Merge dev stabilization into main

Promote STEP_DONE routing, paired telemetry, and bundled runner fixes from dev.
2026-04-24 14:12:00 +09:00
parent 4a1d30d787 3eee842164
commit a36c8a5f07
40 changed files with 1531 additions and 107 deletions
--- a/prompts/arbiter-paired-room.md
+++ b/prompts/arbiter-paired-room.md
@@ -34,7 +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 reading owner/reviewer summaries, treat **TASK_DONE** as full task completion, **STEP_DONE** as intermediate progress that should keep the owner flow alive, and **DONE** as a legacy alias for **TASK_DONE**
+- 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/lint 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
--- a/prompts/claude-paired-room.md
+++ b/prompts/claude-paired-room.md
@@ -22,9 +22,11 @@ If you see a materially better design, debugging path, or scoping choice, propos

 ## Completion status

-**Start your first line** with one of these four statuses. This is required.
+**Start your first line** with one of these six statuses. This is required.

- **DONE** — Approved. The owner's response is correct and complete. Include the evidence
+- **STEP_DONE** — The current step is acceptable, but the original requested task still has remaining work. Send the task back to the owner without escalating to the arbiter
+- **TASK_DONE** — Approved. The owner's work satisfies the full requested task. Include the evidence
+- **DONE** — Legacy alias for **TASK_DONE**. Prefer **TASK_DONE** for new turns
 - **DONE_WITH_CONCERNS** — Approved with concerns. List specific actions the owner must take. If the same concerns repeat for 2+ turns, escalate to BLOCKED
 - **BLOCKED** — Cannot proceed without user decision
 - **NEEDS_CONTEXT** — Missing information from user
@@ -32,10 +34,10 @@ If you see a materially better design, debugging path, or scoping choice, propos
 ## 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 runs against the owner's current workspace in read-only mode. Do not treat the inability to run direct local test/typecheck/build from the reviewer workspace as a product bug by itself
+- Reviewer runs against the owner's current workspace in read-only mode. Do not treat the inability to run direct local test/typecheck/build/lint from the reviewer workspace as a product bug by itself
 - Treat `EJCLAW_WORK_DIR` as the canonical verification root for this turn. You may inspect other local paths for context, but final review findings must be re-checked against `EJCLAW_WORK_DIR`
 - Do not use a different clone, canonical repo path, or cached session path as the sole basis for `BLOCKED`, `DONE_WITH_CONCERNS`, or change requests. If another path disagrees with `EJCLAW_WORK_DIR`, prefer `EJCLAW_WORK_DIR` and explicitly call out the mismatch
- When test/typecheck/build evidence is needed, prefer the dedicated verification path (`run_verification`) over assuming the reviewer workspace should execute the full project locally
+- When test/typecheck/build/lint evidence is needed, prefer the dedicated verification path (`run_verification`) over assuming the reviewer workspace should execute the full project locally
 - Separate correctness issues from improvement ideas. If something is only a better alternative, label it as optional instead of blocking the owner unnecessarily
 - 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
--- a/prompts/owner-common-paired-room.md
+++ b/prompts/owner-common-paired-room.md
@@ -18,16 +18,19 @@ Challenge the reviewer's reasoning. Point out logical gaps, over-engineering, sc

 ## Completion status

-**Start your first line** with one of these four statuses. This is required.
+**Start your first line** with one of these six statuses. This is required.

- **DONE** — All steps completed. Include the evidence (test output, build log, diff)
+- **STEP_DONE** — A meaningful intermediate step is complete, but the original task still has remaining work. This keeps the task active and continues the owner flow without reviewer or arbiter intervention
+- **TASK_DONE** — The original requested task is complete. Include the evidence (test output, build log, diff)
+- **DONE** — Legacy alias for **TASK_DONE**. Prefer **TASK_DONE** for new turns
 - **DONE_WITH_CONCERNS** — Completed, but there are issues worth flagging. If the reviewer raises the same concerns again, fix them or escalate to BLOCKED
 - **BLOCKED** — Cannot proceed. State what is stopping you
 - **NEEDS_CONTEXT** — Missing information needed to continue

 ### Finalize semantics

- When the reviewer already approved and you are finalizing, **DONE** closes the paired turn
+- When the reviewer already approved and you are finalizing, **TASK_DONE** closes the paired turn
+- In that same finalize step, **STEP_DONE** keeps the task active and resumes the owner flow because the original request still has remaining work
 - In that same finalize step, **DONE_WITH_CONCERNS** does not close the turn — it intentionally reopens review
 - Use **DONE_WITH_CONCERNS** on finalize only when you are explicitly asking the reviewer loop to resume

--- a/runners/agent-runner/src/bundled-cli-path.ts
+++ b/runners/agent-runner/src/bundled-cli-path.ts
@@ -45,7 +45,7 @@ function platformCandidates(
    return [
      {
        pkg: `@anthropic-ai/claude-agent-sdk-darwin-${arch}`,
-        file: arch === 'arm64' ? 'cli' : 'cli',
+        file: 'claude',
      },
    ];
  }
@@ -53,7 +53,7 @@ function platformCandidates(
    return [
      {
        pkg: `@anthropic-ai/claude-agent-sdk-win32-${arch}`,
-        file: 'cli.exe',
+        file: 'claude.exe',
      },
    ];
  }
@@ -127,12 +127,22 @@ export function resolveBundledClaudeCodeExecutable(options?: {
 * Default package directory resolver. Uses `require.resolve` against this
 * module's location so it works regardless of whether agent-runner is invoked
 * from its own node_modules layout or via a parent workspace.
+ *
+ * Important: the SDK's optional native-binary packages may not expose a bare
+ * package entrypoint, so `require.resolve(pkg)` can fail even when the package
+ * and binary are installed. Resolve `package.json` first, then fall back to the
+ * bare package only for package layouts that do expose an entrypoint.
 */
 function defaultResolvePackageDir(pkg: string): string | null {
+  const req = createRequire(import.meta.url);
+  try {
+    const pkgJson = req.resolve(`${pkg}/package.json`);
+    return path.dirname(pkgJson);
+  } catch {
+    // Fall through to legacy/bare-entrypoint resolution below.
+  }
+
  try {
-    // Resolve the package entrypoint instead of package.json because the SDK's
-    // exports map does not expose `./package.json`.
-    const req = createRequire(import.meta.url);
    const entrypoint = req.resolve(pkg);
    return path.dirname(entrypoint);
  } catch {
@@ -142,5 +152,6 @@ function defaultResolvePackageDir(pkg: string): string | null {

 export const __test__ = {
  platformCandidates,
+  defaultResolvePackageDir,
  ENV_OVERRIDE,
 };
--- a/runners/agent-runner/src/ipc-mcp-stdio.ts
+++ b/runners/agent-runner/src/ipc-mcp-stdio.ts
@@ -479,12 +479,12 @@ server.tool(

 server.tool(
  'run_verification',
-  'Run a fixed verification profile directly on the host against the current repo snapshot using a restricted environment and snapshot checks. Use this instead of broad shell write access for test/typecheck/build verification.',
+  'Run a fixed verification profile directly on the host against the current repo snapshot using a restricted environment and snapshot checks. Use this instead of broad shell write access for test/typecheck/build/lint verification.',
  {
    profile: z
      .enum(VERIFICATION_PROFILES)
      .describe(
-        'Fixed verification profile. Runs the workspace test/typecheck/build scripts with the repo-configured package manager.',
+        'Fixed verification profile. Runs the workspace test/typecheck/build/lint scripts with the repo-configured package manager.',
      ),
  },
  async (args) => {
--- a/runners/agent-runner/src/verification.ts
+++ b/runners/agent-runner/src/verification.ts
@@ -3,7 +3,12 @@ import path from 'path';
 import { pathToFileURL } from 'url';
 export { computeVerificationSnapshotId } from '../../../shared/verification-snapshot.js';

-export const VERIFICATION_PROFILES = ['test', 'typecheck', 'build'] as const;
+export const VERIFICATION_PROFILES = [
+  'test',
+  'typecheck',
+  'build',
+  'lint',
+] as const;

 export type VerificationProfile = (typeof VERIFICATION_PROFILES)[number];

--- a/runners/agent-runner/test/bundled-cli-path.test.ts
+++ b/runners/agent-runner/test/bundled-cli-path.test.ts
@@ -1,8 +1,12 @@
 import fs from 'fs';
+import { createRequire } from 'module';
 import path from 'path';
 import { afterEach, describe, expect, it } from 'vitest';

-import { resolveBundledClaudeCodeExecutable } from '../src/bundled-cli-path.js';
+import {
+  __test__,
+  resolveBundledClaudeCodeExecutable,
+} from '../src/bundled-cli-path.js';

 describe('resolveBundledClaudeCodeExecutable', () => {
  const origEnv = process.env.EJCLAW_CLAUDE_CLI_PATH;
@@ -83,10 +87,10 @@ describe('resolveBundledClaudeCodeExecutable', () => {
    expect(result).toBe(path.join(muslDir, 'claude'));
  });

-  it('resolves Darwin arm64 binary under `cli`', () => {
+  it('resolves Darwin arm64 binary under `claude`', () => {
    const dir =
      '/fake/node_modules/@anthropic-ai/claude-agent-sdk-darwin-arm64';
-    const existsSync = (p: string): boolean => p === path.join(dir, 'cli');
+    const existsSync = (p: string): boolean => p === path.join(dir, 'claude');
    const result = resolveBundledClaudeCodeExecutable({
      env: {},
      existsSync,
@@ -95,13 +99,13 @@ describe('resolveBundledClaudeCodeExecutable', () => {
      resolvePackageDir: (pkg) =>
        pkg === '@anthropic-ai/claude-agent-sdk-darwin-arm64' ? dir : null,
    });
-    expect(result).toBe(path.join(dir, 'cli'));
+    expect(result).toBe(path.join(dir, 'claude'));
  });

-  it('resolves Windows binary under `cli.exe`', () => {
+  it('resolves Windows binary under `claude.exe`', () => {
    const dir =
      'C:\\\\fake\\\\node_modules\\\\@anthropic-ai\\\\claude-agent-sdk-win32-x64';
-    const binary = path.join(dir, 'cli.exe');
+    const binary = path.join(dir, 'claude.exe');
    const existsSync = (p: string): boolean => p === binary;
    const result = resolveBundledClaudeCodeExecutable({
      env: {},
@@ -114,6 +118,17 @@ describe('resolveBundledClaudeCodeExecutable', () => {
    expect(result).toBe(binary);
  });

+  it('default resolver handles SDK optional packages without a bare entrypoint', () => {
+    if (process.platform !== 'linux' || process.arch !== 'x64') return;
+
+    const dir = __test__.defaultResolvePackageDir(
+      '@anthropic-ai/claude-agent-sdk-linux-x64',
+    );
+    if (!dir) return;
+
+    expect(fs.existsSync(path.join(dir, 'claude'))).toBe(true);
+  });
+
  it('throws a descriptive error when no binary is found', () => {
    const existsSync = () => false;
    expect(() =>
@@ -127,6 +142,34 @@ describe('resolveBundledClaudeCodeExecutable', () => {
    ).toThrow(/Unable to locate a bundled Claude Code CLI binary/);
  });

+  it('resolves the linux x64 optional package even when it has no JS entrypoint', () => {
+    if (process.platform !== 'linux' || process.arch !== 'x64') {
+      return;
+    }
+
+    const req = createRequire(import.meta.url);
+    let packageJsonPath: string;
+    try {
+      packageJsonPath = req.resolve(
+        '@anthropic-ai/claude-agent-sdk-linux-x64/package.json',
+      );
+    } catch {
+      return;
+    }
+
+    try {
+      req.resolve('@anthropic-ai/claude-agent-sdk-linux-x64');
+      return;
+    } catch {
+      // This is the regression case: package.json resolves, bare package does not.
+    }
+
+    const result = resolveBundledClaudeCodeExecutable({ env: {} });
+
+    expect(result).toBe(path.join(path.dirname(packageJsonPath), 'claude'));
+    expect(fs.existsSync(result)).toBe(true);
+  });
+
  it('end-to-end: resolves the actual bundled binary on this host when installed', () => {
    // Real-world smoke test. Some CI/host sandboxes do not install the optional
    // per-platform SDK package, so treat "package not resolvable" as a host
--- a/runners/agent-runner/test/reviewer-runtime.test.ts
+++ b/runners/agent-runner/test/reviewer-runtime.test.ts
@@ -136,7 +136,7 @@ describe('claude reviewer runtime guard', () => {
      ['/repo/work'],
      'linux',
      'best-effort',
-    );
+    ) as unknown as { failIfUnavailable?: boolean };

    expect(sandbox.failIfUnavailable).toBe(false);
  });
@@ -150,7 +150,7 @@ describe('claude reviewer runtime guard', () => {
      ['/repo/work'],
      'darwin',
      'best-effort',
-    );
+    ) as unknown as { failIfUnavailable?: boolean };

    expect(sandbox.failIfUnavailable).toBe(false);
  });
@@ -169,6 +169,9 @@ describe('claude reviewer runtime guard', () => {
    const env = buildReviewerGitGuardEnv({ PATH: process.env.PATH }, true);
    expect(env.PATH).toContain('ejclaw-reviewer-git-');
    expect(env.EJCLAW_REAL_GIT).toBeTruthy();
+    expect(env.GIT_CONFIG_GLOBAL).toContain('global.gitconfig');
+    expect(env.GIT_CONFIG_NOSYSTEM).toBe('1');
+    expect(fs.existsSync(env.GIT_CONFIG_GLOBAL!)).toBe(true);
  });

  it('prefers an executable HOME-scoped wrapper dir before tmp', () => {
@@ -245,6 +248,30 @@ describe('claude reviewer runtime guard', () => {
    }
  });

