test(runner): create explicit review cards in warm continuity prompts

tests/runner-e2e/FIXTURES.md

@@ -100,7 +100,7 @@ canonical Plan revision, capture its pending UI, approve in the browser, and
 prove exactly two successful runs. `warm_three_turn` provides exactly two
 browser follow-up messages, preserves one project/execution-workspace scope,
 verifies host file contents after every turn, and finishes within three
-ten-minute turn deadlines.
+ten-minute turn deadlines. Native warm turns 1 and 2 create a real confirmation card with `request_human_input` before reporting `needs_review`; the pending card supplies the required review gate. A disposition string alone does not create that gate. Turn 3 reports Done without another card.
 
 Every selected case runs in its own isolated Paperclip process, and independent
 cases may run concurrently. Follow-up turns inside one case retain their shared

tests/runner-e2e/catalog.test.ts

@@ -1,4 +1,5 @@
 import { describe, expect, it } from "vitest";
+import { createIssueThreadInteractionSchema } from "../../packages/shared/src/validators/issue.js";
 import {
   connectionReviewSuite,
   runnerEnvironments,
@@ -25,6 +26,30 @@ import {
 } from "./selectors.js";
 
 describe("runner E2E catalog", () => {
+  it("gives native warm turns a real review card before reporting needs_review", () => {
+    const prompts = [daytonaWarmContinuityTask.buildPrompt("nonce"), ...daytonaWarmContinuityTask.buildFollowupMessages!("nonce")];
+    for (const [index, prompt] of prompts.entries()) {
+      const match = prompt.match(/call request_human_input with (\{.*\})\. Reuse/);
+      if (index === 2) {
+        expect(match).toBeNull();
+        continue;
+      }
+      expect(match, `turn ${index + 1} must create its review interaction`).not.toBeNull();
+      const request = JSON.parse(match![1]);
+      const interaction = createIssueThreadInteractionSchema.parse({
+        kind: "request_confirmation", idempotencyKey: request.idempotencyKey,
+        title: request.title, summary: request.prompt, continuationPolicy: request.continuationPolicy,
+        payload: { ...request.payload, prompt: request.prompt },
+      });
+      expect(interaction).toMatchObject({
+        kind: "request_confirmation", continuationPolicy: "wake_assignee",
+        payload: { target: { type: "custom", revisionId: `nonce-T${index + 1}` }, rejectRequiresReason: true },
+      });
+      expect(request.interactionKind).toBe("confirmation");
+      expect(prompt.indexOf("call request_human_input with")).toBeLessThan(prompt.indexOf("then call paperclip_finish"));
+    }
+  });
+
   it("defines sixteen local connection-review journeys without expanding the default matrix", () => {
     expect(connectionReviewSuite.expectedMatrixSize).toBe(16);
     expect(new Set(connectionReviewSuite.profiles.map(profile => profile.id))).toEqual(new Set(["runner-codex", "runner-acpx-claude", "legacy-codex", "legacy-claude"]));

tests/runner-e2e/catalog.ts

@@ -811,6 +811,22 @@ function warmTurnInstructions(turn: 1 | 2 | 3, nonce: string) {
   );
   const marker = warmTurnMarker(turn, nonce);
   const finalTurn = turn === 3;
+  const nativeReviewRequest = {
+    interactionKind: "confirmation",
+    idempotencyKey: `daytona-warm-review-T${turn}-${nonce}`,
+    title: `Warm continuity turn ${turn}`,
+    prompt: `Is this warm continuity task ready to complete after turn ${turn}?`,
+    continuationPolicy: "wake_assignee",
+    payload: {
+      version: 1,
+      acceptLabel: "Approve completion",
+      rejectLabel: "Continue work",
+      rejectRequiresReason: true,
+      allowDeclineReason: true,
+      supersedeOnUserComment: false,
+      target: { type: "custom", key: `daytona_warm_turn_${turn}`, revisionId: `${nonce}-T${turn}`, label: `Warm continuity turn ${turn}` },
+    },
+  };
   const legacyCompletion = finalTurn
     ? `In a legacy runner, make exactly one public-API completion write after verification: PATCH /api/issues/$PAPERCLIP_TASK_ID with {"status":"done","comment":"${marker}"}. Include Authorization and X-Paperclip-Run-Id. Do not POST a separate comment.`
     : `In a legacy runner, after verification POST exactly one request_confirmation to /api/issues/$PAPERCLIP_TASK_ID/interactions with {"kind":"request_confirmation","idempotencyKey":"daytona-warm-review-T${turn}-${nonce}","resolverPolicy":"human_only","title":"Warm continuity turn ${turn}","summary":"Review completed warm continuity turn ${turn}.","continuationPolicy":"wake_assignee","payload":{"version":1,"prompt":"Is this warm continuity task ready to complete after turn ${turn}?","acceptLabel":"Approve completion","rejectLabel":"Continue work","rejectRequiresReason":true,"allowDeclineReason":true,"supersedeOnUserComment":false,"target":{"type":"custom","key":"daytona_warm_turn_${turn}","revisionId":"${nonce}-T${turn}","label":"Warm continuity turn ${turn}"}}}. Capture the returned interaction id. Then make exactly one issue PATCH with {"status":"in_review","comment":"${marker}","reviewInteractionId":"<returned interaction id>"}. Include Authorization and X-Paperclip-Run-Id on both writes. If the issue PATCH fails, retry only that PATCH and never create another interaction. Do not POST a separate comment. After both writes succeed, end the response and heartbeat immediately; do not wait or poll because the reviewer action will start the next turn.`;
@@ -820,6 +836,7 @@ function warmTurnInstructions(turn: 1 | 2 | 3, nonce: string) {
       ? `Create ${file} with exactly this one line followed by a newline: ${lines[0]}`
       : `Before changing anything, read ${file} and verify its content is exactly ${lines.slice(0, -1).join("\\n")} followed by a newline. Then append exactly ${lines.at(-1)} followed by a newline.`,
     `After the write, verify ${file} contains exactly these lines, once each and in order: ${lines.join(" | ")}.`,
+    ...(!finalTurn ? [`In a native runner, after verification and before reporting needs_review, call request_human_input with ${JSON.stringify(nativeReviewRequest)}. Reuse its idempotencyKey on a retry; do not create another card. The pending card is the review gate; do not report done while it is waiting.`] : []),
     `In a native runner, use the current turn completion revision and criterion IDs (CURRENT_TURN_REVISION below is a placeholder, not a literal), then call paperclip_finish exactly once with {reportedWorkDisposition:"${finalTurn ? "done" : "needs_review"}",summary:"${marker}",completionClaim:{contractRevision:CURRENT_TURN_REVISION,objectiveSatisfied:true,criteria:[{criterionId:"objective",status:"satisfied",evidenceRefs:[]}],remainingWork:[]},evidence:[],verification:[{commandOrCheck:"read ${file}",status:"passed"}]}. Wait for that tool call to succeed, then emit exactly ${marker} once as the complete user-facing final response.`,
     legacyCompletion,
     `In a legacy runner, the PATCH comment is the complete visible response. After its 2xx response, finish silently: do not print, echo, or emit ${marker} again as assistant text.`,