the number names the same eval, so it is
+    // normalised here rather than refused as a forged fragment.
+    evalId: z.union([z.string().min(1), z.number().int()]).transform(String),
     arm: z.enum(['with', 'without']).optional(),
     expectations: z.array(EvalFragmentExpectationSchema).min(1),
     friction: z.array(FrictionItemSchema).optional(),
@@ -72,10 +76,10 @@ export const EvalFragmentSchema = z.object({
  * should point at the offending eval, not at "Re-sync the vendored
  * skill-creator".
  */
-export class EvalFragmentError extends Error {
+export class EvalFragmentError extends VatError {
+    reason = 'internal';
     constructor(message) {
-        super(message);
-        this.name = 'EvalFragmentError';
+        super('EVAL_FRAGMENT', message);
     }
 }
 /**
@@ -88,6 +92,8 @@ export class EvalFragmentError extends Error {
 function extractRawEvalId(raw) {
     if (typeof raw === 'object' && raw !== null && 'evalId' in raw) {
         const value = raw['evalId'];
+        if (typeof value === 'number')
+            return `"${String(value)}"`;
         if (typeof value === 'string' && value.length > 0)
             return `"${sanitizeGraderText(value)}"`;
     }
@@ -106,7 +112,7 @@ const UNSANITIZED_FRAGMENT_KEYS = new Set(['runNonce']);
  * Sanitize ONLY the (non-verdict-bearing) `friction` field before the strict
  * fragment parse. Friction is auxiliary advisory data, not a verdict channel —
  * so a grader that wobbles on the friction shape (e.g. emits bare strings, the
- * common failure mode — PR #147) must NOT be allowed to discard the
+ * common failure mode) must NOT be allowed to discard the
  * verdict-bearing grading for the whole run. We drop friction items that don't
  * match {@link FrictionItemSchema} (and a `friction` that isn't an array at
  * all), then hand the rest to the STRICT fragment parse — so the verdict

package/dist/skill-test/eval-fragment.js.map

@@ -1 +1 @@
-{"version":3,"file":"eval-fragment.js","sourceRoot":"","sources":["../../src/skill-test/eval-fragment.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,8BAA8B,EAAE,6BAA6B,EAAE,MAAM,yBAAyB,CAAC;AACxG,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC1D,OAAO,EAAE,kBAAkB,EAAE,sBAAsB,EAAE,MAAM,kBAAkB,CAAC;AAC9E,OAAO,EAAE,qBAAqB,EAAE,MAAM,uBAAuB,CAAC;AAE9D,gEAAgE;AAChE,MAAM,CAAC,MAAM,6BAA6B,GAAG,CAAC,CAAC,MAAM,CAAC;IACpD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE;IACnB,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAChC,CAAC,CAAC,MAAM,EAAE,CAAC;AAIZ;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC3B,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACzB,GAAG,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC,CAAC,QAAQ,EAAE;IAC3C,YAAY,EAAE,CAAC,CAAC,KAAK,CAAC,6BAA6B,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC3D,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC,QAAQ,EAAE;IAChD,IAAI,EAAE,qBAAqB,CAAC,QAAQ,EAAE;IACtC;;;;;OAKG;IACH,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,8BAA8B,CAAC,CAAC,QAAQ,EAAE;IACjE;;;;;;;;;;;;;;OAcG;IACH,QAAQ,EAAE,6BAA6B,CAAC,QAAQ,EAAE;CACnD,CAAC,CAAC,MAAM,EAAE,CAAC;AAIZ;;;;;;;;;GASG;AACH,MAAM,OAAO,iBAAkB,SAAQ,KAAK;IAC1C,YAAY,OAAe;QACzB,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,mBAAmB,CAAC;IAClC,CAAC;CACF;AAED;;;;;;GAMG;AACH,SAAS,gBAAgB,CAAC,GAAY;IACpC,IAAI,OAAO,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI,IAAI,QAAQ,IAAI,GAAG,EAAE,CAAC;QAC/D,MAAM,KAAK,GAAI,GAA+B,CAAC,QAAQ,CAAC,CAAC;QACzD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,IAAI,kBAAkB,CAAC,KAAK,CAAC,GAAG,CAAC;IAC7F,CAAC;IACD,OAAO,WAAW,CAAC;AACrB,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,yBAAyB,GAAwB,IAAI,GAAG,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC;AAE7E;;;;;;;;;;;;;;GAcG;AACH,SAAS,qBAAqB,CAAC,GAAY;IACzC,IAAI,OAAO,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI,IAAI,CAAC,CAAC,UAAU,IAAI,GAAG,CAAC,EAAE,CAAC;QACpE,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IACpC,CAAC;IACD,MAAM,GAAG,GAAG,GAA8B,CAAC;IAC3C,MAAM,QAAQ,GAAG,GAAG,CAAC,UAAU,CAAC,CAAC;IACjC,IAAI,QAAQ,KAAK,SAAS;QAAE,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IAC9D,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;QAC7B,4EAA4E;QAC5E,yDAAyD;QACzD,MAAM,IAAI,GAA4B,EAAE,CAAC;QACzC,KAAK,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;YAC7C,IAAI,GAAG,KAAK,UAAU;gBAAE,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC;QAC1C,CAAC;QACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IACrC,CAAC;IACD,MAAM,IAAI,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,kBAAkB,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC;IACnF,IAAI,IAAI,CAAC,MAAM,KAAK,QAAQ,CAAC,MAAM;QAAE,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IACvE,OAAO,EAAE,KAAK,EAAE,EAAE,GAAG,GAAG,EAAE,QAAQ,EAAE,IAAI,EAAE,EAAE,OAAO,EAAE,QAAQ,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;AACvF,CAAC;AAED;;;;;;;;;;;;;;;;;;GAkBG;AACH,MAAM,UAAU,iBAAiB,CAAC,GAAY,EAAE,MAAiC;IAC/E,MAAM,SAAS,GAAG,sBAAsB,CAAC,GAAG,EAAE,yBAAyB,CAAC,CAAC;IACzE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,qBAAqB,CAAC,SAAS,CAAC,CAAC;IAC5D,IAAI,OAAO,GAAG,CAAC,EAAE,CAAC;QAChB,MAAM,CACJ,4BAA4B,gBAAgB,CAAC,SAAS,CAAC,QAAQ,OAAO,gCAAgC;YACpG,6DAA6D,CAChE,CAAC;IACJ,CAAC;IACD,MAAM,MAAM,GAAG,kBAAkB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACnD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,UAAU,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QAC1C,4EAA4E;QAC5E,0EAA0E;QAC1E,6EAA6E;QAC7E,4EAA4E;QAC5E,uEAAuE;QACvE,MAAM,IAAI,GAAG,kBAAkB,CAAC,UAAU,EAAE,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,QAAQ,CAAC,CAAC;QACxE,MAAM,KAAK,GAAG,kBAAkB,CAAC,UAAU,EAAE,OAAO,IAAI,SAAS,CAAC,CAAC;QACnE,MAAM,MAAM,GAAG,gBAAgB,CAAC,SAAS,CAAC,CAAC;QAC3C,MAAM,IAAI,iBAAiB,CACzB,4BAA4B,MAAM,4CAA4C,IAAI,MAAM,KAAK,GAAG,CACjG,CAAC;IACJ,CAAC;IACD,OAAO,MAAM,CAAC,IAAI,CAAC;AACrB,CAAC"}
+{"version":3,"file":"eval-fragment.js","sourceRoot":"","sources":["../../src/skill-test/eval-fragment.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AACrD,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,8BAA8B,EAAE,6BAA6B,EAAE,MAAM,yBAAyB,CAAC;AACxG,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAC1D,OAAO,EAAE,kBAAkB,EAAE,sBAAsB,EAAE,MAAM,kBAAkB,CAAC;AAC9E,OAAO,EAAE,qBAAqB,EAAE,MAAM,uBAAuB,CAAC;AAE9D,gEAAgE;AAChE,MAAM,CAAC,MAAM,6BAA6B,GAAG,CAAC,CAAC,MAAM,CAAC;IACpD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE;IACnB,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAChC,CAAC,CAAC,MAAM,EAAE,CAAC;AAIZ;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC3B,wEAAwE;IACxE,wEAAwE;IACxE,4DAA4D;IAC5D,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,MAAM,CAAC;IACxE,GAAG,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC,CAAC,QAAQ,EAAE;IAC3C,YAAY,EAAE,CAAC,CAAC,KAAK,CAAC,6BAA6B,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC3D,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC,QAAQ,EAAE;IAChD,IAAI,EAAE,qBAAqB,CAAC,QAAQ,EAAE;IACtC;;;;;OAKG;IACH,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,8BAA8B,CAAC,CAAC,QAAQ,EAAE;IACjE;;;;;;;;;;;;;;OAcG;IACH,QAAQ,EAAE,6BAA6B,CAAC,QAAQ,EAAE;CACnD,CAAC,CAAC,MAAM,EAAE,CAAC;AAIZ;;;;;;;;;GASG;AACH,MAAM,OAAO,iBAAkB,SAAQ,QAAQ;IACpC,MAAM,GAAG,UAAmB,CAAC;IACtC,YAAY,OAAe;QACzB,KAAK,CAAC,eAAe,EAAE,OAAO,CAAC,CAAC;IAClC,CAAC;CACF;AAED;;;;;;GAMG;AACH,SAAS,gBAAgB,CAAC,GAAY;IACpC,IAAI,OAAO,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI,IAAI,QAAQ,IAAI,GAAG,EAAE,CAAC;QAC/D,MAAM,KAAK,GAAI,GAA+B,CAAC,QAAQ,CAAC,CAAC;QACzD,IAAI,OAAO,KAAK,KAAK,QAAQ;YAAE,OAAO,IAAI,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC;QAC3D,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,IAAI,kBAAkB,CAAC,KAAK,CAAC,GAAG,CAAC;IAC7F,CAAC;IACD,OAAO,WAAW,CAAC;AACrB,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,yBAAyB,GAAwB,IAAI,GAAG,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC;AAE7E;;;;;;;;;;;;;;GAcG;AACH,SAAS,qBAAqB,CAAC,GAAY;IACzC,IAAI,OAAO,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI,IAAI,CAAC,CAAC,UAAU,IAAI,GAAG,CAAC,EAAE,CAAC;QACpE,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IACpC,CAAC;IACD,MAAM,GAAG,GAAG,GAA8B,CAAC;IAC3C,MAAM,QAAQ,GAAG,GAAG,CAAC,UAAU,CAAC,CAAC;IACjC,IAAI,QAAQ,KAAK,SAAS;QAAE,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IAC9D,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;QAC7B,4EAA4E;QAC5E,yDAAyD;QACzD,MAAM,IAAI,GAA4B,EAAE,CAAC;QACzC,KAAK,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;YAC7C,IAAI,GAAG,KAAK,UAAU;gBAAE,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC;QAC1C,CAAC;QACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IACrC,CAAC;IACD,MAAM,IAAI,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,kBAAkB,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC;IACnF,IAAI,IAAI,CAAC,MAAM,KAAK,QAAQ,CAAC,MAAM;QAAE,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IACvE,OAAO,EAAE,KAAK,EAAE,EAAE,GAAG,GAAG,EAAE,QAAQ,EAAE,IAAI,EAAE,EAAE,OAAO,EAAE,QAAQ,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;AACvF,CAAC;AAED;;;;;;;;;;;;;;;;;;GAkBG;AACH,MAAM,UAAU,iBAAiB,CAAC,GAAY,EAAE,MAAiC;IAC/E,MAAM,SAAS,GAAG,sBAAsB,CAAC,GAAG,EAAE,yBAAyB,CAAC,CAAC;IACzE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,qBAAqB,CAAC,SAAS,CAAC,CAAC;IAC5D,IAAI,OAAO,GAAG,CAAC,EAAE,CAAC;QAChB,MAAM,CACJ,4BAA4B,gBAAgB,CAAC,SAAS,CAAC,QAAQ,OAAO,gCAAgC;YACpG,6DAA6D,CAChE,CAAC;IACJ,CAAC;IACD,MAAM,MAAM,GAAG,kBAAkB,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACnD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,UAAU,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QAC1C,4EAA4E;QAC5E,0EAA0E;QAC1E,6EAA6E;QAC7E,4EAA4E;QAC5E,uEAAuE;QACvE,MAAM,IAAI,GAAG,kBAAkB,CAAC,UAAU,EAAE,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,QAAQ,CAAC,CAAC;QACxE,MAAM,KAAK,GAAG,kBAAkB,CAAC,UAAU,EAAE,OAAO,IAAI,SAAS,CAAC,CAAC;QACnE,MAAM,MAAM,GAAG,gBAAgB,CAAC,SAAS,CAAC,CAAC;QAC3C,MAAM,IAAI,iBAAiB,CACzB,4BAA4B,MAAM,4CAA4C,IAAI,MAAM,KAAK,GAAG,CACjG,CAAC;IACJ,CAAC;IACD,OAAO,MAAM,CAAC,IAAI,CAAC;AACrB,CAAC"}

package/dist/skill-test/eval-grader.d.ts

@@ -1,3 +1,4 @@
+import { VatError } from '@vibe-agent-toolkit/utils';
 import { spawnHeadlessClaude } from '@vibe-agent-toolkit/utils/skill-test';
 import { type EvalFragment } from './eval-fragment.js';
 import type { ToolExpectations } from './eval-inputs.js';
@@ -10,7 +11,7 @@ export interface RunGraderInput {
     /** Path to the vendored skill-creator grader rubric (references/grader.md). */
     rubricPath: string;
     /**
-     * The eval's declared tool expectations (issue #145 Phase T). Optional — when
+     * The eval's declared tool expectations. Optional — when
      * absent, `buildGraderPrompt` emits no tool-verdict instruction and the
      * returned fragment omits `tool`. See `grader-prompt.ts`'s
      * `BuildGraderPromptOptions.toolExpectations`.
@@ -49,12 +50,63 @@ export interface RunGraderInput {
      */
     onProgress: (chunk: string) => void;
     /**
-     * Reports this grader session's `total_cost_usd` (parsed from its stream-json
+     * Reports each grader session's `total_cost_usd` (parsed from its stream-json
      * transcript's terminal result; `undefined` when the transcript carried none,
-     * e.g. a mock spawn). Called once after a successful grader spawn so the run can
-     * aggregate total spend across all executor+grader sessions (adopter follow-up).
+     * e.g. a mock spawn). Called once per successful grader spawn — twice when the
+     * unparseable-fragment re-grade fires — so the run can aggregate total spend
+     * across all executor+grader sessions (adopter follow-up).