+  it('overrides problematic git config paths so read-only git queries still work', () => {
+    const cwd = createTempRepo('ejclaw-reviewer-readonly-git-');
+    const env = buildReviewerGitGuardEnv(
+      {
+        PATH: process.env.PATH,
+        HOME: cwd,
+        EJCLAW_WORK_DIR: cwd,
+        GIT_CONFIG_GLOBAL: path.join(cwd, '.gitconfig'),
+      },
+      true,
+    );
+
+    expect(env.GIT_CONFIG_GLOBAL).not.toBe(path.join(cwd, '.gitconfig'));
+    expect(() =>
+      execFileSync('git', ['log', '--oneline', '-1'], {
+        cwd,
+        env,
+        encoding: 'utf-8',
+        stdio: ['ignore', 'pipe', 'pipe'],
+      }),
+    ).not.toThrow();
+    expect(fs.existsSync(path.join(cwd, '.gitconfig'))).toBe(false);
+  });
+
  it('accepts a mounted local origin path that resolves as a git repo', () => {
    const originDir = createTempRepo('ejclaw-reviewer-origin-');
    const cwd = createTempRepo('ejclaw-reviewer-workspace-');
--- a/runners/codex-runner/bun.lock
+++ b/runners/codex-runner/bun.lock
@@ -5,7 +5,7 @@
    "": {
      "name": "ejclaw-codex-runner",
      "dependencies": {
-        "@openai/codex": "^0.120.0",
+        "@openai/codex": "^0.124.0",
        "ejclaw-runners-shared": "file:../shared",
      },
      "devDependencies": {
@@ -15,19 +15,19 @@
    },
  },
  "packages": {
-    "@openai/codex": ["@openai/codex@0.120.0", "", { "optionalDependencies": { "@openai/codex-darwin-arm64": "npm:@openai/codex@0.120.0-darwin-arm64", "@openai/codex-darwin-x64": "npm:@openai/codex@0.120.0-darwin-x64", "@openai/codex-linux-arm64": "npm:@openai/codex@0.120.0-linux-arm64", "@openai/codex-linux-x64": "npm:@openai/codex@0.120.0-linux-x64", "@openai/codex-win32-arm64": "npm:@openai/codex@0.120.0-win32-arm64", "@openai/codex-win32-x64": "npm:@openai/codex@0.120.0-win32-x64" }, "bin": { "codex": "bin/codex.js" } }, "sha512-e2P1Gya3dwsRe9IPOiswVz5JfR700u+/sWCqDc3jkqv2QViPkNiBmZoGhFnZL5jBpKakSjehC4/Fpspg70nHTw=="],
+    "@openai/codex": ["@openai/codex@0.124.0", "", { "optionalDependencies": { "@openai/codex-darwin-arm64": "npm:@openai/codex@0.124.0-darwin-arm64", "@openai/codex-darwin-x64": "npm:@openai/codex@0.124.0-darwin-x64", "@openai/codex-linux-arm64": "npm:@openai/codex@0.124.0-linux-arm64", "@openai/codex-linux-x64": "npm:@openai/codex@0.124.0-linux-x64", "@openai/codex-win32-arm64": "npm:@openai/codex@0.124.0-win32-arm64", "@openai/codex-win32-x64": "npm:@openai/codex@0.124.0-win32-x64" }, "bin": { "codex": "bin/codex.js" } }, "sha512-1EVAuPyAQZ8zIVMw3bPJ6a4R8ifLAZ7LGsOyknj5c2he9AFXVRCmWx12WrdZJ25wcBvOEKt1n1Zx+QAj0EVGbQ=="],

-    "@openai/codex-darwin-arm64": ["@openai/codex@0.120.0-darwin-arm64", "", { "os": "darwin", "cpu": "arm64" }, "sha512-7CU+I5kBaMuoqfG3xisq0mUWzxoEHvfu34cB8a0KpBiIhAgu12fKpmYgZ4/DvRP6Wm9Fu6LJYKVF5apUHFp8nQ=="],
+    "@openai/codex-darwin-arm64": ["@openai/codex@0.124.0-darwin-arm64", "", { "os": "darwin", "cpu": "arm64" }, "sha512-lnuqeAdl+RjPWsqVZ8rrPskRIOZmzlyr8e5q/wFVEnMPsm9dWgjRW1PKa84UmDSQVOuz9GObF28DEHFyC4A8NA=="],

-    "@openai/codex-darwin-x64": ["@openai/codex@0.120.0-darwin-x64", "", { "os": "darwin", "cpu": "x64" }, "sha512-d7joNYuwrmd5iIdp/xAE5f8bZT1r82MnmU6Hzgxq3G+xClwEyhxU737ZWnstFSpnZNfxJ5zXCuFUJh4CAkHNtQ=="],
+    "@openai/codex-darwin-x64": ["@openai/codex@0.124.0-darwin-x64", "", { "os": "darwin", "cpu": "x64" }, "sha512-Br+VlL83IOu96Urw+mkZ1PQXLsQXGAYEH6kXNt4ULuVt7qAdQY2FKYwIgLLVLl0vpMk8qWxdfdVMqhiu2uCHlg=="],

-    "@openai/codex-linux-arm64": ["@openai/codex@0.120.0-linux-arm64", "", { "os": "linux", "cpu": "arm64" }, "sha512-sVYY25/URlpZPtb0Q0ryLh+lcq9UTEtHAkdZKa0a/R7mAdyPuhpU9V6jWmxwiUh7s53XZOEVFoKmLfH8YIDWCQ=="],
+    "@openai/codex-linux-arm64": ["@openai/codex@0.124.0-linux-arm64", "", { "os": "linux", "cpu": "arm64" }, "sha512-QXafFVQJxPfU1LSCI5afuHsF5evKAcbQpFuKou0Kl241/HG/zYHdwoWj546zH844gJgegIPAxPf36+RSkUsbbA=="],

-    "@openai/codex-linux-x64": ["@openai/codex@0.120.0-linux-x64", "", { "os": "linux", "cpu": "x64" }, "sha512-VcP9B/c/O+EFEgqoetCzvHrLfAdo8vrt09Gx1lJ8ikewctqAuJ/ozj/6wuvlz7XaaK64ib5cge01pOAeCyt2Sg=="],
+    "@openai/codex-linux-x64": ["@openai/codex@0.124.0-linux-x64", "", { "os": "linux", "cpu": "x64" }, "sha512-cJ4QfQIrz2gkXVWephiGo9JDyD3qLKhvkTTLk7gI8rKd7MZpVXn9/1e7pZCQnJVA7Dk6ocA5G+sxnR78SoIejw=="],

-    "@openai/codex-win32-arm64": ["@openai/codex@0.120.0-win32-arm64", "", { "os": "win32", "cpu": "arm64" }, "sha512-SAaTQU1XHa1qDnmQldmbyROIY5SiaspF+Cw3ziWeeTgyAET3rWusm4ELOElx6QiY1ugYW5ZD+7AFufS2z1xtpQ=="],
+    "@openai/codex-win32-arm64": ["@openai/codex@0.124.0-win32-arm64", "", { "os": "win32", "cpu": "arm64" }, "sha512-oDSI/CQh0kagBwWVPG9EiUBfHiY27ju3LPrjTVZg4VMpVJVNA31JTK8rHMZ6G/2gfNmOl6DMBA6+0ICxSkkshg=="],

-    "@openai/codex-win32-x64": ["@openai/codex@0.120.0-win32-x64", "", { "os": "win32", "cpu": "x64" }, "sha512-zja1GNrbHyOUTvOy5FVMa+rAYIs3m+FOS8rAXftxMEhodMmkMw2O8zcvso657SHhZR0hIEiZ6T70lcyH2YX0mQ=="],
+    "@openai/codex-win32-x64": ["@openai/codex@0.124.0-win32-x64", "", { "os": "win32", "cpu": "x64" }, "sha512-t+lAwmCWwIWQKk6dKaYetRhQsmA+uDmQy1NLka1xAAv3PlNOH8iNz9Lsisr3qYkBR8Tf7tbQlt+pl9tw/A5mfA=="],

    "@types/node": ["@types/node@22.19.15", "", { "dependencies": { "undici-types": "~6.21.0" } }, "sha512-F0R/h2+dsy5wJAUe3tAU6oqa2qbWY5TpNfL/RGmo1y38hiyO1w3x2jPtt76wmuaJI4DQnOBu21cNXQ2STIUUWg=="],

--- a/runners/codex-runner/package.json
+++ b/runners/codex-runner/package.json
@@ -11,7 +11,7 @@
  },
  "dependencies": {
    "ejclaw-runners-shared": "file:../shared",
-    "@openai/codex": "^0.120.0"
+    "@openai/codex": "^0.124.0"
  },
  "devDependencies": {
    "@types/node": "^22.10.7",
--- a/runners/codex-runner/test/reviewer-runtime.test.ts
+++ b/runners/codex-runner/test/reviewer-runtime.test.ts
@@ -75,6 +75,9 @@ describe('codex reviewer runtime guard', () => {
    const env = buildReviewerGitGuardEnv({ PATH: process.env.PATH }, true);
    expect(env.PATH).toContain('ejclaw-reviewer-git-');
    expect(env.EJCLAW_REAL_GIT).toBeTruthy();
+    expect(env.GIT_CONFIG_GLOBAL).toContain('global.gitconfig');
+    expect(env.GIT_CONFIG_NOSYSTEM).toBe('1');
+    expect(fs.existsSync(env.GIT_CONFIG_GLOBAL!)).toBe(true);
  });

  it('prefers an executable HOME-scoped wrapper dir before tmp', () => {
@@ -151,6 +154,30 @@ describe('codex reviewer runtime guard', () => {
    }
  });

+  it('overrides problematic git config paths so read-only git queries still work', () => {
+    const cwd = createTempRepo('ejclaw-reviewer-readonly-git-');
+    const env = buildReviewerGitGuardEnv(
+      {
+        PATH: process.env.PATH,
+        HOME: cwd,
+        EJCLAW_WORK_DIR: cwd,
+        GIT_CONFIG_GLOBAL: path.join(cwd, '.gitconfig'),
+      },
+      true,
+    );
+
+    expect(env.GIT_CONFIG_GLOBAL).not.toBe(path.join(cwd, '.gitconfig'));
+    expect(() =>
+      execFileSync('git', ['log', '--oneline', '-1'], {
+        cwd,
+        env,
+        encoding: 'utf-8',
+        stdio: ['ignore', 'pipe', 'pipe'],
+      }),
+    ).not.toThrow();
+    expect(fs.existsSync(path.join(cwd, '.gitconfig'))).toBe(false);
+  });
+
  it('accepts a mounted local origin path that resolves as a git repo', () => {
    const originDir = createTempRepo('ejclaw-reviewer-origin-');
    const cwd = createTempRepo('ejclaw-reviewer-workspace-');
--- a/runners/shared/src/reviewer-git-guard.ts
+++ b/runners/shared/src/reviewer-git-guard.ts
@@ -78,6 +78,18 @@ function createExecutableWrapperDir(baseEnv: NodeJS.ProcessEnv): string {
  );
 }

+function prepareGitConfigSandbox(wrapperDir: string): {
+  gitConfigGlobalPath: string;
+} {
+  const configDir = path.join(wrapperDir, 'git-config');
+  fs.mkdirSync(configDir, { recursive: true });
+  const gitConfigGlobalPath = path.join(configDir, 'global.gitconfig');
+  if (!fs.existsSync(gitConfigGlobalPath)) {
+    fs.writeFileSync(gitConfigGlobalPath, '');
+  }
+  return { gitConfigGlobalPath };
+}
+
 export function buildReviewerGitGuardEnv(
  baseEnv: NodeJS.ProcessEnv,
  reviewerRuntime: boolean,
@@ -90,6 +102,7 @@ export function buildReviewerGitGuardEnv(
  const protectedWorkDir = baseEnv.EJCLAW_WORK_DIR || '';
  const wrapperDir = createExecutableWrapperDir(baseEnv);
  const wrapperPath = path.join(wrapperDir, 'git');
+  const { gitConfigGlobalPath } = prepareGitConfigSandbox(wrapperDir);
  const blocked = [...BLOCKED_GIT_SUBCOMMANDS]
    .map((value) => `'${value}'`)
    .join(' ');
@@ -179,6 +192,8 @@ exec "$real_git" "$@"
    ...baseEnv,
    EJCLAW_REAL_GIT: realGitPath,
    EJCLAW_PROTECTED_WORK_DIR: protectedWorkDir,
+    GIT_CONFIG_GLOBAL: gitConfigGlobalPath,
+    GIT_CONFIG_NOSYSTEM: '1',
    PATH: `${wrapperDir}:${baseEnv.PATH || ''}`,
  };
 }
--- a/src/db.test.ts
+++ b/src/db.test.ts
@@ -1242,10 +1242,53 @@ describe('paired task state', () => {
    expect(outputs.map((output) => output.turn_number)).toEqual([1, 2]);
    expect(outputs[0].role).toBe('owner');
    expect(outputs[0].output_text).toHaveLength(50_000);
+    expect(outputs[0].verdict).toBe('continue');
    expect(outputs[1].output_text).toBe('review turn');
+    expect(outputs[1].verdict).toBe('continue');
    expect(getLatestTurnNumber('paired-task-turn-output')).toBe(2);
  });

+  it('stores the parsed visible verdict with paired turn outputs', () => {
+    createPairedTask({
+      id: 'paired-task-turn-output-verdict',
+      chat_jid: 'dc:paired',
+      group_folder: 'paired-room',
+      owner_service_id: 'codex-main',
+      reviewer_service_id: 'codex-review',
+      title: null,
+      source_ref: null,
+      plan_notes: null,
+      round_trip_count: 0,
+      review_requested_at: null,
+      status: 'active',
+      arbiter_verdict: null,
+      arbiter_requested_at: null,
+      completion_reason: null,
+      created_at: '2026-03-28T00:00:00.000Z',
+      updated_at: '2026-03-28T00:00:00.000Z',
+    });
+
+    insertPairedTurnOutput(
+      'paired-task-turn-output-verdict',
+      1,
+      'owner',
+      'STEP_DONE\n1단계 완료',
+    );
+    insertPairedTurnOutput(
+      'paired-task-turn-output-verdict',
+      2,
+      'owner',
+      'TASK_DONE\n요청 범위 전체 완료',
+    );
+
+    const outputs = getPairedTurnOutputs('paired-task-turn-output-verdict');
+
+    expect(outputs.map((output) => output.verdict)).toEqual([
+      'step_done',
+      'task_done',
+    ]);
+  });
+
  it('preserves explicit created_at when inserting a paired turn output', () => {
    createPairedTask({
      id: 'paired-task-turn-output-created-at',
--- a/src/db/base-schema.ts
+++ b/src/db/base-schema.ts
@@ -119,6 +119,10 @@ export function applyBaseSchema(database: Database): void {
      review_requested_at TEXT,
      round_trip_count INTEGER NOT NULL DEFAULT 0,
      owner_failure_count INTEGER NOT NULL DEFAULT 0,
+      owner_step_done_streak INTEGER NOT NULL DEFAULT 0,
+      finalize_step_done_count INTEGER NOT NULL DEFAULT 0,
+      task_done_then_user_reopen_count INTEGER NOT NULL DEFAULT 0,
+      empty_step_done_streak INTEGER NOT NULL DEFAULT 0,
      status TEXT NOT NULL DEFAULT 'active',
      arbiter_verdict TEXT,
      arbiter_requested_at TEXT,
@@ -154,6 +158,7 @@ export function applyBaseSchema(database: Database): void {
      turn_number INTEGER NOT NULL,
      role TEXT NOT NULL,
      output_text TEXT NOT NULL,
+      verdict TEXT,
      created_at TEXT NOT NULL,
      UNIQUE(task_id, turn_number, role)
    );
--- a/src/db/bootstrap.test.ts
+++ b/src/db/bootstrap.test.ts
@@ -36,6 +36,7 @@ function getExpectedSchemaMigrations(): Array<{
    { version: 9, name: 'paired_workspace_project_schema_cleanup' },
    { version: 10, name: 'paired_turn_provenance_upgrade' },
    { version: 11, name: 'owner_failure_count' },
+    { version: 12, name: 'paired_verdict_and_step_telemetry' },
  ];
 }

