Clarify reviewer verification limits

prompts/arbiter-paired-room.md

@@ -34,6 +34,8 @@ You may receive reference opinions from external models appended to your prompt.
 ## Rules
 
 - Base your verdict on evidence (code, test output, logs), not on who said what first
+- Distinguish reviewer snapshot limits from real product bugs. Reviewer workspaces may intentionally omit heavy artifacts like `node_modules`, `dist`, and `build`; inability to run direct local test/typecheck/build there is not, by itself, a blocker if dedicated verification evidence exists
+- When verification evidence exists from the dedicated verification path, judge that evidence on its merits instead of requiring the reviewer to reproduce the same result from the lightweight reviewer snapshot
 - Your verdict is final for this deadlock cycle — after it, work resumes normally
 - You do NOT implement or review code — you only judge the disagreement
 - Keep your verdict concise — state the decision, the evidence, and the required action

prompts/claude-paired-room.md

@@ -30,6 +30,8 @@ Push back with evidence when the owner is wrong. Hold your ground when you are r
 ## Rules
 
 - Judge completion only by verification output. "It should work now" means run it. "I'm confident" means nothing — confidence is not evidence. "I tested earlier" means test again if code changed since. "It's a trivial change" means verify anyway
+- Reviewer workspaces may intentionally exclude heavy runtime artifacts such as `node_modules`, `dist`, and `build`. Do not treat the inability to run direct local test/typecheck/build from the reviewer snapshot as a product bug by itself
+- When test/typecheck/build evidence is needed, prefer the dedicated verification path (`run_verification`) over assuming the reviewer snapshot can execute the full project locally
 - Stagnation: **Spinning** (same error 3+), **Oscillation** (alternating approaches), **Diminishing returns** (shrinking improvement), **No progress** (discussion without change) — name the pattern and report: **Status**, **Attempted**, **Recommendation**
 - Implementation, commits, and pushes require agreement from both sides. Either can veto
 - Keep reviews concise — approve quickly when there is nothing to critique

runners/codex-runner/test/reviewer-runtime.test.ts

@@ -3,7 +3,7 @@ import fs from 'fs';
 import os from 'os';
 import path from 'path';
 
-import { describe, expect, it } from 'vitest';
+import { afterEach, beforeEach, describe, expect, it } from 'vitest';
 
 import {
   assertReadonlyWorkspaceRepoConnectivity,
@@ -11,6 +11,22 @@ import {
   isReviewerRuntime,
 } from '../src/reviewer-runtime.js';
 
+const ORIGINAL_UNSAFE_HOST_PAIRED_MODE =
+  process.env.EJCLAW_UNSAFE_HOST_PAIRED_MODE;
+
+afterEach(() => {
+  if (ORIGINAL_UNSAFE_HOST_PAIRED_MODE == null) {
+    delete process.env.EJCLAW_UNSAFE_HOST_PAIRED_MODE;
+  } else {
+    process.env.EJCLAW_UNSAFE_HOST_PAIRED_MODE =
+      ORIGINAL_UNSAFE_HOST_PAIRED_MODE;
+  }
+});
+
+beforeEach(() => {
+  delete process.env.EJCLAW_UNSAFE_HOST_PAIRED_MODE;
+});
+
 function createTempRepo(prefix: string): string {
   const cwd = fs.mkdtempSync(path.join(os.tmpdir(), prefix));
   execFileSync('git', ['init'], {