      */
     costSink?: (totalCostUsd: number | undefined) => void;
 }
+/** The reason named in {@link GraderFragmentUnparseableError}'s message and in every `evidence` of the failing fragment vat synthesizes from it. */
+export declare const GRADER_FRAGMENT_UNPARSEABLE = "grader-fragment-unparseable";
+/**
+ * Thrown when the grader's fragment was not valid JSON on BOTH attempts (see
+ * {@link runGraderForEval}). A per-eval GRADING failure, deliberately neither
+ * {@link InternalHarnessError} nor `EvalFragmentError`: the harness did its
+ * job twice and the model could not write strict JSON, which is a verdict about
+ * this eval's grading, not about the suite — so it ends on the verdict path
+ * (`FINDINGS`), never on `ERROR` with `Reason: internal`. `runEvalWorker` (run-harness.ts)
+ * turns it into that eval's failing fragment (reason
+ * {@link GRADER_FRAGMENT_UNPARSEABLE}) on the treatment arm, and into a recorded
+ * control-arm failure on the other — where a fabricated `fail` would read as
+ * skill lift. `detail` is already sanitized. Carries no `reason` on purpose:
+ * `skillTestFailureReason` would read it only if this escaped, and escaping IS
+ * the bug — the worker catches it on both arms.
+ */
+export declare class GraderFragmentUnparseableError extends VatError {
+    readonly evalId: string;
+    readonly detail: string;
+    constructor(evalId: string, detail: string);
+}
+/**
+ * Run ONE eval's grader: a blind `claude -p` spawn (grader model, NO
+ * skill/plugin loaded) that reads the executor's captured transcript via its
+ * prompt and writes a nonce'd fragment JSON to `graderOutDir`.
+ * `graderOutDir` is vat-only and separate from the skill sandbox the
+ * executor ran in — the grader's `sandboxDir` --add-dir is
+ * `graderOutDir` itself, and `pluginDirs: []` means no skill code ever runs
+ * inside the grader's process, so skill code cannot forge or tamper with the
+ * fragment this run trusts.
+ *
+ * The grader is vat infrastructure, not the thing under test: unlike the
+ * executor (see eval-executor.ts), there is NO clean-failure path here. A
+ * spawn rejection, watchdog kill (`timedOut`/`stalled`), non-zero exit, or a
+ * missing fragment file all mean the grader failed to do its ONE job and are
+ * thrown as {@link InternalHarnessError} (`ERROR`, `Reason: internal`) — never laundered into a
+ * passing (or even a valid failing) verdict.
+ *
+ * A fragment that exists but is not valid JSON is the one exception, because
+ * the grader MODEL writes it: one bad escape in a free-text `evidence` took a
+ * six-eval suite down as harness-broke after five evals had graded fine. That
+ * attempt is re-run ONCE — same transcript, a FRESH nonce (so the stale file, or
+ * anything left behind, cannot satisfy the retry), the bad file consumed first,
+ * and the prompt told what went wrong. Only a second unparseable fragment
+ * surfaces, as {@link GraderFragmentUnparseableError}. Once a fragment parses,
+ * shape problems throw `EvalFragmentError` (via {@link parseEvalFragment})
+ * and a nonce that doesn't match the nonce THAT attempt was prompted with throws
+ * {@link GradingNonceError} — both fail-closed with no retry, because the grader
+ * DID produce JSON, just not JSON we can trust.
+ */
 export declare function runGraderForEval(input: RunGraderInput): Promise<EvalFragment>;
 //# sourceMappingURL=eval-grader.d.ts.map

package/dist/skill-test/eval-grader.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"eval-grader.d.ts","sourceRoot":"","sources":["../../src/skill-test/eval-grader.ts"],"names":[],"mappings":"AAGA,OAAO,EAEL,mBAAmB,EACpB,MAAM,sCAAsC,CAAC;AAE9C,OAAO,EAAwC,KAAK,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAC7F,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AAOzD,MAAM,WAAW,cAAc;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,+EAA+E;IAC/E,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,+EAA+E;IAC/E,UAAU,EAAE,MAAM,CAAC;IACnB;;;;;OAKG;IACH,gBAAgB,CAAC,EAAE,gBAAgB,CAAC;IACpC,uGAAuG;IACvG,mBAAmB,CAAC,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAChF;;;;;OAKG;IACH,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,yEAAyE;IACzE,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,iEAAiE;IACjE,GAAG,EAAE,MAAM,CAAC,UAAU,CAAC;IACvB,mFAAmF;IACnF,KAAK,CAAC,EAAE,OAAO,mBAAmB,CAAC;IACnC;;;;;;OAMG;IACH,UAAU,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,CAAC;IACpC;;;;;OAKG;IACH,QAAQ,CAAC,EAAE,CAAC,YAAY,EAAE,MAAM,GAAG,SAAS,KAAK,IAAI,CAAC;CACvD;AAkCD,wBAAsB,gBAAgB,CAAC,KAAK,EAAE,cAAc,GAAG,OAAO,CAAC,YAAY,CAAC,CAyHnF"}