--- a/src/db/migrations/012_paired-verdict-and-step-telemetry.ts
+++ b/src/db/migrations/012_paired-verdict-and-step-telemetry.ts
@@ -0,0 +1,75 @@
+import type { Database } from 'bun:sqlite';
+
+import { tableHasColumn } from './helpers.js';
+import type { SchemaMigrationDefinition } from './types.js';
+
+export const PAIRED_VERDICT_AND_STEP_TELEMETRY_MIGRATION: SchemaMigrationDefinition =
+  {
+    version: 12,
+    name: 'paired_verdict_and_step_telemetry',
+    apply(database: Database) {
+      if (!tableHasColumn(database, 'paired_turn_outputs', 'verdict')) {
+        database.exec(`
+        ALTER TABLE paired_turn_outputs
+        ADD COLUMN verdict TEXT
+      `);
+      }
+
+      if (!tableHasColumn(database, 'paired_tasks', 'owner_step_done_streak')) {
+        database.exec(`
+        ALTER TABLE paired_tasks
+        ADD COLUMN owner_step_done_streak INTEGER NOT NULL DEFAULT 0
+      `);
+      }
+
+      if (
+        !tableHasColumn(database, 'paired_tasks', 'finalize_step_done_count')
+      ) {
+        database.exec(`
+        ALTER TABLE paired_tasks
+        ADD COLUMN finalize_step_done_count INTEGER NOT NULL DEFAULT 0
+      `);
+      }
+
+      if (
+        !tableHasColumn(
+          database,
+          'paired_tasks',
+          'task_done_then_user_reopen_count',
+        )
+      ) {
+        database.exec(`
+        ALTER TABLE paired_tasks
+        ADD COLUMN task_done_then_user_reopen_count INTEGER NOT NULL DEFAULT 0
+      `);
+      }
+
+      if (!tableHasColumn(database, 'paired_tasks', 'empty_step_done_streak')) {
+        database.exec(`
+        ALTER TABLE paired_tasks
+        ADD COLUMN empty_step_done_streak INTEGER NOT NULL DEFAULT 0
+      `);
+      }
+
+      database.exec(`
+      UPDATE paired_tasks
+         SET owner_step_done_streak = 0
+       WHERE owner_step_done_streak IS NULL
+    `);
+      database.exec(`
+      UPDATE paired_tasks
+         SET finalize_step_done_count = 0
+       WHERE finalize_step_done_count IS NULL
+    `);
+      database.exec(`
+      UPDATE paired_tasks
+         SET task_done_then_user_reopen_count = 0
+       WHERE task_done_then_user_reopen_count IS NULL
+    `);
+      database.exec(`
+      UPDATE paired_tasks
+         SET empty_step_done_streak = 0
+       WHERE empty_step_done_streak IS NULL
+    `);
+    },
+  };
--- a/src/db/migrations/index.ts
+++ b/src/db/migrations/index.ts
@@ -11,6 +11,7 @@ import { PAIRED_TASK_SCHEMA_CLEANUP_MIGRATION } from './008_paired-task-schema-c
 import { PAIRED_WORKSPACE_PROJECT_SCHEMA_CLEANUP_MIGRATION } from './009_paired-workspace-project-schema-cleanup.js';
 import { PAIRED_TURN_PROVENANCE_UPGRADE_MIGRATION } from './010_paired-turn-provenance-upgrade.js';
 import { OWNER_FAILURE_COUNT_MIGRATION } from './011_owner-failure-count.js';
+import { PAIRED_VERDICT_AND_STEP_TELEMETRY_MIGRATION } from './012_paired-verdict-and-step-telemetry.js';
 import type {
  SchemaMigrationArgs,
  SchemaMigrationDefinition,
@@ -30,6 +31,7 @@ const ORDERED_SCHEMA_MIGRATIONS: readonly SchemaMigrationDefinition[] = [
  PAIRED_WORKSPACE_PROJECT_SCHEMA_CLEANUP_MIGRATION,
  PAIRED_TURN_PROVENANCE_UPGRADE_MIGRATION,
  OWNER_FAILURE_COUNT_MIGRATION,
+  PAIRED_VERDICT_AND_STEP_TELEMETRY_MIGRATION,
 ];

 function ensureSchemaMigrationsTable(database: Database): void {
--- a/src/db/paired-state.ts
+++ b/src/db/paired-state.ts
@@ -51,6 +51,10 @@ export type PairedTaskUpdates = Partial<
    | 'review_requested_at'
    | 'round_trip_count'
    | 'owner_failure_count'
+    | 'owner_step_done_streak'
+    | 'finalize_step_done_count'
+    | 'task_done_then_user_reopen_count'
+    | 'empty_step_done_streak'
    | 'status'
    | 'arbiter_verdict'
    | 'arbiter_requested_at'
@@ -173,6 +177,10 @@ export function createPairedTaskInDatabase(
          review_requested_at,
          round_trip_count,
          owner_failure_count,
+          owner_step_done_streak,
+          finalize_step_done_count,
+          task_done_then_user_reopen_count,
+          empty_step_done_streak,
          status,
          arbiter_verdict,
          arbiter_requested_at,
@@ -180,7 +188,7 @@ export function createPairedTaskInDatabase(
          created_at,
          updated_at
        )
-        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
      `,
    )
    .run(
@@ -198,6 +206,10 @@ export function createPairedTaskInDatabase(
      task.review_requested_at,
      task.round_trip_count,
      task.owner_failure_count ?? 0,
+      task.owner_step_done_streak ?? 0,
+      task.finalize_step_done_count ?? 0,
+      task.task_done_then_user_reopen_count ?? 0,
+      task.empty_step_done_streak ?? 0,
      task.status,
      task.arbiter_verdict,
      task.arbiter_requested_at,
@@ -306,6 +318,22 @@ export function updatePairedTaskInDatabase(
    fields.push('owner_failure_count = ?');
    values.push(updates.owner_failure_count);
  }
+  if (updates.owner_step_done_streak !== undefined) {
+    fields.push('owner_step_done_streak = ?');
+    values.push(updates.owner_step_done_streak);
+  }
+  if (updates.finalize_step_done_count !== undefined) {
+    fields.push('finalize_step_done_count = ?');
+    values.push(updates.finalize_step_done_count);
+  }
+  if (updates.task_done_then_user_reopen_count !== undefined) {
+    fields.push('task_done_then_user_reopen_count = ?');
+    values.push(updates.task_done_then_user_reopen_count);
+  }
+  if (updates.empty_step_done_streak !== undefined) {
+    fields.push('empty_step_done_streak = ?');
+    values.push(updates.empty_step_done_streak);
+  }
  if (updates.status !== undefined) {
    fields.push('status = ?');
    values.push(updates.status);
@@ -369,6 +397,22 @@ export function updatePairedTaskIfUnchangedInDatabase(
    fields.push('owner_failure_count = ?');
    values.push(updates.owner_failure_count);
  }
+  if (updates.owner_step_done_streak !== undefined) {
+    fields.push('owner_step_done_streak = ?');
+    values.push(updates.owner_step_done_streak);
+  }
+  if (updates.finalize_step_done_count !== undefined) {
+    fields.push('finalize_step_done_count = ?');
+    values.push(updates.finalize_step_done_count);
+  }
+  if (updates.task_done_then_user_reopen_count !== undefined) {
+    fields.push('task_done_then_user_reopen_count = ?');
+    values.push(updates.task_done_then_user_reopen_count);
+  }
+  if (updates.empty_step_done_streak !== undefined) {
+    fields.push('empty_step_done_streak = ?');
+    values.push(updates.empty_step_done_streak);
+  }
  if (updates.status !== undefined) {
    fields.push('status = ?');
    values.push(updates.status);
--- a/src/db/paired-turn-outputs.ts
+++ b/src/db/paired-turn-outputs.ts
@@ -1,6 +1,7 @@
 import { Database } from 'bun:sqlite';

 import { logger } from '../logger.js';
+import { parseVisibleVerdict } from '../paired-verdict.js';
 import { PairedRoomRole, PairedTurnOutput } from '../types.js';

 const MAX_TURN_OUTPUT_CHARS = 50_000;
@@ -29,14 +30,15 @@ export function insertPairedTurnOutputInDatabase(
  database
    .prepare(
      `INSERT OR REPLACE INTO paired_turn_outputs
-         (task_id, turn_number, role, output_text, created_at)
-       VALUES (?, ?, ?, ?, ?)`,
+         (task_id, turn_number, role, output_text, verdict, created_at)
+       VALUES (?, ?, ?, ?, ?, ?)`,
    )
    .run(
      taskId,
      turnNumber,
      role,
      outputText.slice(0, MAX_TURN_OUTPUT_CHARS),
+      parseVisibleVerdict(outputText),
      createdAt ?? new Date().toISOString(),
    );
 }
--- a/src/message-agent-executor-paired.ts
+++ b/src/message-agent-executor-paired.ts
@@ -14,6 +14,7 @@ import {
  completePairedExecutionContext,
  type PreparedPairedExecutionContext,
 } from './paired-execution-context.js';
+import { parseVisibleVerdict } from './paired-execution-context-shared.js';
 import { resolvePairedFollowUpQueueAction } from './message-agent-executor-rules.js';
 import { enqueuePairedFollowUpAfterEvent } from './message-runtime-follow-up.js';
 import type { PairedTurnIdentity } from './paired-turn-identity.js';
@@ -438,6 +439,7 @@ export function createPairedExecutionLifecycle(args: {
              executionStatus: effectiveStatus,
              sawOutput: sawOutputForFollowUp,
              taskStatus: finishedTask?.status ?? null,
+              outputSummary: pairedExecutionSummary,
            });
      if (queueAction !== 'pending' || !finishedTask) {
        return;
@@ -452,6 +454,10 @@ export function createPairedExecutionLifecycle(args: {
        executionStatus: effectiveStatus,
        sawOutput: sawOutputForFollowUp,
        fallbackLastTurnOutputRole: sawOutputForFollowUp ? completedRole : null,
+        fallbackLastTurnOutputVerdict:
+          sawOutputForFollowUp && pairedExecutionSummary
+            ? parseVisibleVerdict(pairedExecutionSummary)
+            : null,
        enqueueMessageCheck,
      });
      if (followUpResult.kind !== 'paired-follow-up') {
--- a/src/message-agent-executor-rules.test.ts
+++ b/src/message-agent-executor-rules.test.ts
@@ -216,6 +216,18 @@ describe('message-agent-executor-rules', () => {
    ).toBe('none');
  });

+  it('does not request a duplicate executor-side follow-up after successful owner STEP_DONE output', () => {
+    expect(
+      resolvePairedFollowUpQueueAction({
+        completedRole: 'owner',
+        executionStatus: 'succeeded',
+        sawOutput: true,
+        taskStatus: 'active',
+        outputSummary: 'STEP_DONE\nkeep going',
+      }),
+    ).toBe('none');
+  });
+
  it('returns none after successful output when no next-turn action is needed', () => {
    expect(
      resolvePairedFollowUpQueueAction({
--- a/src/message-agent-executor-rules.ts
+++ b/src/message-agent-executor-rules.ts
@@ -2,6 +2,7 @@ import {
  resolveFollowUpDispatch,
  resolveNextTurnAction,
 } from './message-runtime-rules.js';
+import { parseVisibleVerdict } from './paired-execution-context-shared.js';
 import type { PairedRoomRole, PairedTaskStatus } from './types.js';
 export {
  isRetryableClaudeSessionFailureAttempt,
@@ -20,10 +21,15 @@ export function resolvePairedFollowUpQueueAction(args: {
  executionStatus: 'succeeded' | 'failed';
  sawOutput: boolean;
  taskStatus: PairedTaskStatus | null;
+  outputSummary?: string | null;
 }): PairedFollowUpQueueAction {
  const nextTurnAction = resolveNextTurnAction({
    taskStatus: args.taskStatus,
    lastTurnOutputRole: args.sawOutput ? args.completedRole : null,
+    lastTurnOutputVerdict:
+      args.sawOutput && args.outputSummary
+        ? parseVisibleVerdict(args.outputSummary)
+        : null,
  });
  const dispatch = resolveFollowUpDispatch({
    source: 'executor-recovery',
--- a/src/message-runtime-flow.ts
+++ b/src/message-runtime-flow.ts
@@ -24,6 +24,7 @@ import {
 } from './message-runtime-rules.js';
 import type { ExecuteTurnFn } from './message-runtime-types.js';
 import { buildPairedTurnIdentity } from './paired-turn-identity.js';
+import { resolveStoredVisibleVerdict } from './paired-verdict.js';
 import {
  claimPairedTurnExecution,
  type ScheduledPairedFollowUpIntentKind,
@@ -127,6 +128,10 @@ export function buildPendingPairedTurn(args: {
  const nextTurnAction = resolveNextTurnAction({
    taskStatus,
    lastTurnOutputRole: lastTurnOutput?.role ?? null,
+    lastTurnOutputVerdict: resolveStoredVisibleVerdict({
+      verdict: lastTurnOutput?.verdict ?? null,
+      outputText: lastTurnOutput?.output_text ?? null,
+    }),
  });
  const recentMessages = getRecentChatMessages(chatJid, 20);
  const lastHumanMessage = getLastHumanMessageContent(chatJid);
--- a/src/message-runtime-follow-up.test.ts
+++ b/src/message-runtime-follow-up.test.ts
@@ -83,6 +83,33 @@ describe('message-runtime-follow-up', () => {
    expect(enqueue).toHaveBeenCalledTimes(1);
  });

