fix(evals): accept concise bound personal setup confirmations

Co-Authored-By: Paperclip <noreply@paperclip.ing>

tests/runner-e2e/chat-catalog-domains/external-personal-secret-question.test.ts

@@ -53,3 +53,16 @@ it("does not treat a pending or wrong-person answer as the saved-value acknowled
   const answered = { ...row, status: "answered", resolvedByUserId: "other-person", result: { version: 1, answers: [{ questionId: "saved", optionIds: ["yes"] }] } };
   expect(personalSecretQuestionHandoff(answered, binding, "answered")).toBeUndefined();
 });
+it("uses bound completion context for terse answers without treating permission or negation as completion", () => {
+  const withChoice = (label: string, prompt = "Have you saved your own value?") => {
+    const row = question(), qs = row.payload.questionSet!;
+    row.payload = questionSetToAskUserQuestionsPayload({ ...qs, questions: [{ ...qs.questions[0]!, prompt,
+      options: [{ id: "yes", label }, { id: "later", label: "Not yet" }] }] });
+    return personalSecretQuestionHandoff(row, binding);
+  };
+  expect(withChoice("Done")).toMatchObject({ optionId: "yes" });
+  expect(withChoice("Yes")).toMatchObject({ optionId: "yes" });
+  expect(withChoice("Not done")).toBeUndefined();
+  expect(withChoice("Done for Riley")).toBeUndefined();
+  expect(withChoice("Yes", "May I save the value for you?")).toBeUndefined();
+});

tests/runner-e2e/chat-catalog-domains/external-personal-secret.ts

@@ -121,10 +121,13 @@ export function personalSecretQuestionHandoff(value: unknown, b: PersonalSecretQ
   if (!text.split(/[^A-Za-z0-9_-]+/).includes(b.definitionId) || !text.split(/[^A-Za-z0-9_-]+/).includes("PERSONAL_PUBLISHING_KEY") || !personalSecretGuidance(text)
     || !/settings/i.test(text) || !/my secrets/i.test(text) || !/set my value/i.test(text)
     || solicitsChatCredential([text, ...q.options.map(o => `${o.label} ${o.description ?? ""}`)].join("\n"))) return;
+  const completedSetupQuestion = /\b(?:have you|did you)\b[^?]{0,80}\b(?:saved|set|entered|added|configured|completed)\b/i.test(q.prompt)
+    && !/\b(?:not|never|haven't|didn't|want|permission)\b|\b(?:may|can|should|could) (?:i|we)\b/i.test(q.prompt);
   const confirmations = q.options.filter(o => {
     const answer = `${o.label} ${o.description ?? ""}`;
-    return /\b(?:yes|saved|entered|added|done)\b/i.test(answer) && /\b(?:my|own|personal)\b/i.test(answer)
-      && !/\b(?:not|never|cannot|can't|yet|need|later|will|another|their)\b/i.test(answer);
+    return (!/^yes\b/i.test(o.label.trim()) || completedSetupQuestion)
+      && (/\b(?:saved|entered|added|done)\b/i.test(answer) || (completedSetupQuestion && /^yes[.!]?$/i.test(o.label.trim())))
+      && !/\b(?:no|not|never|cannot|can.t|didn.t|haven.t|hasn.t|yet|need|later|will|would|Riley|another|other|their)\b/i.test(answer);
   });
   if (confirmations.length !== 1) return;
   const option = confirmations[0]!;