+{"version":3,"file":"eval-grader.d.ts","sourceRoot":"","sources":["../../src/skill-test/eval-grader.ts"],"names":[],"mappings":"AAGA,OAAO,EAA2B,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAC9E,OAAO,EAEL,mBAAmB,EACpB,MAAM,sCAAsC,CAAC;AAE9C,OAAO,EAAqB,KAAK,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAC1E,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AAOzD,MAAM,WAAW,cAAc;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,+EAA+E;IAC/E,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,EAAE,CAAC;IACvB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,+EAA+E;IAC/E,UAAU,EAAE,MAAM,CAAC;IACnB;;;;;OAKG;IACH,gBAAgB,CAAC,EAAE,gBAAgB,CAAC;IACpC,uGAAuG;IACvG,mBAAmB,CAAC,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAChF;;;;;OAKG;IACH,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,yEAAyE;IACzE,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,iEAAiE;IACjE,GAAG,EAAE,MAAM,CAAC,UAAU,CAAC;IACvB,mFAAmF;IACnF,KAAK,CAAC,EAAE,OAAO,mBAAmB,CAAC;IACnC;;;;;;OAMG;IACH,UAAU,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,CAAC;IACpC;;;;;;OAMG;IACH,QAAQ,CAAC,EAAE,CAAC,YAAY,EAAE,MAAM,GAAG,SAAS,KAAK,IAAI,CAAC;CACvD;AAED,oJAAoJ;AACpJ,eAAO,MAAM,2BAA2B,gCAAgC,CAAC;AAEzE;;;;;;;;;;;;;GAaG;AACH,qBAAa,8BAA+B,SAAQ,QAAQ;aAExC,MAAM,EAAE,MAAM;aACd,MAAM,EAAE,MAAM;gBADd,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,MAAM;CAQjC;AAWD;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AACH,wBAAsB,gBAAgB,CAAC,KAAK,EAAE,cAAc,GAAG,OAAO,CAAC,YAAY,CAAC,CAwDnF"}

package/dist/skill-test/eval-grader.js

@@ -1,18 +1,53 @@
-import { existsSync, readFileSync, unlinkSync } from 'node:fs';
-import { mkdirSyncReal, safePath } from '@vibe-agent-toolkit/utils';
+import { randomBytes } from 'node:crypto';
+import { existsSync, readFileSync, rmSync } from 'node:fs';
+import { mkdirSyncReal, safePath, VatError } from '@vibe-agent-toolkit/utils';
 import { parseStreamJsonTranscript, spawnHeadlessClaude, } from '@vibe-agent-toolkit/utils/skill-test';
-import { EvalFragmentError, parseEvalFragment } from './eval-fragment.js';
-import { InternalHarnessError } from './exit-codes.js';
+import { parseEvalFragment } from './eval-fragment.js';
+import { InternalHarnessError } from './failure-reason.js';
 import { assertGraderPromptInvariants, buildGraderPrompt } from './grader-prompt.js';
 import { sanitizeGraderText } from './grader-text.js';
 import { GradingNonceError } from './grading-adapter.js';
 import { computeToolPassed } from './tool-eval-schema.js';
+/** The reason named in {@link GraderFragmentUnparseableError}'s message and in every `evidence` of the failing fragment vat synthesizes from it. */
+export const GRADER_FRAGMENT_UNPARSEABLE = 'grader-fragment-unparseable';
+/**
+ * Thrown when the grader's fragment was not valid JSON on BOTH attempts (see
+ * {@link runGraderForEval}). A per-eval GRADING failure, deliberately neither
+ * {@link InternalHarnessError} nor `EvalFragmentError`: the harness did its
+ * job twice and the model could not write strict JSON, which is a verdict about
+ * this eval's grading, not about the suite — so it ends on the verdict path
+ * (`FINDINGS`), never on `ERROR` with `Reason: internal`. `runEvalWorker` (run-harness.ts)
+ * turns it into that eval's failing fragment (reason
+ * {@link GRADER_FRAGMENT_UNPARSEABLE}) on the treatment arm, and into a recorded
+ * control-arm failure on the other — where a fabricated `fail` would read as
+ * skill lift. `detail` is already sanitized. Carries no `reason` on purpose:
+ * `skillTestFailureReason` would read it only if this escaped, and escaping IS
+ * the bug — the worker catches it on both arms.
+ */
+export class GraderFragmentUnparseableError extends VatError {
+    evalId;
+    detail;
+    constructor(evalId, detail) {
+        super('SKILL_TEST_GRADER_FRAGMENT_UNPARSEABLE', `${GRADER_FRAGMENT_UNPARSEABLE}: the grader for eval "${evalId}" wrote a fragment that was not valid JSON ` +
+            `on either attempt (${detail}).`);
+        this.evalId = evalId;
+        this.detail = detail;
+    }
+}
+/**
+ * Adapt the grader's stdout `onProgress` sink into the `onWarn` callback
+ * {@link parseEvalFragment} requires (called only when it drops malformed
+ * friction items).