+  it('does not use fallback owner STEP_DONE verdict to schedule owner follow-ups while active', () => {
+    const enqueue = vi.fn();
+    const result = dispatchPairedFollowUpForEvent({
+      chatJid: 'group@test',
+      runId: 'run-owner-step-done',
+      task: {
+        id: 'task-owner-step-done',
+        status: 'active',
+        round_trip_count: 1,
+        updated_at: '2026-03-30T00:00:00.000Z',
+      },
+      source: 'owner-delivery-success',
+      completedRole: 'owner',
+      fallbackLastTurnOutputRole: 'owner',
+      fallbackLastTurnOutputVerdict: 'step_done',
+      enqueue,
+    });
+
+    expect(result).toMatchObject({
+      kind: 'none',
+      taskStatus: 'active',
+      lastTurnOutputRole: 'owner',
+      lastTurnOutputVerdict: 'step_done',
+    });
+    expect(enqueue).not.toHaveBeenCalled();
+  });
+
  it('prefers the latest persisted turn output over the fallback delivery role when both are present', () => {
    const enqueue = vi.fn();
    vi.mocked(getPairedTurnOutputs).mockReturnValue([
--- a/src/message-runtime-follow-up.ts
+++ b/src/message-runtime-follow-up.ts
@@ -1,4 +1,6 @@
 import { getPairedTurnOutputs } from './db.js';
+import { type VisibleVerdict } from './paired-execution-context-shared.js';
+import { resolveStoredVisibleVerdict } from './paired-verdict.js';
 import {
  matchesExpectedPairedFollowUpIntent,
  resolveFollowUpDispatch,
@@ -20,6 +22,7 @@ export interface PairedFollowUpDecision {
  taskId: string | null;
  taskStatus: PairedTaskStatus | null;
  lastTurnOutputRole: PairedRoomRole | null;
+  lastTurnOutputVerdict: VisibleVerdict | null;
  nextTurnAction: NextTurnAction;
  dispatch: FollowUpDispatch;
 }
@@ -37,15 +40,27 @@ export type PairedFollowUpDispatchResult =
      scheduled: boolean;
    });

-export function resolveLatestPairedTurnOutputRole(args: {
+export function resolveLatestPairedTurnOutputContext(args: {
  task: Pick<PairedTask, 'id'> | null | undefined;
  fallbackLastTurnOutputRole?: PairedRoomRole | null;
-}): PairedRoomRole | null {
-  return (
-    (args.task ? getPairedTurnOutputs(args.task.id).at(-1)?.role : null) ??
-    args.fallbackLastTurnOutputRole ??
-    null
-  );
+  fallbackLastTurnOutputVerdict?: VisibleVerdict | null;
+}): {
+  role: PairedRoomRole | null;
+  verdict: VisibleVerdict | null;
+} {
+  const latestOutput = args.task
+    ? getPairedTurnOutputs(args.task.id).at(-1)
+    : null;
+  return {
+    role: latestOutput?.role ?? args.fallbackLastTurnOutputRole ?? null,
+    verdict:
+      resolveStoredVisibleVerdict({
+        verdict: latestOutput?.verdict ?? null,
+        outputText: latestOutput?.output_text ?? null,
+      }) ??
+      args.fallbackLastTurnOutputVerdict ??
+      null,
+  };
 }

 export function resolvePairedFollowUpDecision(args: {
@@ -55,14 +70,18 @@ export function resolvePairedFollowUpDecision(args: {
  executionStatus?: 'succeeded' | 'failed';
  sawOutput?: boolean;
  fallbackLastTurnOutputRole?: PairedRoomRole | null;
+  fallbackLastTurnOutputVerdict?: VisibleVerdict | null;
 }): PairedFollowUpDecision {
-  const lastTurnOutputRole = resolveLatestPairedTurnOutputRole({
-    task: args.task,
-    fallbackLastTurnOutputRole: args.fallbackLastTurnOutputRole,
-  });
+  const { role: lastTurnOutputRole, verdict: lastTurnOutputVerdict } =
+    resolveLatestPairedTurnOutputContext({
+      task: args.task,
+      fallbackLastTurnOutputRole: args.fallbackLastTurnOutputRole,
+      fallbackLastTurnOutputVerdict: args.fallbackLastTurnOutputVerdict,
+    });
  const nextTurnAction = resolveNextTurnAction({
    taskStatus: args.task?.status ?? null,
    lastTurnOutputRole,
+    lastTurnOutputVerdict,
  });
  const dispatch = resolveFollowUpDispatch({
    source: args.source,
@@ -76,6 +95,7 @@ export function resolvePairedFollowUpDecision(args: {
    taskId: args.task?.id ?? null,
    taskStatus: args.task?.status ?? null,
    lastTurnOutputRole,
+    lastTurnOutputVerdict,
    nextTurnAction,
    dispatch,
  };
@@ -91,23 +111,31 @@ export function schedulePairedFollowUpIntent(args: {
  intentKind: ScheduledPairedFollowUpIntentKind;
  enqueue: () => void;
  fallbackLastTurnOutputRole?: PairedRoomRole | null;
+  fallbackLastTurnOutputVerdict?: VisibleVerdict | null;
  lastTurnOutputRole?: PairedRoomRole | null;
+  lastTurnOutputVerdict?: VisibleVerdict | null;
 }): boolean {
  if (!args.task) {
    return false;
  }

-  const lastTurnOutputRole =
-    args.lastTurnOutputRole ??
-    resolveLatestPairedTurnOutputRole({
-      task: args.task,
-      fallbackLastTurnOutputRole: args.fallbackLastTurnOutputRole,
-    });
+  const latestOutputContext =
+    args.lastTurnOutputRole != null || args.lastTurnOutputVerdict != null
+      ? {
+          role: args.lastTurnOutputRole ?? null,
+          verdict: args.lastTurnOutputVerdict ?? null,
+        }
+      : resolveLatestPairedTurnOutputContext({
+          task: args.task,
+          fallbackLastTurnOutputRole: args.fallbackLastTurnOutputRole,
+          fallbackLastTurnOutputVerdict: args.fallbackLastTurnOutputVerdict,
+        });

  if (
    !matchesExpectedPairedFollowUpIntent({
      taskStatus: args.task.status,
-      lastTurnOutputRole,
+      lastTurnOutputRole: latestOutputContext.role,
+      lastTurnOutputVerdict: latestOutputContext.verdict,
      intentKind: args.intentKind,
    })
  ) {
@@ -133,7 +161,9 @@ export function schedulePairedFollowUpWithMessageCheck(args: {
  intentKind: ScheduledPairedFollowUpIntentKind;
  enqueueMessageCheck: () => void;
  fallbackLastTurnOutputRole?: PairedRoomRole | null;
+  fallbackLastTurnOutputVerdict?: VisibleVerdict | null;
  lastTurnOutputRole?: PairedRoomRole | null;
+  lastTurnOutputVerdict?: VisibleVerdict | null;
 }): boolean {
  return schedulePairedFollowUpIntent({
    chatJid: args.chatJid,
@@ -142,7 +172,9 @@ export function schedulePairedFollowUpWithMessageCheck(args: {
    intentKind: args.intentKind,
    enqueue: args.enqueueMessageCheck,
    fallbackLastTurnOutputRole: args.fallbackLastTurnOutputRole,
+    fallbackLastTurnOutputVerdict: args.fallbackLastTurnOutputVerdict,
    lastTurnOutputRole: args.lastTurnOutputRole,
+    lastTurnOutputVerdict: args.lastTurnOutputVerdict,
  });
 }

@@ -158,6 +190,7 @@ export function dispatchPairedFollowUpForEvent(args: {
  executionStatus?: 'succeeded' | 'failed';
  sawOutput?: boolean;
  fallbackLastTurnOutputRole?: PairedRoomRole | null;
+  fallbackLastTurnOutputVerdict?: VisibleVerdict | null;
  enqueue: () => void;
  enqueueMessageCheck?: () => void;
 }): PairedFollowUpDispatchResult {
@@ -168,6 +201,7 @@ export function dispatchPairedFollowUpForEvent(args: {
    executionStatus: args.executionStatus,
    sawOutput: args.sawOutput,
    fallbackLastTurnOutputRole: args.fallbackLastTurnOutputRole,
+    fallbackLastTurnOutputVerdict: args.fallbackLastTurnOutputVerdict,
  });

  if (
@@ -193,6 +227,7 @@ export function dispatchPairedFollowUpForEvent(args: {
      intentKind: decision.nextTurnAction.kind,
      enqueue: args.enqueue,
      lastTurnOutputRole: decision.lastTurnOutputRole,
+      lastTurnOutputVerdict: decision.lastTurnOutputVerdict,
    });

    return {
@@ -221,6 +256,7 @@ export function enqueuePairedFollowUpAfterEvent(args: {
  executionStatus?: 'succeeded' | 'failed';
  sawOutput?: boolean;
  fallbackLastTurnOutputRole?: PairedRoomRole | null;
+  fallbackLastTurnOutputVerdict?: VisibleVerdict | null;
  enqueueMessageCheck: () => void;
 }): PairedFollowUpDispatchResult {
  return dispatchPairedFollowUpForEvent({
@@ -232,6 +268,7 @@ export function enqueuePairedFollowUpAfterEvent(args: {
    executionStatus: args.executionStatus,
    sawOutput: args.sawOutput,
    fallbackLastTurnOutputRole: args.fallbackLastTurnOutputRole,
+    fallbackLastTurnOutputVerdict: args.fallbackLastTurnOutputVerdict,
    enqueue: args.enqueueMessageCheck,
    enqueueMessageCheck: args.enqueueMessageCheck,
  });
--- a/src/message-runtime-prompts.ts
+++ b/src/message-runtime-prompts.ts
@@ -212,7 +212,8 @@ export function buildFinalizePendingPrompt(args: {
    ? `\n\nReviewer's final assessment:\n${lastReviewerOutput.output_text.slice(0, 2000)}`
    : '';

-  return `The reviewer approved your work (DONE). Finalize and report the result.
-If you intend to close this paired turn now, your first line must be DONE.
+  return `The reviewer approved the current task scope (TASK_DONE / legacy DONE). Finalize and report the result.
+If you intend to close this paired turn now, your first line must be TASK_DONE.
+If the original request still has remaining work and the owner flow should continue, your first line may be STEP_DONE.
 If your first line is DONE_WITH_CONCERNS, the system will reopen review instead of finishing.${reviewerSummary}`;
 }
--- a/src/message-runtime-queue.ts
+++ b/src/message-runtime-queue.ts
@@ -11,6 +11,7 @@ import {
  isBotOnlyPairedRoomTurn,
 } from './message-runtime-flow.js';
 import { buildPairedTurnIdentity } from './paired-turn-identity.js';
+import { resolveStoredVisibleVerdict } from './paired-verdict.js';
 import {
  advanceLastAgentCursor,
  resolveActiveRole,
@@ -183,6 +184,12 @@ export async function runQueuedGroupTurn(args: {
  const lastTurnOutputRole = currentTask
    ? (turnOutputs.at(-1)?.role ?? null)
    : null;
+  const lastTurnOutputVerdict = currentTask
+    ? resolveStoredVisibleVerdict({
+        verdict: turnOutputs.at(-1)?.verdict ?? null,
+        outputText: turnOutputs.at(-1)?.output_text ?? null,
+      })
+    : null;
  const turnRole = currentTask
    ? hasHumanMsg
      ? resolveQueuedTurnRole({
@@ -193,6 +200,7 @@ export async function runQueuedGroupTurn(args: {
          taskStatus,
          hasHumanMessage: false,
          lastTurnOutputRole,
+          lastTurnOutputVerdict,
        })
    : 'owner';
  if (!turnRole) {
--- a/src/message-runtime-rules.test.ts
+++ b/src/message-runtime-rules.test.ts
@@ -113,6 +113,14 @@ describe('message-runtime-rules', () => {
        intentKind: 'owner-follow-up',
      }),
    ).toBe(true);
+    expect(
+      matchesExpectedPairedFollowUpIntent({
+        taskStatus: 'active',
+        lastTurnOutputRole: 'owner',
+        lastTurnOutputVerdict: 'step_done',
+        intentKind: 'owner-follow-up',
+      }),
+    ).toBe(false);
  });

  it('maps active tasks with reviewer output to an owner follow-up', () => {
@@ -124,6 +132,16 @@ describe('message-runtime-rules', () => {
    ).toEqual({ kind: 'owner-follow-up' });
  });

+  it('does not map active owner STEP_DONE output to an owner follow-up', () => {
+    expect(
+      resolveNextTurnAction({
+        taskStatus: 'active',
+        lastTurnOutputRole: 'owner',
+        lastTurnOutputVerdict: 'step_done',
+      }),
+    ).toEqual({ kind: 'none' });
+  });
+
  it('returns none when an active task has no reviewer or arbiter handoff output', () => {
    expect(
      resolveNextTurnAction({
@@ -152,6 +170,15 @@ describe('message-runtime-rules', () => {
      kind: 'enqueue',
      queueKind: 'paired-follow-up',
    });
+    expect(
+      resolveFollowUpDispatch({
+        source: 'owner-delivery-success',
+        nextTurnAction: { kind: 'owner-follow-up' },
+      }),
+    ).toEqual({
+      kind: 'enqueue',
+      queueKind: 'paired-follow-up',
+    });
    expect(
      resolveFollowUpDispatch({
        source: 'owner-delivery-success',
@@ -295,6 +322,14 @@ describe('message-runtime-rules', () => {
        lastTurnOutputRole: 'owner',
      }),
    ).toBe('reviewer');
+    expect(
+      resolveQueuedPairedTurnRole({
+        taskStatus: 'active',
+        hasHumanMessage: false,
+        lastTurnOutputRole: 'owner',
+        lastTurnOutputVerdict: 'step_done',
+      }),
+    ).toBeNull();
  });

  it('resolves reviewer execution target from review_ready task status', () => {
--- a/src/message-runtime-rules.ts
+++ b/src/message-runtime-rules.ts
@@ -4,6 +4,7 @@ import { normalizeStoredSeqCursor } from './message-cursor.js';
 import { isTriggerAllowed, loadSenderAllowlist } from './sender-allowlist.js';
 import { isTaskStatusControlMessage } from './task-watch-status.js';
 import { ARBITER_AGENT_TYPE, REVIEWER_AGENT_TYPE } from './config.js';
+import type { VisibleVerdict } from './paired-execution-context-shared.js';
 import {
  hasReviewerLease,
  resolveLeaseServiceId,
@@ -79,6 +80,7 @@ export function resolveQueuedPairedTurnRole(args: {
  taskStatus?: PairedTaskStatus | null;
  hasHumanMessage: boolean;
  lastTurnOutputRole?: PairedRoomRole | null;
+  lastTurnOutputVerdict?: VisibleVerdict | null;
 }): 'owner' | 'reviewer' | 'arbiter' | null {
  if (args.hasHumanMessage) {
    return resolveQueuedTurnRole({
@@ -90,6 +92,7 @@ export function resolveQueuedPairedTurnRole(args: {
  const nextTurnAction = resolveNextTurnAction({
    taskStatus: args.taskStatus,
    lastTurnOutputRole: args.lastTurnOutputRole,
+    lastTurnOutputVerdict: args.lastTurnOutputVerdict,
  });

