Clarify the Hermes native question fixture objective

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

tests/runner-e2e/README.md

@@ -1607,6 +1607,13 @@ budgets, settled reported billing, pre-credential source admission and cleanup
 pipeline. It allows one attempt and records its harness source digest. Its
 browser-reconnect scope does not qualify controller restart, cancellation,
 Stop, subscriptions, or Daytona until each corresponding live check passes.
+Definition version 2 checks the question-only objective against the production
+file/document delivery guard before any live execution. The first version's
+comma-separated prohibition triggered a file-output requirement: its native
+callback/reconnect checks passed, but completion rejected and the run timed out.
+That failed attempt is retained. Standalone prohibitions clarify the same
+question-only task; the production delivery guard and positive file requirements
+remain in force.
 
 ```sh
 pnpm test:e2e:runner -- --list --suite hermes-native-interactions

tests/runner-e2e/catalog.ts

@@ -1219,8 +1219,9 @@ export const runnerSuites: readonly RunnerSuiteFixture[] = [
     groups: ["native"], profiles: [extendedHarnessProfiles.find(profile => profile.qualificationCandidate === "hermes")!],
     environments: runnerEnvironments, tasks: [hermesNativeQuestionTask], expectedMatrixSize: 2,
     definitionMetadata: {
-      version: 1, qualification: "pending", scheduling: "explicit-only", accountMethod: "api_key", accountMode: "responsible_user",
+      version: 2, qualification: "pending", scheduling: "explicit-only", accountMethod: "api_key", accountMode: "responsible_user",
       providerTurns: 1, lifecycle: "per-turn", nativeMethod: "_hermes/ask_questions", maximumAttemptsPerCell: 1,
+      objectiveAdmission: "production-delivery-guard-question-only",
       budgetMonthlyCents: HERMES_API_CONNECTION_BUDGET_CENTS, billing: "reported-cost-and-budget-health",
       coverage: "native-question-batch-browser-reconnect-exact-delivery", sourceDigest: hermesApiConnectionDefinitionDigest,
     },

tests/runner-e2e/hermes-api-connections.test.ts

@@ -2,6 +2,7 @@ import { describe, expect, it } from "vitest";
 import { runnerMatrix, runnerSuites, suiteDefinitionHash } from "./catalog.js";
 import { buildMatrixJobs, parseRunnerSelectors, selectRunnerExecutions } from "./selectors.js";
 import { buildRunnerE2EProcessEnvironment } from "./harness-env.js";
+import { explicitlyRequestsFileOutput, explicitlyRequestsTaskDocumentOutput } from "../../server/src/services/native-runtime/native-deliverable-feedback.js";
 import { captureHermesApiAccountOwner, captureHermesApiBudgets, captureHermesOpenRouterSettlement, gradeHermesApiConnection, isHermesOpenRouterWorkflow, isHermesConnectionSuite, HERMES_NATIVE_INTERACTION_SUITE, hasExactHermesNativeQuestionResponse, hasHermesNativeQuestionBatch, hermesNativeAnswerText } from "./hermes-api-connections.js";
 
 describe("Hermes native browser questions", () => {
@@ -48,6 +49,13 @@ describe("Hermes native browser questions", () => {
     expect(cells[0]!.task.buildPrompt("fixture-1")).not.toContain(answer);
     expect(cells[0]!.task.buildVisibleMarker("fixture-1")).toContain(answer);
   });
+  it("admits a question-only objective through the production delivery guard without suppressing file requirements", () => {
+    const prompt = cells[0]!.task.buildPrompt("fixture-guard");
+    expect(explicitlyRequestsFileOutput(prompt)).toBe(false);
+    expect(explicitlyRequestsTaskDocumentOutput(prompt)).toBe(false);
+    expect(explicitlyRequestsFileOutput("Write a JSON file with the returned answers.")).toBe(true);
+    expect(explicitlyRequestsTaskDocumentOutput("Write a document on this task with the returned answers.")).toBe(true);
+  });
   it.each(["acpx-runtime", "acpx-runtime-sidecar"])("accepts a complete native delivery from %s", adapter => {
     expect(hasHermesNativeQuestionBatch(questionSet)).toBe(true);
     expect(grade(native(adapter))).toBe(true);

tests/runner-e2e/hermes-api-connections.ts

@@ -40,9 +40,9 @@ export const hermesNativeQuestionTask: RunnerTaskFixture = {
       { id: "targets", question: "Choose the fixture targets", choices: ["Linux", "Mac"], multi_select: true },
       { id: "notes", question: "Describe the fixture constraint" },
     ]),
-    "Wait for the actual native callback. Do not answer your own questions, guess the reviewer's selections, substitute request_human_input, ask through prose, or create a plan or files. If clarify is unavailable, report that limitation without claiming completion.",
+    "Wait for the actual native callback. Do not answer your own questions. Do not guess the reviewer's selections. Do not substitute request_human_input. Do not ask through prose. Do not create files or plans. If clarify is unavailable, report that limitation without claiming completion.",
     `After the actual native result arrives, form your summary as HERMES-NATIVE-${nonce}-<exact returned notes user_response>. The notes are supplied by the reviewer; copy that returned text verbatim.`,
-    "Call paperclip_finish once with the current completion contract revision, satisfied objective, this computed summary and no remaining work. Wait for acceptance, then emit only that exact summary as your final answer.",
+    "Use the current turn completion revision and criterion IDs. Call paperclip_finish exactly once with {reportedWorkDisposition:'done',summary:COMPUTED_SUMMARY,completionClaim:{contractRevision:CURRENT_TURN_REVISION,objectiveSatisfied:true,criteria:[{criterionId:'objective',status:'satisfied',evidenceRefs:[]}],remainingWork:[]},evidence:[],verification:[]}. COMPUTED_SUMMARY and CURRENT_TURN_REVISION are placeholders for the actual returned-text summary and current revision. Wait for acceptance, then emit only that exact summary as your final answer.",
   ].join("\n"),
   buildVisibleMarker: nonce => `HERMES-NATIVE-${nonce}-${hermesNativeAnswerText(nonce)}`,
   buildMatchers: (nonce, execution) => [