+ */
+function fragmentWarnRouter(onProgress) {
+    return (message) => onProgress(`[skill-test] ${message}\n`);
+}
 /**
  * Run ONE eval's grader: a blind `claude -p` spawn (grader model, NO
  * skill/plugin loaded) that reads the executor's captured transcript via its
- * prompt and writes a nonce'd fragment JSON to `graderOutDir` (issue #145
- * Task 8). `graderOutDir` is vat-only and separate from the skill sandbox the
- * executor ran in (Task 7) — the grader's `sandboxDir` --add-dir is
+ * prompt and writes a nonce'd fragment JSON to `graderOutDir`.
+ * `graderOutDir` is vat-only and separate from the skill sandbox the
+ * executor ran in — the grader's `sandboxDir` --add-dir is
  * `graderOutDir` itself, and `pluginDirs: []` means no skill code ever runs
  * inside the grader's process, so skill code cannot forge or tamper with the
  * fragment this run trusts.
@@ -21,23 +56,21 @@ import { computeToolPassed } from './tool-eval-schema.js';
  * executor (see eval-executor.ts), there is NO clean-failure path here. A
  * spawn rejection, watchdog kill (`timedOut`/`stalled`), non-zero exit, or a
  * missing fragment file all mean the grader failed to do its ONE job and are
- * thrown as {@link InternalHarnessError} (exit 1) — never laundered into a
+ * thrown as {@link InternalHarnessError} (`ERROR`, `Reason: internal`) — never laundered into a
  * passing (or even a valid failing) verdict.
  *
- * Once the fragment file exists, shape/JSON problems throw
- * {@link EvalFragmentError} (via {@link parseEvalFragment}) and a nonce that
- * doesn't match this run's expected nonce throws {@link GradingNonceError} —
- * both distinct from harness breakage because the grader DID run and produce
- * *something*, just not something we can trust.
+ * A fragment that exists but is not valid JSON is the one exception, because
+ * the grader MODEL writes it: one bad escape in a free-text `evidence` took a
+ * six-eval suite down as harness-broke after five evals had graded fine. That
+ * attempt is re-run ONCE — same transcript, a FRESH nonce (so the stale file, or
+ * anything left behind, cannot satisfy the retry), the bad file consumed first,
+ * and the prompt told what went wrong. Only a second unparseable fragment
+ * surfaces, as {@link GraderFragmentUnparseableError}. Once a fragment parses,
+ * shape problems throw `EvalFragmentError` (via {@link parseEvalFragment})
+ * and a nonce that doesn't match the nonce THAT attempt was prompted with throws
+ * {@link GradingNonceError} — both fail-closed with no retry, because the grader
+ * DID produce JSON, just not JSON we can trust.
  */
-/**
- * Adapt the grader's stdout `onProgress` sink into the `onWarn` callback
- * {@link parseEvalFragment} requires (called only when it drops malformed
- * friction items).
- */
-function fragmentWarnRouter(onProgress) {
-    return (message) => onProgress(`[skill-test] ${message}\n`);
-}
 export async function runGraderForEval(input) {
     // 0700, like every other vat-only directory this codebase creates (the grader
     // root in `resolveGraderOutDir`, the workspaces root and every eval workspace in
@@ -53,17 +86,60 @@ export async function runGraderForEval(input) {
     // join is safe, but `joinUnderRoot` is used defensively — it costs nothing
     // and closes off any future path where an unvalidated id reaches here.
     const fragmentOut = safePath.joinUnderRoot(input.graderOutDir, `${input.evalId}.json`);
-    const prompt = buildGraderPrompt({
+    let attempt = await runGraderAttempt(input, fragmentOut, input.nonce, undefined);
+    if (attempt.unparseable !== undefined) {
+        input.onProgress(`[skill-test] grader fragment for eval "${input.evalId}" was not valid JSON (${attempt.unparseable}); re-grading once.\n`);
+        // Minted here, not derived from the run nonce: the retry must accept nothing
+        // written under the first prompt.
+        attempt = await runGraderAttempt(input, fragmentOut, randomBytes(16).toString('hex'), attempt.unparseable);
+        if (attempt.unparseable !== undefined) {
+            throw new GraderFragmentUnparseableError(input.evalId, attempt.unparseable);
+        }
+    }
+    const fragment = parseEvalFragment(attempt.parsed, fragmentWarnRouter(input.onProgress));
+    // Integrity gate: a missing/wrong nonce means this fragment was not produced
+    // by the grader we prompted for THIS attempt — most likely forged or left
+    // behind by untrusted skill code — so the verdict cannot be trusted, hard error.