  switch (nextTurnAction.kind) {
@@ -125,6 +128,7 @@ export type FollowUpDispatch =
 export function resolveNextTurnAction(args: {
  taskStatus?: PairedTaskStatus | null;
  lastTurnOutputRole?: PairedRoomRole | null;
+  lastTurnOutputVerdict?: VisibleVerdict | null;
 }): NextTurnAction {
  switch (args.taskStatus) {
    case 'review_ready':
@@ -154,12 +158,14 @@ export function resolveNextTurnAction(args: {
 export function matchesExpectedPairedFollowUpIntent(args: {
  taskStatus?: PairedTaskStatus | null;
  lastTurnOutputRole?: PairedRoomRole | null;
+  lastTurnOutputVerdict?: VisibleVerdict | null;
  intentKind: ScheduledNextTurnActionKind;
 }): boolean {
  return (
    resolveNextTurnAction({
      taskStatus: args.taskStatus,
      lastTurnOutputRole: args.lastTurnOutputRole,
+      lastTurnOutputVerdict: args.lastTurnOutputVerdict,
    }).kind === args.intentKind
  );
 }
@@ -184,7 +190,8 @@ export function resolveFollowUpDispatch(args: {
        args.completedRole === 'owner'
      ) {
        return args.nextTurnAction.kind === 'reviewer-turn' ||
-          args.nextTurnAction.kind === 'arbiter-turn'
+          args.nextTurnAction.kind === 'arbiter-turn' ||
+          args.nextTurnAction.kind === 'owner-follow-up'
          ? { kind: 'enqueue', queueKind: 'paired-follow-up' }
          : { kind: 'none' };
      }
@@ -219,6 +226,7 @@ export function resolveFollowUpDispatch(args: {
        return { kind: 'none' };
      }
      return args.nextTurnAction.kind === 'reviewer-turn' ||
+        args.nextTurnAction.kind === 'owner-follow-up' ||
        args.nextTurnAction.kind === 'arbiter-turn' ||
        args.nextTurnAction.kind === 'finalize-owner-turn'
        ? { kind: 'enqueue', queueKind: 'paired-follow-up' }
--- a/src/message-runtime.test.ts
+++ b/src/message-runtime.test.ts
@@ -842,6 +842,127 @@ describe('createMessageRuntime', () => {
    );
  });

+  it('suppresses a stale owner work item when a new human message arrives while the paired task is still active', async () => {
+    const chatJid = 'group@test';
+    const group = makeGroup('codex');
+    const channel = makeChannel(chatJid);
+    const enqueueMessageCheck = vi.fn();
+    const activeTask = {
+      id: 'task-active-owner-follow-up',
+      chat_jid: chatJid,
+      group_folder: group.folder,
+      owner_service_id: 'claude',
+      reviewer_service_id: 'codex-main',
+      title: null,
+      source_ref: 'HEAD',
+      plan_notes: null,
+      review_requested_at: null,
+      round_trip_count: 0,
+      status: 'active',
+      arbiter_verdict: null,
+      arbiter_requested_at: null,
+      completion_reason: null,
+      created_at: '2026-03-30T00:00:00.000Z',
+      updated_at: '2026-03-30T00:00:05.000Z',
+    } as any;
+
+    vi.mocked(serviceRouting.hasReviewerLease).mockReturnValue(true);
+    vi.mocked(db.getOpenWorkItem).mockReturnValue({
+      id: 109,
+      group_folder: group.folder,
+      chat_jid: chatJid,
+      agent_type: 'codex',
+      service_id: 'claude',
+      delivery_role: 'owner',
+      status: 'delivery_retry',
+      start_seq: 1,
+      end_seq: 1,
+      result_payload: '이전 step 결과입니다.',
+      delivery_attempts: 1,
+      created_at: '2026-03-30T00:00:05.000Z',
+      updated_at: '2026-03-30T00:00:05.000Z',
+      delivered_at: null,
+      delivery_message_id: null,
+      last_error: 'discord send failed',
+    });
+    vi.mocked(db.getMessagesSince).mockReturnValue([
+      {
+        id: 'msg-2',
+        chat_jid: chatJid,
+        sender: 'user@test',
+        sender_name: 'User',
+        content: '이전 답 말고 이 방향으로 진행해줘',
+        timestamp: '2026-03-30T00:00:10.000Z',
+        seq: 2,
+      },
+    ]);
+    vi.mocked(db.getLatestOpenPairedTaskForChat).mockReturnValue(activeTask);
+    vi.mocked(agentRunner.runAgentProcess).mockImplementation(
+      async (_group, input, _onProcess, onOutput) => {
+        expect(input.prompt).toContain('이 방향으로 진행해줘');
+        expect(input.prompt).not.toContain('이전 step 결과입니다.');
+        await onOutput?.({
+          status: 'success',
+          phase: 'final',
+          result: 'STEP_DONE\n새 입력 기준으로 계속 진행',
+          output: {
+            visibility: 'public',
+            text: 'STEP_DONE\n새 입력 기준으로 계속 진행',
+          },
+        } as any);
+        return {
+          status: 'success',
+          result: 'STEP_DONE\n새 입력 기준으로 계속 진행',
+        };
+      },
+    );
+
+    const runtime = createMessageRuntime({
+      assistantName: 'Andy',
+      idleTimeout: 1_000,
+      pollInterval: 1_000,
+      timezone: 'UTC',
+      triggerPattern: /^@Andy\b/i,
+      channels: [channel],
+      queue: {
+        registerProcess: vi.fn(),
+        closeStdin: vi.fn(),
+        notifyIdle: vi.fn(),
+        enqueueMessageCheck,
+      } as any,
+      getRoomBindings: () => ({ [chatJid]: group }),
+      getSessions: () => ({}),
+      getLastTimestamp: () => '',
+      setLastTimestamp: vi.fn(),
+      getLastAgentTimestamps: () => ({}),
+      saveState: vi.fn(),
+      persistSession: vi.fn(),
+      clearSession: vi.fn(),
+    });
+
+    const result = await runtime.processGroupMessages(chatJid, {
+      runId: 'run-suppress-stale-active-owner-work-item',
+      reason: 'messages',
+    });
+
+    expect(result).toBe(true);
+    expect(db.markWorkItemDelivered).toHaveBeenCalledWith(109);
+    expect(channel.sendMessage).not.toHaveBeenCalledWith(
+      chatJid,
+      '이전 step 결과입니다.',
+    );
+    expect(agentRunner.runAgentProcess).toHaveBeenCalledTimes(1);
+    expect(logger.info).toHaveBeenCalledWith(
+      expect.objectContaining({
+        chatJid,
+        workItemId: 109,
+        taskId: 'task-active-owner-follow-up',
+        taskStatus: 'active',
+      }),
+      'Suppressed stale owner delivery retry because a new human message arrived while the paired task was still active',
+    );
+  });
+
  it('suppresses duplicate stale work item delivery and logs the suppression reason', async () => {
    const chatJid = 'group@test';
    const group = makeGroup('codex');
@@ -1632,8 +1753,9 @@ describe('createMessageRuntime', () => {
    vi.mocked(agentRunner.runAgentProcess).mockImplementation(
      async (_group, input, _onProcess, onOutput) => {
        expect(input.prompt).toBe(
-          `The reviewer approved your work (DONE). Finalize and report the result.
-If you intend to close this paired turn now, your first line must be DONE.
+          `The reviewer approved the current task scope (TASK_DONE / legacy DONE). Finalize and report the result.
+If you intend to close this paired turn now, your first line must be TASK_DONE.
+If the original request still has remaining work and the owner flow should continue, your first line may be STEP_DONE.
 If your first line is DONE_WITH_CONCERNS, the system will reopen review instead of finishing.\n\nReviewer's final assessment:\n${truncatedReviewerOutput}`,
        );
        expect(input.prompt).not.toContain(longReviewerOutput);
@@ -1741,7 +1863,7 @@ If your first line is DONE_WITH_CONCERNS, the system will reopen review instead
    vi.mocked(agentRunner.runAgentProcess).mockImplementation(
      async (_group, input, _onProcess, onOutput) => {
        expect(input.prompt).toBe(
-          "The reviewer approved your work (DONE). Finalize and report the result.\nIf you intend to close this paired turn now, your first line must be DONE.\nIf your first line is DONE_WITH_CONCERNS, the system will reopen review instead of finishing.\n\nReviewer's final assessment:\n리뷰 승인 요약",
+          "The reviewer approved the current task scope (TASK_DONE / legacy DONE). Finalize and report the result.\nIf you intend to close this paired turn now, your first line must be TASK_DONE.\nIf the original request still has remaining work and the owner flow should continue, your first line may be STEP_DONE.\nIf your first line is DONE_WITH_CONCERNS, the system will reopen review instead of finishing.\n\nReviewer's final assessment:\n리뷰 승인 요약",
        );
        expect(input.prompt).not.toContain('DONE\n승인합니다.');
        await onOutput?.({
@@ -2004,6 +2126,105 @@ If your first line is DONE_WITH_CONCERNS, the system will reopen review instead
    expect(pending?.prompt).toContain('이전 reviewer 피드백');
  });

+  it('does not build an owner follow-up pending turn from owner STEP_DONE on an active task', () => {
+    const chatJid = 'group@test';
+    const group = makeGroup('claude-code');
+    const task = {
+      id: 'task-owner-step-done-pending',
+      chat_jid: chatJid,
+      group_folder: group.folder,
+      owner_service_id: 'claude',
+      reviewer_service_id: 'codex-main',
+      title: null,
+      source_ref: 'HEAD',
+      plan_notes: null,
+      review_requested_at: null,
+      round_trip_count: 1,
+      status: 'active',
+      arbiter_verdict: null,
+      arbiter_requested_at: null,
+      completion_reason: null,
+      created_at: '2026-03-30T00:00:10.000Z',
+      updated_at: '2026-03-30T00:00:10.000Z',
+    } as any;
+
+    vi.mocked(db.getLatestPreviousPairedTaskForChat).mockReturnValue(undefined);
+    vi.mocked(db.getPairedTurnOutputs).mockReturnValue([
+      {
+        id: 1,
+        task_id: task.id,
+        turn_number: 1,
+        role: 'owner',
+        output_text: 'STEP_DONE\n1단계는 끝났고 2단계로 이어가야 함',
+        created_at: '2026-03-30T00:00:11.000Z',
+      },
+    ] as any);
+
+    const pending = buildPendingPairedTurn({
+      chatJid,
+      timezone: 'UTC',
+      task,
+      rawMissedMessages: [{ seq: 42, timestamp: '2026-03-30T00:00:12.000Z' }],
+      recentHumanMessages: [],
+      labeledRecentMessages: [],
+      resolveChannel: () => makeChannel(chatJid),
+    });
+
+    expect(pending).toBeNull();
+  });
+
+  it('builds a reviewer pending turn when a review_ready task follows owner TASK_DONE output', () => {
+    const chatJid = 'group@test';
+    const group = makeGroup('claude-code');
+    const task = {
+      id: 'task-owner-task-done-pending',
+      chat_jid: chatJid,
+      group_folder: group.folder,
+      owner_service_id: 'claude',
+      reviewer_service_id: 'codex-main',
+      title: null,
+      source_ref: 'HEAD',
+      plan_notes: null,
+      review_requested_at: '2026-03-30T00:00:10.000Z',
+      round_trip_count: 1,
+      status: 'review_ready',
+      arbiter_verdict: null,
+      arbiter_requested_at: null,
+      completion_reason: null,
+      created_at: '2026-03-30T00:00:10.000Z',
+      updated_at: '2026-03-30T00:00:10.000Z',
+    } as any;
+
+    vi.mocked(db.getLatestPreviousPairedTaskForChat).mockReturnValue(undefined);
+    vi.mocked(db.getPairedTurnOutputs).mockReturnValue([
+      {
+        id: 1,
+        task_id: task.id,
+        turn_number: 1,
+        role: 'owner',
+        output_text: 'TASK_DONE\n요청 범위 전체 완료',
+        created_at: '2026-03-30T00:00:11.000Z',
+      },
+    ] as any);
+
+    const pending = buildPendingPairedTurn({
+      chatJid,
+      timezone: 'UTC',
+      task,
+      rawMissedMessages: [{ seq: 42, timestamp: '2026-03-30T00:00:12.000Z' }],
+      recentHumanMessages: [],
+      labeledRecentMessages: [],
+      resolveChannel: () => makeChannel(chatJid, 'discord-review', false),
+    });
+
+    expect(pending).not.toBeNull();
+    expect(pending).toMatchObject({
+      role: 'reviewer',
+      intentKind: 'reviewer-turn',
+      taskId: task.id,
+    });
+  });
+
  it('re-enqueues reviewer after a successful owner delivery moves the task to review_ready', async () => {
    const chatJid = 'group@test';
    const group = makeGroup('codex');
@@ -2109,6 +2330,128 @@ If your first line is DONE_WITH_CONCERNS, the system will reopen review instead
    );
  });