+    if (fragment.runNonce !== attempt.nonce) {
+        throw new GradingNonceError(`eval "${input.evalId}" fragment \`runNonce\` does not match this run`);
+    }
+    assertExpectationCountDeclared(input.evalId, input.expectations, fragment.expectations);
+    assertToolVerdictConsistent(input.evalId, input.toolExpectations, fragment.tool);
+    // The grader's only input is the executor transcript, which untrusted skill code
+    // controls, and `evalId` is schema-typed as any non-empty string. It is echoed
+    // verbatim into the run summary and into `baseline.json` — including the
+    // CONTAMINATED banner — so a grader talked into emitting an id containing
+    // newlines and ANSI escapes can paint a reassuring "known false positive, the
+    // delta is valid" line directly beneath vat's own warning. VAT knows which eval
+    // it asked about; take the id from the request, never from the answer.
+    //
+    // `runNonce` is re-stamped for the same reason: the merge re-verifies every
+    // fragment against the ONE nonce the run minted, and a re-graded fragment has
+    // already passed the gate above against the nonce its own prompt carried.
+    return { ...fragment, evalId: input.evalId, runNonce: input.nonce };
+}
+/**
+ * ONE grader spawn under `nonce`: build and assert the prompt, spawn, apply the
+ * harness-breakage gates, report cost, consume the fragment file. `previousFailure`
+ * is the re-grade's only difference — a trailing line telling the grader what its
+ * first fragment got wrong (sanitized where it was caught, at the bottom of this
+ * function). Required, not defaulted, so both call sites say which attempt they are.
+ */
+async function runGraderAttempt(input, fragmentOut, nonce, previousFailure) {
+    const built = buildGraderPrompt({
         evalId: input.evalId,
         transcript: input.transcript,
         expectations: input.expectations,
         ...(input.expectedOutput === undefined ? {} : { expectedOutput: input.expectedOutput }),
         rubricPath: input.rubricPath,
         fragmentOut,
-        nonce: input.nonce,
+        nonce,
         ...(input.toolExpectations === undefined ? {} : { toolExpectations: input.toolExpectations }),
         ...(input.declaredExecutables === undefined ? {} : { declaredExecutables: input.declaredExecutables }),
     });
+    const prompt = previousFailure === undefined
+        ? built
+        : `${built}\n\nYour previous fragment was not valid JSON: ${previousFailure}. Write strict JSON.`;
     // Defense-in-depth (parity with eval-executor.ts's build→assert→use pattern):
     // re-verify the built grader prompt still carries its required invariants
     // (STOP, fragment path, browser/iteration forbids, nonce directive, and the
@@ -73,7 +149,7 @@ export async function runGraderForEval(input) {
     // manifest, and the suite's own expectations/expected_output — and none of
     // that attacker-controlled text can satisfy an invariant on the builder's
     // behalf.
-    assertGraderPromptInvariants(prompt, input.nonce);
+    assertGraderPromptInvariants(prompt, nonce);
     const spawn = input.spawn ?? spawnHeadlessClaude;
     // Accumulate the grader's stream-json stdout so we can read its terminal
     // `total_cost_usd` for run-wide spend aggregation (adopter follow-up). Held in
@@ -121,70 +197,59 @@ export async function runGraderForEval(input) {
     // the in-memory transcript (not the fragment) so it reflects real API spend even
     // under subscription auth; `undefined` when the transcript carried no result.
     input.costSink?.(parseStreamJsonTranscript(graderTranscript).result?.totalCostUsd);
-    const raw = readAndConsumeFragmentFile(fragmentOut, input.evalId, spawnResult.status);
-    const fragment = parseEvalFragment(raw, fragmentWarnRouter(input.onProgress));
-    // Integrity gate: a missing/wrong nonce means this fragment was not produced
-    // by the grader we prompted for THIS run — most likely forged or left behind
-    // by untrusted skill code — so the verdict cannot be trusted, hard error.
-    if (fragment.runNonce !== input.nonce) {
-        throw new GradingNonceError(`eval "${input.evalId}" fragment \`runNonce\` does not match this run`);
+    const warn = fragmentWarnRouter(input.onProgress);
+    const text = readAndConsumeFragmentFile(fragmentOut, input.evalId, spawnResult.status, warn);
+    try {
+        return { nonce, parsed: JSON.parse(text) };
+    }
+    catch (err) {
+        if (!(err instanceof SyntaxError))
+            throw err;
+        // V8 embeds a VERBATIM slice of the offending bytes in its SyntaxError
+        // message, unescaped — and this is one of the few grader-controlled strings
+        // that reaches an operator WITHOUT passing `parseEvalFragment`, the
+        // documented text boundary, because the parse it would have gone through is
+        // the one that just failed. It lands in a thrown message (which the CLI writes
+        // as `Error: ${err.message}`, so an unsanitized `ESC[2K CR ESC[32m…` wipes the
+        // line vat just printed and continues in vat's own colour) AND in the re-grade
+        // prompt.
+        return { nonce, unparseable: sanitizeGraderText(err.message) };
     }
-    assertExpectationCountDeclared(input.evalId, input.expectations, fragment.expectations);
-    assertToolVerdictConsistent(input.evalId, input.toolExpectations, fragment.tool);
-    // The grader's only input is the executor transcript, which untrusted skill code
-    // controls, and `evalId` is schema-typed as any non-empty string. It is echoed
-    // verbatim into the run summary and into `baseline.json` — including the
-    // CONTAMINATED banner — so a grader talked into emitting an id containing
-    // newlines and ANSI escapes can paint a reassuring "known false positive, the
-    // delta is valid" line directly beneath vat's own warning. VAT knows which eval
-    // it asked about; take the id from the request, never from the answer.
-    return { ...fragment, evalId: input.evalId };
 }
 /**
- * Read the grader's fragment JSON into memory and unlink it immediately —
- * consume-on-read. Split out of {@link runGraderForEval} to keep its cognitive
- * complexity within budget. Throws {@link InternalHarnessError} if the grader
- * wrote no fragment, {@link EvalFragmentError} if it is not valid JSON.
+ * Read the grader's fragment file into memory and remove it immediately —
+ * consume-on-read, whether or not it will parse. Throws {@link InternalHarnessError}
+ * if the grader wrote no fragment.
  *
- * The unlink is the security-relevant half: the grader dir is same-uid (see
+ * The removal is the security-relevant half: the grader dir is same-uid (see
  * `resolveGraderOutDir`), so a fragment left on disk lets skill code that
  * survived the process-group kill read the echoed nonce at leisure and forge
  * LATER fragments. Consuming it on read leaves no persisted copy — exposure
- * shrinks to each fragment's own read window (no cross-eval harvest). The read
- * here is the ONLY read of the file; all downstream logic runs off the returned
- * value. Best-effort: a failed unlink is not a run failure (end-of-run cleanup
- * removes the whole dir), and true isolation from same-uid code is issue #149.
+ * shrinks to each fragment's own read window (no cross-eval harvest). It is also
+ * what makes the re-grade honest: a retry that writes nothing finds no file, not
+ * the unparseable one. The read here is the ONLY read of the file; all downstream
+ * logic runs off the returned text. A failed removal is not a run failure
+ * (end-of-run cleanup removes the whole dir) but it IS the one degradation this
+ * design exists to prevent, so it is reported through `onWarn` rather than
+ * swallowed: the operator learns the fragment lingered, with the errno. True
+ * isolation from same-uid code is a separate, unbuilt mechanism (a different uid
+ * or a sandbox), not this removal.
  */
-function readAndConsumeFragmentFile(fragmentOut, evalId, status) {
-    // eslint-disable-next-line security/detect-non-literal-fs-filename -- fragmentOut is our own derived path (joinUnderRoot-guarded)
+function readAndConsumeFragmentFile(fragmentOut, evalId, status, onWarn) {
     if (!existsSync(fragmentOut)) {
         throw new InternalHarnessError(`Grader exited (status ${status}) without writing a fragment at ${fragmentOut} for eval "${evalId}".`);
     }
-    let raw;
+    const text = readFileSync(fragmentOut, 'utf-8');
     try {
-        // eslint-disable-next-line security/detect-non-literal-fs-filename -- fragmentOut is our own derived path (joinUnderRoot-guarded)
-        raw = JSON.parse(readFileSync(fragmentOut, 'utf-8'));
+        // `force` covers the one benign case — the file is already gone — so what the
+        // catch sees is a refusal (EACCES/EPERM/EBUSY), never an absence.
+        rmSync(fragmentOut, { force: true });
     }
     catch (err) {
-        // V8 embeds a VERBATIM slice of the offending bytes in its SyntaxError
-        // message, unescaped — and this is one of the few grader-controlled strings
-        // that reaches an operator WITHOUT passing `parseEvalFragment`, the
-        // documented text boundary, because the parse it would have gone through is
-        // the one that just failed. The CLI writes it as `Error: ${err.message}`, so
-        // an unsanitized fragment of `ESC[2K CR ESC[32m…` wipes the line vat just
-        // printed and continues in vat's own colour (and a <=16-byte file gets its
-        // whole `ESC[2J ESC[H` echoed, clearing the screen).
-        throw new EvalFragmentError(`grader fragment for eval "${evalId}" at ${fragmentOut} is not valid JSON: ` +
-            sanitizeGraderText(err instanceof Error ? err.message : String(err)));
-    }
-    try {
-        // eslint-disable-next-line security/detect-non-literal-fs-filename -- fragmentOut is our own derived path (joinUnderRoot-guarded)
-        unlinkSync(fragmentOut);
-    }
-    catch {
-        // Swallow: cleanup removes the grader dir regardless.
+        onWarn(`grader fragment for eval "${evalId}" at ${fragmentOut} could not be removed after it was read ` +
+            `(${err instanceof Error ? err.message : String(err)}); it stays on disk until end-of-run cleanup.`);
     }
-    return raw;
+    return text;
 }
 /**
  * The OUTPUT channel's declared-vs-reported check — the exact thing