+  it('does not re-enqueue owner follow-up after a successful owner STEP_DONE delivery keeps the task active', async () => {
+    const chatJid = 'group@test';
+    const group = makeGroup('codex');
+    const ownerChannel: Channel = {
+      ...makeChannel(chatJid),
+      isOwnMessage: vi.fn((msg) => msg.sender === 'owner-bot@test'),
+    };
+    const reviewerChannel = makeChannel(chatJid, 'discord-review', false);
+    const enqueueMessageCheck = vi.fn();
+    const pairedTask = {
+      id: 'task-owner-step-done-follow-up',
+      chat_jid: chatJid,
+      group_folder: group.folder,
+      owner_service_id: 'claude',
+      reviewer_service_id: 'codex-main',
+      title: null,
+      source_ref: 'HEAD',
+      plan_notes: null,
+      review_requested_at: null,
+      round_trip_count: 1,
+      owner_failure_count: 2,
+      status: 'active',
+      arbiter_verdict: null,
+      arbiter_requested_at: null,
+      completion_reason: null,
+      created_at: '2026-03-30T00:00:00.000Z',
+      updated_at: '2026-03-30T00:00:00.000Z',
+    } as any;
+    let turnOutputs: any[] = [];
+
+    vi.mocked(serviceRouting.hasReviewerLease).mockReturnValue(true);
+    vi.mocked(db.getLatestOpenPairedTaskForChat).mockImplementation(
+      () => pairedTask,
+    );
+    vi.mocked(db.getPairedTurnOutputs).mockImplementation((taskId: string) =>
+      taskId === pairedTask.id ? turnOutputs : [],
+    );
+    vi.mocked(db.getMessagesSince).mockReturnValue([
+      {
+        id: 'human-step-done-1',
+        chat_jid: chatJid,
+        sender: 'user@test',
+        sender_name: 'User',
+        content: '이 리팩토링 이어서 해줘',
+        timestamp: '2026-03-30T00:00:00.000Z',
+        seq: 1,
+        is_bot_message: false,
+      } as any,
+    ]);
+    vi.mocked(agentRunner.runAgentProcess).mockImplementation(
+      async (_group, _input, _onProcess, onOutput) => {
+        pairedTask.status = 'active';
+        pairedTask.owner_failure_count = 0;
+        turnOutputs = [
+          {
+            id: 1,
+            task_id: pairedTask.id,
+            turn_number: 1,
+            role: 'owner',
+            output_text: 'STEP_DONE\n리팩토링 1단계 완료, 다음 단계 진행',
+            created_at: '2026-03-30T00:00:01.000Z',
+          },
+        ];
+        await onOutput?.({
+          status: 'success',
+          phase: 'final',
+          result: 'STEP_DONE\n리팩토링 1단계 완료, 다음 단계 진행',
+          newSessionId: 'session-owner-step-done-follow-up',
+        });
+        return {
+          status: 'success',
+          result: 'STEP_DONE\n리팩토링 1단계 완료, 다음 단계 진행',
+          newSessionId: 'session-owner-step-done-follow-up',
+        };
+      },
+    );
+
+    const runtime = createMessageRuntime({
+      assistantName: 'Andy',
+      idleTimeout: 1_000,
+      pollInterval: 1_000,
+      timezone: 'UTC',
+      triggerPattern: /^@Andy\b/i,
+      channels: [ownerChannel, reviewerChannel],
+      queue: {
+        registerProcess: vi.fn(),
+        closeStdin: vi.fn(),
+        notifyIdle: vi.fn(),
+        enqueueMessageCheck,
+      } as any,
+      getRoomBindings: () => ({ [chatJid]: group }),
+      getSessions: () => ({}),
+      getLastTimestamp: () => '',
+      setLastTimestamp: vi.fn(),
+      getLastAgentTimestamps: () => ({}),
+      saveState: vi.fn(),
+      persistSession: vi.fn(),
+      clearSession: vi.fn(),
+    });
+
+    const result = await runtime.processGroupMessages(chatJid, {
+      runId: 'run-owner-step-done-follow-up',
+      reason: 'messages',
+    });
+
+    expect(result).toBe(true);
+    expect(ownerChannel.sendMessage).toHaveBeenCalledWith(
+      chatJid,
+      'STEP_DONE\n리팩토링 1단계 완료, 다음 단계 진행',
+    );
+    expect(enqueueMessageCheck).not.toHaveBeenCalled();
+    expect(pairedTask.status).toBe('active');
+    expect(pairedTask.round_trip_count).toBe(1);
+    expect(pairedTask.owner_failure_count).toBe(0);
+    expect(logger.info).not.toHaveBeenCalledWith(
+      expect.objectContaining({
+        intentKind: 'owner-follow-up',
+      }),
+      'Queued paired follow-up after successful owner delivery',
+    );
+  });
+
  it('defers reviewer enqueue after owner delivery when a CI watcher is still active', async () => {
    const chatJid = 'group@test';
    const group = makeGroup('codex');
--- a/src/message-runtime.ts
+++ b/src/message-runtime.ts
@@ -56,6 +56,7 @@ import {
 } from './types.js';
 import { createScopedLogger, logger } from './logger.js';
 import { hasReviewerLease } from './service-routing.js';
+import type { WorkItem } from './db/work-items.js';
 export {
  resolveHandoffCursorKey,
  resolveHandoffRoleOverride,
@@ -113,6 +114,54 @@ export function createMessageRuntime(deps: MessageRuntimeDeps): {
    deps.queue,
  );

+  const getFreshHumanPreflightMessages = (
+    chatJid: string,
+    channel: Channel,
+  ): NewMessage[] => {
+    const sinceSeqCursor = deps.getLastAgentTimestamps()[chatJid] || '0';
+    const preflightRawMessages = getMessagesSinceSeq(
+      chatJid,
+      sinceSeqCursor,
+      deps.assistantName,
+    );
+    const preflightMessages = filterLoopingPairedBotMessages(
+      chatJid,
+      getProcessableMessages(chatJid, preflightRawMessages, channel),
+      FAILURE_FINAL_TEXT,
+    );
+    return preflightMessages.filter(
+      (message) => message.is_from_me !== true && !message.is_bot_message,
+    );
+  };
+
+  const hasHumanMessageAfterWorkItem = (
+    openWorkItem: WorkItem,
+    freshHumanMessages: NewMessage[],
+  ): boolean => {
+    if (freshHumanMessages.length === 0) {
+      return false;
+    }
+
+    const workItemSeq = openWorkItem.end_seq ?? openWorkItem.start_seq ?? null;
+    const workItemUpdatedAt = Date.parse(openWorkItem.updated_at);
+
+    return freshHumanMessages.some((message) => {
+      if (message.seq != null && workItemSeq != null) {
+        return message.seq > workItemSeq;
+      }
+
+      const messageTimestamp = Date.parse(message.timestamp);
+      if (
+        Number.isFinite(messageTimestamp) &&
+        Number.isFinite(workItemUpdatedAt)
+      ) {
+        return messageTimestamp > workItemUpdatedAt;
+      }
+
+      return true;
+    });
+  };
+
  const scheduleQueuedPairedFollowUp = (args: {
    chatJid: string;
    runId: string;
@@ -351,25 +400,15 @@ export function createMessageRuntime(deps: MessageRuntimeDeps): {
      ? getLatestOpenPairedTaskForChat(chatJid)
      : null;
    let openWorkItem = getOpenWorkItemForChat(chatJid, SERVICE_SESSION_SCOPE);
-    if (
-      pendingTask?.status === 'merge_ready' &&
-      openWorkItem?.delivery_role === 'owner'
-    ) {
-      const sinceSeqCursor = deps.getLastAgentTimestamps()[chatJid] || '0';
-      const preflightRawMessages = getMessagesSinceSeq(
+    if (openWorkItem?.delivery_role === 'owner' && pendingTask) {
+      const freshHumanMessages = getFreshHumanPreflightMessages(
        chatJid,
-        sinceSeqCursor,
-        deps.assistantName,
+        channel,
      );
-      const preflightMessages = filterLoopingPairedBotMessages(
-        chatJid,
-        getProcessableMessages(chatJid, preflightRawMessages, channel),
-        FAILURE_FINAL_TEXT,
-      );
-      const hasFreshHumanMessage = preflightMessages.some(
-        (message) => message.is_from_me !== true && !message.is_bot_message,
-      );
-      if (hasFreshHumanMessage) {
+      if (
+        pendingTask.status === 'merge_ready' &&
+        freshHumanMessages.length > 0
+      ) {
        const resolvedTask = resolveOwnerTaskForHumanMessage({
          group,
          chatJid,
@@ -389,6 +428,23 @@ export function createMessageRuntime(deps: MessageRuntimeDeps): {
          );
          openWorkItem = undefined;
        }
+      } else if (
+        pendingTask.status === 'active' &&
+        hasHumanMessageAfterWorkItem(openWorkItem, freshHumanMessages)
+      ) {
+        markWorkItemDelivered(openWorkItem.id);
+        log.info(
+          {
+            chatJid,
+            workItemId: openWorkItem.id,
+            taskId: pendingTask.id,
+            taskStatus: pendingTask.status,
+            workItemEndSeq: openWorkItem.end_seq ?? null,
+            freshHumanMessageCount: freshHumanMessages.length,
+          },
+          'Suppressed stale owner delivery retry because a new human message arrived while the paired task was still active',
+        );
+        openWorkItem = undefined;
      }
    }
    const openWorkItemOutcome = await processOpenWorkItemDelivery({
--- a/src/paired-execution-context-owner.ts
+++ b/src/paired-execution-context-owner.ts
@@ -22,6 +22,7 @@ import type { PairedTask } from './types.js';

 type OwnerFinalizeOutcome = 'stop' | 're_review';
 const OWNER_FAILURE_ESCALATION_THRESHOLD = 2;
+const EMPTY_STEP_DONE_THRESHOLD = 2;

 export function handleFailedOwnerExecution(args: {
  task: PairedTask;
@@ -101,6 +102,14 @@ function handleOwnerFinalizeCompletion(args: {
  const hasNewChanges = workspace?.workspace_dir
    ? hasCodeChangesSinceRef(workspace.workspace_dir, task.source_ref)
    : null;
+  const nextFinalizeStepDoneCount =
+    ownerVerdict === 'step_done'
+      ? (task.finalize_step_done_count ?? 0) + 1
+      : (task.finalize_step_done_count ?? 0);
+  const nextEmptyStepDoneStreak =
+    ownerVerdict === 'step_done' && hasNewChanges === false
+      ? (task.empty_step_done_streak ?? 0) + 1
+      : 0;
  const signal = resolveOwnerCompletionSignal({
    phase: 'finalize',
    visibleVerdict: ownerVerdict,
@@ -138,6 +147,41 @@ function handleOwnerFinalizeCompletion(args: {
      },
      patch: {
        owner_failure_count: 0,
+        owner_step_done_streak: 0,
+        finalize_step_done_count: nextFinalizeStepDoneCount,
+        empty_step_done_streak: nextEmptyStepDoneStreak,
+      },
+    });
+    return 'stop';
+  }
+
+  if (
+    ownerVerdict === 'step_done' &&
+    hasNewChanges === false &&
+    nextEmptyStepDoneStreak >= EMPTY_STEP_DONE_THRESHOLD
+  ) {
+    requestArbiterOrEscalate({
+      taskId,
+      currentStatus: task.status,
+      expectedUpdatedAt: task.updated_at,
+      now,
+      arbiterLogMessage:
+        'Owner repeated STEP_DONE during finalize without code changes — requesting arbiter',
+      escalateLogMessage:
+        'Owner repeated STEP_DONE during finalize without code changes — escalating to user',
+      logContext: {
+        taskId,
+        ownerVerdict,
+        hasNewChanges,
+        emptyStepDoneStreak: nextEmptyStepDoneStreak,
+        finalizeStepDoneCount: nextFinalizeStepDoneCount,
+        summary: summary?.slice(0, 100),
+      },
+      patch: {
+        owner_failure_count: 0,
+        owner_step_done_streak: 0,
+        finalize_step_done_count: nextFinalizeStepDoneCount,
+        empty_step_done_streak: nextEmptyStepDoneStreak,
      },
    });
    return 'stop';
@@ -153,6 +197,10 @@ function handleOwnerFinalizeCompletion(args: {
        updatedAt: now,
        patch: {
          owner_failure_count: 0,
+          owner_step_done_streak: 0,
+          finalize_step_done_count: nextFinalizeStepDoneCount,
+          empty_step_done_streak:
+            ownerVerdict === 'step_done' ? nextEmptyStepDoneStreak : 0,
        },
      });
    }
@@ -165,9 +213,27 @@ function handleOwnerFinalizeCompletion(args: {
      },
      ownerVerdict === 'done_with_concerns'
        ? 'Owner raised concerns during finalize — task set back to active'
-        : 'Owner made changes after reviewer approval — task set back to active before re-review',
+        : ownerVerdict === 'step_done'
+          ? 'Owner reported STEP_DONE during finalize — task set back to active before review'
+          : 'Owner made changes after reviewer approval — task set back to active before re-review',
    );
-    return 're_review';
+    maybeAutoTriggerReviewerAfterOwnerCompletion({
+      task,
+      taskId,
+      now,
+      logMessage:
+        ownerVerdict === 'step_done'
+          ? 'Auto-triggered reviewer after owner finalize STEP_DONE'
+          : 'Auto-triggered reviewer after owner finalize required re-review',
+      patch:
+        ownerVerdict === 'step_done'
+          ? {
+              finalize_step_done_count: nextFinalizeStepDoneCount,
+              empty_step_done_streak: nextEmptyStepDoneStreak,
+            }
+          : undefined,
+    });
+    return 'stop';
  }

  transitionPairedTaskStatus({
@@ -179,6 +245,9 @@ function handleOwnerFinalizeCompletion(args: {
    patch: {
      completion_reason: 'done',
      owner_failure_count: 0,
+      owner_step_done_streak: 0,
+      finalize_step_done_count: nextFinalizeStepDoneCount,
+      empty_step_done_streak: 0,
    },
  });
  logger.info(
@@ -193,6 +262,11 @@ function maybeAutoTriggerReviewerAfterOwnerCompletion(args: {
  taskId: string;
  now: string;
  logMessage: string;
+  patch?: {
+    owner_step_done_streak?: number;
+    finalize_step_done_count?: number;
+    empty_step_done_streak?: number;
+  };
 }): void {
  const { task, taskId, now, logMessage } = args;
  if (task.round_trip_count >= PAIRED_MAX_ROUND_TRIPS) {
@@ -220,6 +294,9 @@ function maybeAutoTriggerReviewerAfterOwnerCompletion(args: {
      patch: {
        round_trip_count: task.round_trip_count + 1,
        owner_failure_count: 0,
+        owner_step_done_streak: 0,
+        empty_step_done_streak: 0,
+        ...args.patch,
      },
    });
    if (hasActiveCiWatcherForChat(task.chat_jid)) {
@@ -265,6 +342,8 @@ export function handleOwnerCompletion(args: {
  }

  const ownerVerdict = parseVisibleVerdict(summary);
+  const nextOwnerStepDoneStreak =
+    ownerVerdict === 'step_done' ? (task.owner_step_done_streak ?? 0) + 1 : 0;
  const signal = resolveOwnerCompletionSignal({
    phase: 'normal',
    visibleVerdict: ownerVerdict,
@@ -281,15 +360,27 @@ export function handleOwnerCompletion(args: {
      logContext: {
        taskId,
        ownerVerdict,
+        ownerStepDoneStreak: nextOwnerStepDoneStreak,
        summary: summary?.slice(0, 100),
      },
      patch: {
        owner_failure_count: 0,
+        owner_step_done_streak: nextOwnerStepDoneStreak,
      },
    });
    return;
  }

+  if (nextOwnerStepDoneStreak !== (task.owner_step_done_streak ?? 0)) {
+    applyPairedTaskPatch({
+      taskId,
+      expectedUpdatedAt: task.updated_at,
+      updatedAt: now,
+      patch: {
+        owner_step_done_streak: nextOwnerStepDoneStreak,
+      },
+    });
+  }
  maybeAutoTriggerReviewerAfterOwnerCompletion({
    task,
    taskId,
--- a/src/paired-execution-context-shared.test.ts
+++ b/src/paired-execution-context-shared.test.ts
@@ -9,6 +9,8 @@ import {

 describe('paired execution context shared verdict helpers', () => {
  it('parses visible verdicts from the first summary line only', () => {
+    expect(parseVisibleVerdict('STEP_DONE\nmore to do')).toBe('step_done');
+    expect(parseVisibleVerdict('TASK_DONE\nall done')).toBe('task_done');
    expect(
      parseVisibleVerdict(
        'DONE_WITH_CONCERNS\n\nfollow-up detail that should not affect parsing',
@@ -19,6 +21,15 @@ describe('paired execution context shared verdict helpers', () => {
  });

  it('maps normal owner completion verdicts to reviewer or arbiter signals', () => {
+    expect(
+      resolveOwnerCompletionSignal({
+        phase: 'normal',
+        visibleVerdict: 'step_done',
+      }),
+    ).toEqual({
+      kind: 'request_reviewer',
+      resetStatusToActive: false,
+    });
    expect(
      resolveOwnerCompletionSignal({
        phase: 'normal',
@@ -37,6 +48,18 @@ describe('paired execution context shared verdict helpers', () => {
  });

  it('maps finalize owner outcomes to complete, re-review, or arbiter', () => {
+    expect(
+      resolveOwnerCompletionSignal({
+        phase: 'finalize',
+        visibleVerdict: 'step_done',
+        hasChangesSinceApproval: false,
+        roundTripCount: 0,
+        deadlockThreshold: 2,
+      }),
+    ).toEqual({
+      kind: 'request_reviewer',
+      resetStatusToActive: true,
+    });
    expect(
      resolveOwnerCompletionSignal({
        phase: 'finalize',
@@ -70,6 +93,20 @@ describe('paired execution context shared verdict helpers', () => {
  });

  it('maps reviewer completion verdicts to finalize, owner changes, or arbiter', () => {
+    expect(
+      resolveReviewerCompletionSignal({
+        visibleVerdict: 'task_done',
+        roundTripCount: 0,
+        deadlockThreshold: 3,
+      }),
+    ).toEqual({ kind: 'request_owner_finalize' });
+    expect(
+      resolveReviewerCompletionSignal({
+        visibleVerdict: 'step_done',
+        roundTripCount: 0,
+        deadlockThreshold: 3,
+      }),
+    ).toEqual({ kind: 'request_owner_changes' });
    expect(
      resolveReviewerCompletionSignal({
        visibleVerdict: 'done',
@@ -94,6 +131,11 @@ describe('paired execution context shared verdict helpers', () => {
  });

  it('maps reviewer failure verdicts to explicit failure signals', () => {
+    expect(
+      resolveReviewerFailureSignal({
+        visibleVerdict: 'task_done',
+      }),
+    ).toEqual({ kind: 'request_owner_finalize' });
    expect(
      resolveReviewerFailureSignal({
        visibleVerdict: 'done',
--- a/src/paired-execution-context-shared.ts
+++ b/src/paired-execution-context-shared.ts
@@ -3,15 +3,9 @@ import { execFileSync } from 'child_process';
 import { isArbiterEnabled } from './config.js';
 import { updatePairedTaskIfUnchanged } from './db.js';
 import { logger } from './logger.js';
+import { parseVisibleVerdict, type VisibleVerdict } from './paired-verdict.js';
 import type { PairedTaskStatus } from './types.js';

-export type VisibleVerdict =
-  | 'done'
-  | 'done_with_concerns'
-  | 'blocked'
-  | 'needs_context'
-  | 'continue';
-
 export type CompletionSignal =
  | { kind: 'request_reviewer'; resetStatusToActive: boolean }
  | { kind: 'request_owner_finalize' }
@@ -20,22 +14,8 @@ export type CompletionSignal =
  | { kind: 'complete'; completionReason: 'done' | 'escalated' }
  | { kind: 'preserve_review_ready' };

-export function parseVisibleVerdict(
-  summary: string | null | undefined,
-): VisibleVerdict {
-  if (!summary) return 'continue';
-  const cleaned = summary.replace(/<internal>[\s\S]*?<\/internal>/g, '').trim();
-  if (!cleaned) return 'continue';
-  const firstLine = cleaned.split('\n')[0].trim();
-  if (/^\*{0,2}BLOCKED\*{0,2}\b/i.test(firstLine)) return 'blocked';
-  if (/^\*{0,2}NEEDS_CONTEXT\*{0,2}\b/i.test(firstLine)) return 'needs_context';
-  if (/^\*{0,2}DONE_WITH_CONCERNS\*{0,2}\b/i.test(firstLine))
-    return 'done_with_concerns';
-  if (/^\*{0,2}DONE\*{0,2}\b/i.test(firstLine)) return 'done';
-  if (/^\*{0,2}Approved\.?\*{0,2}/i.test(firstLine)) return 'done';
-  if (/^\*{0,2}LGTM\*{0,2}/i.test(firstLine)) return 'done';
-  return 'continue';
-}
+export { parseVisibleVerdict };
+export type { VisibleVerdict };

 export function resolveOwnerCompletionSignal(args: {
  phase: 'normal' | 'finalize';
@@ -63,6 +43,13 @@ export function resolveOwnerCompletionSignal(args: {
    };
  }

+  if (visibleVerdict === 'step_done') {
+    return {
+      kind: 'request_reviewer',
+      resetStatusToActive: true,
+    };
+  }
+
  const needsReReview =
    visibleVerdict === 'done_with_concerns' || hasChangesSinceApproval === true;

@@ -90,8 +77,11 @@ export function resolveReviewerCompletionSignal(args: {
  const { visibleVerdict, roundTripCount, deadlockThreshold } = args;

  switch (visibleVerdict) {
+    case 'task_done':
    case 'done':
      return { kind: 'request_owner_finalize' };
+    case 'step_done':
+      return { kind: 'request_owner_changes' };
    case 'blocked':
    case 'needs_context':
      return { kind: 'request_arbiter' };
@@ -111,6 +101,7 @@ export function resolveReviewerFailureSignal(args: {
  const { visibleVerdict } = args;

  switch (visibleVerdict) {
+    case 'task_done':
    case 'done':
      return { kind: 'request_owner_finalize' };
    case 'blocked':
@@ -248,6 +239,10 @@ export function transitionPairedTaskStatus(args: {
    review_requested_at?: string | null;
    round_trip_count?: number;
    owner_failure_count?: number;
+    owner_step_done_streak?: number;
+    finalize_step_done_count?: number;
+    task_done_then_user_reopen_count?: number;
+    empty_step_done_streak?: number;
    arbiter_verdict?: string | null;
    arbiter_requested_at?: string | null;
    completion_reason?: string | null;
@@ -292,6 +287,10 @@ export function applyPairedTaskPatch(args: {
    review_requested_at?: string | null;
    round_trip_count?: number;
    owner_failure_count?: number;
+    owner_step_done_streak?: number;
+    finalize_step_done_count?: number;
+    task_done_then_user_reopen_count?: number;
+    empty_step_done_streak?: number;
    status?: PairedTaskStatus;
    arbiter_verdict?: string | null;
    arbiter_requested_at?: string | null;
@@ -333,6 +332,10 @@ export function requestArbiterOrEscalate(args: {
    review_requested_at?: string | null;
    round_trip_count?: number;
    owner_failure_count?: number;
+    owner_step_done_streak?: number;
+    finalize_step_done_count?: number;
+    task_done_then_user_reopen_count?: number;
+    empty_step_done_streak?: number;
    arbiter_verdict?: string | null;
    arbiter_requested_at?: string | null;
    completion_reason?: string | null;
--- a/src/paired-execution-context.test.ts
+++ b/src/paired-execution-context.test.ts
@@ -143,6 +143,10 @@ function buildPairedTask(overrides: Partial<PairedTask> = {}): PairedTask {
    review_requested_at: null,
    round_trip_count: 0,
    owner_failure_count: 0,
+    owner_step_done_streak: 0,
+    finalize_step_done_count: 0,
+    task_done_then_user_reopen_count: 0,
+    empty_step_done_streak: 0,
    status: 'active',
    arbiter_verdict: null,
    arbiter_requested_at: null,
@@ -293,6 +297,62 @@ describe('paired execution context', () => {
    expect(db.insertPairedTurnOutput).not.toHaveBeenCalled();
  });

+  it('records a quick reopen when a new owner task starts shortly after TASK_DONE completion', () => {
+    const previousTask = buildPairedTask({
+      id: 'task-completed',
+      status: 'completed',
+      completion_reason: 'done',
+      updated_at: new Date(Date.now() - 60_000).toISOString(),
+      task_done_then_user_reopen_count: 0,
+    });
+    vi.mocked(db.getLatestOpenPairedTaskForChat).mockReturnValue(undefined);
+    vi.mocked(db.getLatestPairedTaskForChat).mockReturnValue(previousTask);
+
+    const result = resolveOwnerTaskForHumanMessage({
+      group,
+      chatJid: 'dc:test',
+      roomRoleContext: ownerContext,
+      existingTask: null,
+    });
+
+    expect(result.task).not.toBeNull();
+    expect(db.updatePairedTask).toHaveBeenCalledWith(
+      'task-completed',
+      expect.objectContaining({
+        task_done_then_user_reopen_count: 1,
+      }),
+    );
+  });
+
+  it('resets active STEP_DONE loop counters when a new human message continues an active task', () => {
+    vi.mocked(db.getLatestOpenPairedTaskForChat).mockReturnValue(
+      buildPairedTask({
+        status: 'active',
+        owner_failure_count: 1,
+        owner_step_done_streak: 2,
+        empty_step_done_streak: 2,
+      }),
+    );
+
+    preparePairedExecutionContext({
+      group,
+      chatJid: 'dc:test',
+      runId: 'run-human-reset-step-done-loop',
+      roomRoleContext: ownerContext,
+      hasHumanMessage: true,
+    });
+
+    expect(db.updatePairedTask).toHaveBeenCalledWith(
+      'task-1',
+      expect.objectContaining({
+        round_trip_count: 0,
+        owner_failure_count: 0,
+        owner_step_done_streak: 0,
+        empty_step_done_streak: 0,
+      }),
+    );
+  });
+
  it('uses room role context agent overrides when creating a paired task', () => {
    preparePairedExecutionContext({
      group,
@@ -742,6 +802,184 @@ describe('paired execution context', () => {
    ).not.toHaveBeenCalled();
  });

+  it('requests review when the owner reports STEP_DONE in active mode', () => {
+    vi.mocked(db.getPairedTaskById).mockReturnValue(
+      buildPairedTask({
+        status: 'active',
+        round_trip_count: 1,
+      }),
+    );
+    vi.mocked(pairedWorkspaceManager.markPairedTaskReviewReady).mockReturnValue(
+      {
+        ownerWorkspace: buildWorkspace('owner', '/tmp/paired/task-1/owner'),
+        reviewerWorkspace: buildWorkspace(
+          'reviewer',
+          '/tmp/paired/task-1/reviewer',
+        ),
+      },
+    );
+
+    completePairedExecutionContext({
+      taskId: 'task-1',
+      role: 'owner',
+      status: 'succeeded',
+      summary: 'STEP_DONE\n1단계 완료, 후속 작업 계속',
+    });
+
+    expect(db.updatePairedTask).toHaveBeenCalledWith(
+      'task-1',
+      expect.objectContaining({
+        round_trip_count: 2,
+        owner_step_done_streak: 0,
+        empty_step_done_streak: 0,
+      }),
+    );
+    expect(
+      pairedWorkspaceManager.markPairedTaskReviewReady,
+    ).toHaveBeenCalledWith('task-1');
+  });
+
+  it('requests review instead of arbiter when active STEP_DONE repeats without code changes', () => {
+    vi.spyOn(config, 'isArbiterEnabled').mockReturnValue(true);
+
+    const repoDir = createCanonicalRepoWithCommit('active step done loop');
+    const sourceRef = resolveTreeRef(repoDir);
+    vi.mocked(db.getPairedTaskById).mockReturnValue(
+      buildPairedTask({
+        status: 'active',
+        source_ref: sourceRef,
+        owner_step_done_streak: 2,
+        empty_step_done_streak: 2,
+      }),
+    );
+    vi.mocked(db.getPairedWorkspace).mockImplementation((_taskId, role) =>
+      role === 'owner' ? buildWorkspace('owner', repoDir) : undefined,
+    );
+    vi.mocked(pairedWorkspaceManager.markPairedTaskReviewReady).mockReturnValue(
+      {
+        ownerWorkspace: buildWorkspace('owner', repoDir),
+        reviewerWorkspace: buildWorkspace(
+          'reviewer',
+          '/tmp/paired/task-1/reviewer',
+        ),
+      },
+    );
+
+    completePairedExecutionContext({
+      taskId: 'task-1',
+      role: 'owner',
+      status: 'succeeded',
+      summary: 'STEP_DONE\n요약만 반복되고 코드 변경은 없음',
+    });
+
+    expect(db.updatePairedTask).toHaveBeenCalledWith(
+      'task-1',
+      expect.objectContaining({
+        round_trip_count: 1,
+        owner_step_done_streak: 0,
+        empty_step_done_streak: 0,
+      }),
+    );
+    expect(
+      pairedWorkspaceManager.markPairedTaskReviewReady,
+    ).toHaveBeenCalledWith('task-1');
+    expect(db.updatePairedTask).not.toHaveBeenCalledWith(
+      'task-1',
+      expect.objectContaining({
+        status: 'arbiter_requested',
+      }),
+    );
+  });
+
+  it('re-triggers review when the owner reports STEP_DONE during finalize', () => {
+    const repoDir = createCanonicalRepoWithCommit('reviewed');
+    const approvedSourceRef = resolveTreeRef(repoDir);
+    vi.mocked(db.getPairedTaskById).mockReturnValue(
+      buildPairedTask({
+        status: 'merge_ready',
+        source_ref: approvedSourceRef,
+        round_trip_count: 1,
+      }),
+    );
+    vi.mocked(db.getPairedWorkspace).mockImplementation((_taskId, role) =>
+      role === 'owner' ? buildWorkspace('owner', repoDir) : undefined,
+    );
+    vi.mocked(pairedWorkspaceManager.markPairedTaskReviewReady).mockReturnValue(
+      {
+        ownerWorkspace: buildWorkspace('owner', repoDir),
+        reviewerWorkspace: buildWorkspace(
+          'reviewer',
+          '/tmp/paired/task-1/reviewer',
+        ),
+      },
+    );
+
+    completePairedExecutionContext({
+      taskId: 'task-1',
+      role: 'owner',
+      status: 'succeeded',
+      summary: 'STEP_DONE\n남은 범위가 있어서 계속 진행',
+    });
+
+    expect(db.updatePairedTask).toHaveBeenCalledWith(
+      'task-1',
+      expect.objectContaining({
+        status: 'active',
+        owner_failure_count: 0,
+        finalize_step_done_count: 1,
+        empty_step_done_streak: 1,
+      }),
+    );
+    expect(db.updatePairedTask).toHaveBeenCalledWith(
+      'task-1',
+      expect.objectContaining({
+        round_trip_count: 2,
+        finalize_step_done_count: 1,
+        empty_step_done_streak: 1,
+      }),
+    );
+    expect(
+      pairedWorkspaceManager.markPairedTaskReviewReady,
+    ).toHaveBeenCalledWith('task-1');
+  });
+
+  it('requests arbiter when finalize STEP_DONE repeats without code changes', () => {
+    vi.spyOn(config, 'isArbiterEnabled').mockReturnValue(true);
+
+    const repoDir = createCanonicalRepoWithCommit('reviewed');
+    const approvedSourceRef = resolveTreeRef(repoDir);
+    vi.mocked(db.getPairedTaskById).mockReturnValue(
+      buildPairedTask({
+        status: 'merge_ready',
+        source_ref: approvedSourceRef,
+        empty_step_done_streak: 1,
+        finalize_step_done_count: 1,
+      }),
+    );
+    vi.mocked(db.getPairedWorkspace).mockImplementation((_taskId, role) =>
+      role === 'owner' ? buildWorkspace('owner', repoDir) : undefined,
+    );
+
+    completePairedExecutionContext({
+      taskId: 'task-1',
+      role: 'owner',
+      status: 'succeeded',
+      summary: 'STEP_DONE\n아직 남았지만 코드 변경은 없음',
+    });
+
+    expect(db.updatePairedTask).toHaveBeenCalledWith(
+      'task-1',
+      expect.objectContaining({
+        status: 'arbiter_requested',
+        empty_step_done_streak: 2,
+        finalize_step_done_count: 2,
+      }),
+    );
+    expect(
+      pairedWorkspaceManager.markPairedTaskReviewReady,
+    ).not.toHaveBeenCalled();
+  });
+
  it('re-triggers review once when owner reports DONE_WITH_CONCERNS during finalize', () => {
    vi.mocked(db.getPairedTaskById).mockReturnValue(
      buildPairedTask({
--- a/src/paired-execution-context.ts
+++ b/src/paired-execution-context.ts
@@ -30,6 +30,7 @@ import {
  hasActiveCiWatcherForChat,
  insertPairedTurnOutput,
  releasePairedTaskExecutionLease,
+  updatePairedTask,
  upsertPairedProject,
 } from './db.js';
 import { logger } from './logger.js';
@@ -72,6 +73,8 @@ import type {
 } from './types.js';
 import { resolveRoleAgentPlan } from './role-agent-plan.js';

+const TASK_DONE_REOPEN_WINDOW_MS = 10 * 60_000;
+
 function ensurePairedProject(
  group: RegisteredGroup,
  chatJid: string,
@@ -144,6 +147,10 @@ function createActiveTaskForRoom(args: {
    review_requested_at: null,
    round_trip_count: 0,
    owner_failure_count: 0,
+    owner_step_done_streak: 0,
+    finalize_step_done_count: 0,
+    task_done_then_user_reopen_count: 0,
+    empty_step_done_streak: 0,
    status: 'active',
    arbiter_verdict: null,
    arbiter_requested_at: null,
@@ -164,6 +171,39 @@ function createActiveTaskForRoom(args: {
  return task;
 }

+function maybeRecordTaskDoneReopen(previousTask: PairedTask | null): void {
+  if (
+    !previousTask ||
+    previousTask.status !== 'completed' ||
+    previousTask.completion_reason !== 'done'
+  ) {
+    return;
+  }
+
+  const completedAt = Date.parse(previousTask.updated_at);
+  if (!Number.isFinite(completedAt)) {
+    return;
+  }
+  if (Date.now() - completedAt > TASK_DONE_REOPEN_WINDOW_MS) {
+    return;
+  }
+
+  updatePairedTask(previousTask.id, {
+    task_done_then_user_reopen_count:
+      (previousTask.task_done_then_user_reopen_count ?? 0) + 1,
+    updated_at: new Date().toISOString(),
+  });
+  logger.info(
+    {
+      taskId: previousTask.id,
+      chatJid: previousTask.chat_jid,
+      reopenCount: (previousTask.task_done_then_user_reopen_count ?? 0) + 1,
+      completionReason: previousTask.completion_reason,
+    },
+    'Recorded paired task reopen shortly after TASK_DONE completion',
+  );
+}
+
 function cancelOutstandingFinalizeOwnerTurn(task: PairedTask): void {
  const turnIdentity = buildPairedTurnIdentity({
    taskId: task.id,
@@ -253,6 +293,7 @@ export function resolveOwnerTaskForHumanMessage(args: {
    args.existingTask ?? getLatestOpenPairedTaskForChat(args.chatJid) ?? null;

  if (!existing) {
+    maybeRecordTaskDoneReopen(getLatestPairedTaskForChat(args.chatJid) ?? null);
    return {
      task: canonicalWorkDir
        ? createActiveTaskForRoom({
@@ -432,7 +473,12 @@ export function preparePairedExecutionContext(args: {
          updatedAt: now,
          patch: {
            ...(hasHuman
-              ? { round_trip_count: 0, owner_failure_count: 0 }
+              ? {
+                  round_trip_count: 0,
+                  owner_failure_count: 0,
+                  owner_step_done_streak: 0,
+                  empty_step_done_streak: 0,
+                }
              : {}),
          },
        });
@@ -443,7 +489,12 @@ export function preparePairedExecutionContext(args: {
          updatedAt: now,
          patch: {
            ...(hasHuman
-              ? { round_trip_count: 0, owner_failure_count: 0 }
+              ? {
+                  round_trip_count: 0,
+                  owner_failure_count: 0,
+                  owner_step_done_streak: 0,
+                  empty_step_done_streak: 0,
+                }
              : {}),
          },
        });
--- a/src/paired-verdict.ts
+++ b/src/paired-verdict.ts
@@ -0,0 +1,40 @@
+export type VisibleVerdict =
+  | 'step_done'
+  | 'task_done'
+  | 'done'
+  | 'done_with_concerns'
+  | 'blocked'
+  | 'needs_context'
+  | 'continue';
+
+export function parseVisibleVerdict(
+  summary: string | null | undefined,
+): VisibleVerdict {
+  if (!summary) return 'continue';
+  const cleaned = summary.replace(/<internal>[\s\S]*?<\/internal>/g, '').trim();
+  if (!cleaned) return 'continue';
+  const firstLine = cleaned.split('\n')[0].trim();
+  if (/^\*{0,2}BLOCKED\*{0,2}\b/i.test(firstLine)) return 'blocked';
+  if (/^\*{0,2}NEEDS_CONTEXT\*{0,2}\b/i.test(firstLine)) return 'needs_context';
+  if (/^\*{0,2}STEP_DONE\*{0,2}\b/i.test(firstLine)) return 'step_done';
+  if (/^\*{0,2}TASK_DONE\*{0,2}\b/i.test(firstLine)) return 'task_done';
+  if (/^\*{0,2}DONE_WITH_CONCERNS\*{0,2}\b/i.test(firstLine))
+    return 'done_with_concerns';
+  if (/^\*{0,2}DONE\*{0,2}\b/i.test(firstLine)) return 'done';
+  if (/^\*{0,2}Approved\.?\*{0,2}/i.test(firstLine)) return 'done';
+  if (/^\*{0,2}LGTM\*{0,2}/i.test(firstLine)) return 'done';
+  return 'continue';
+}
+
+export function resolveStoredVisibleVerdict(args: {
+  verdict?: VisibleVerdict | null;
+  outputText?: string | null;
+}): VisibleVerdict | null {
+  if (args.verdict) {
+    return args.verdict;
+  }
+  if (!args.outputText) {
+    return null;
+  }
+  return parseVisibleVerdict(args.outputText);
+}
--- a/src/types.ts
+++ b/src/types.ts
@@ -1,3 +1,5 @@
+import type { VisibleVerdict } from './paired-verdict.js';
+
 export interface AgentConfig {
  timeout?: number; // Default: 300000 (5 minutes)
  // Per-group model/effort overrides (take precedence over global env vars)
@@ -83,6 +85,10 @@ export interface PairedTask {
  review_requested_at: string | null;
  round_trip_count: number;
  owner_failure_count?: number | null;
+  owner_step_done_streak?: number | null;
+  finalize_step_done_count?: number | null;
+  task_done_then_user_reopen_count?: number | null;
+  empty_step_done_streak?: number | null;
  status: PairedTaskStatus;
  arbiter_verdict: string | null;
  arbiter_requested_at: string | null;
@@ -97,6 +103,7 @@ export interface PairedTurnOutput {
  turn_number: number;
  role: PairedRoomRole;
  output_text: string;
+  verdict?: VisibleVerdict | null;
  created_at: string;
 }

--- a/src/verification.test.ts
+++ b/src/verification.test.ts
@@ -20,7 +20,7 @@ describe('verification helpers', () => {
    expect(isVerificationProfile('test')).toBe(true);
    expect(isVerificationProfile('typecheck')).toBe(true);
    expect(isVerificationProfile('build')).toBe(true);
-    expect(isVerificationProfile('lint')).toBe(false);
+    expect(isVerificationProfile('lint')).toBe(true);
  });

  it('builds deterministic commands for each profile', () => {
@@ -44,6 +44,49 @@ describe('verification helpers', () => {
      args: ['run', 'build'],
      requiredScript: 'build',
    });
+    expect(buildVerificationCommand('lint', repoDir)).toMatchObject({
+      file: 'npm',
+      args: ['run', 'lint'],
+      requiredScript: 'lint',
+    });
+  });
+
+  it('prefers lint:ci and lint:check before plain lint', () => {
+    const repoDir = fs.mkdtempSync(
+      path.join(os.tmpdir(), 'ejclaw-verification-lint-'),
+    );
+    fs.writeFileSync(
+      path.join(repoDir, 'package.json'),
+      JSON.stringify({
+        scripts: {
+          lint: 'eslint .',
+          'lint:check': 'eslint --max-warnings 0 .',
+          'lint:ci': 'eslint --format compact .',
+        },
+      }),
+    );
+
+    expect(buildVerificationCommand('lint', repoDir)).toMatchObject({
+      file: 'npm',
+      args: ['run', 'lint:ci'],
+      requiredScript: 'lint:ci',
+    });
+
+    fs.writeFileSync(
+      path.join(repoDir, 'package.json'),
+      JSON.stringify({
+        scripts: {
+          lint: 'eslint .',
+          'lint:check': 'eslint --max-warnings 0 .',
+        },
+      }),
+    );
+
+    expect(buildVerificationCommand('lint', repoDir)).toMatchObject({
+      file: 'npm',
+      args: ['run', 'lint:check'],
+      requiredScript: 'lint:check',
+    });
  });

  it('selects the workspace package manager for verification commands', () => {
@@ -250,6 +293,53 @@ describe('verification helpers', () => {
    expect(result.snapshotId).toBe(expectedSnapshotId);
  });

+  it('runs lint verification using the preferred lint script', async () => {
+    const repoDir = fs.mkdtempSync(
+      path.join(os.tmpdir(), 'ejclaw-verification-direct-lint-'),
+    );
+    fs.mkdirSync(path.join(repoDir, 'node_modules', '.bin'), {
+      recursive: true,
+    });
+    fs.mkdirSync(path.join(repoDir, 'src'), { recursive: true });
+    fs.writeFileSync(
+      path.join(repoDir, 'package.json'),
+      JSON.stringify({
+        name: 'verification-direct-lint',
+        packageManager: 'bun@1.3.11',
+        scripts: {
+          'lint:check':
+            "node -e \"process.stdout.write('lint-check:' + (process.env.CI || 'missing'))\"",
+        },
+      }),
+    );
+    fs.writeFileSync(path.join(repoDir, 'bun.lock'), '');
+    fs.writeFileSync(
+      path.join(repoDir, 'node_modules', '.bin', 'placeholder'),
+      '',
+    );
+    fs.writeFileSync(
+      path.join(repoDir, 'src', 'index.ts'),
+      'export const value = 1;\n',
+    );
+    ensureWorkspaceDependenciesInstalled(repoDir);
+
+    const expectedSnapshotId = computeVerificationSnapshot(repoDir).snapshotId;
+    const result = await runVerificationRequest(
+      {
+        requestId: 'req-direct-lint',
+        profile: 'lint',
+        expectedSnapshotId,
+      },
+      { repoDir },
+    );
+
+    expect(result.ok).toBe(true);
+    expect(result.exitCode).toBe(0);
+    expect(result.command).toContain('lint:check');
+    expect(result.stdout).toContain('lint-check:1');
+    expect(result.snapshotId).toBe(expectedSnapshotId);
+  });
+
  it('does not leak host secrets into direct verification commands', async () => {
    const repoDir = fs.mkdtempSync(
      path.join(os.tmpdir(), 'ejclaw-verification-direct-secret-'),
--- a/src/verification.ts
+++ b/src/verification.ts
@@ -11,7 +11,12 @@ import {
 } from './workspace-package-manager.js';
 import { computeVerificationSnapshotId } from '../shared/verification-snapshot.js';

-export const VERIFICATION_PROFILES = ['test', 'typecheck', 'build'] as const;
+export const VERIFICATION_PROFILES = [
+  'test',
+  'typecheck',
+  'build',
+  'lint',
+] as const;

 export type VerificationProfile = (typeof VERIFICATION_PROFILES)[number];

@@ -91,12 +96,19 @@ export function buildVerificationCommand(
  profile: VerificationProfile,
  repoDir: string = process.cwd(),
 ): VerificationCommandSpec {
+  const scripts = readPackageScripts(repoDir);
  const scriptName =
    profile === 'test'
      ? 'test'
      : profile === 'typecheck'
        ? 'typecheck'
-        : 'build';
+        : profile === 'build'
+          ? 'build'
+          : scripts['lint:ci']
+            ? 'lint:ci'
+            : scripts['lint:check']
+              ? 'lint:check'
+              : 'lint';
  const command = buildWorkspaceScriptCommand(repoDir, scriptName);
  return {
    file: command.file,