Kiro CLI TUIを共通live E2E policy/lifecycleへ接続する - #2233

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

tests/harness/live-e2e/tmux.ts (1)

35-41: 🩺 Stability & Availability | 🟠 Major | ⚡ Quick win

spawnSync に timeout を設定してください。

run は同期呼び出しです。tmux クライアントがサーバー応答を待って停止すると、呼び出しスレッドが無期限にブロックします。この間 #waitForReady の deadline も AbortSignal も評価されません。kiro-tui.ts の #probeKiro は spawnSync に timeout: 15_000 を設定しています。共通ポートにも同じ境界を入れてください。

🛡️ 提案する修正

 export interface TmuxCommandOptions {
   readonly cwd?: string;
   readonly env?: Readonly<Record<string, string>>;
+  readonly timeoutMs?: number;
 }

     const result = spawnSync(this.#tmuxBin, [...args], {
       cwd: options.cwd,
       env: options.env,
       encoding: "utf8",
       maxBuffer: MAX_PANE_BYTES + 1,
+      timeout: options.timeoutMs ?? 15_000,
     });

timeout 超過時は status が null、error が設定されます。commandFailed と stderr の組み立ては既にその形を吸収します。

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/harness/live-e2e/tmux.ts` around lines 35 - 41, Update the spawnSync
call in TmuxCommandRunner.run to include a 15,000 ms timeout, matching the
existing `#probeKiro` boundary. Preserve the current commandFailed and stderr
handling for timeout results.

tests/harness/live-e2e/kiro-tui.ts (1)

463-471: 🩺 Stability & Availability | 🟠 Major | ⚡ Quick win

#reapPrivateServer が恒久的なコマンド失敗を「未 reap」と誤診断します。

行 467 はサーバー消滅を commandFailed(listed) && absentPrivateServer(listed) でのみ判定します。list-sessions が別の理由で失敗する場合 — ソケットへのアクセス拒否、tmux バイナリの不在、コマンドタイムアウト — このループは同じ失敗を deadline (既定 10 秒) まで polling します。その後 "private tmux server was not reaped" を返し、cleanup barrier が PASS を落とします。サーバーは既に消えている可能性があります。

失敗の種類で分岐してください。absentPrivateServer ではない失敗は、その stderr をそのまま診断として返してください。原因の追跡が可能になります。

♻️ 提案する修正

   async `#reapPrivateServer`(): Promise<string | null> {
     const deadline = Date.now() + (this.#options.reapTimeoutMs ?? 10_000);
     for (;;) {
       const listed = this.#privateCommand(["list-sessions"]);
-      if (commandFailed(listed) && absentPrivateServer(listed)) return null;
+      if (commandFailed(listed)) {
+        if (absentPrivateServer(listed)) return null;
+        return sanitizeText(listed.stderr || listed.stdout || "private tmux server liveness is unknown");
+      }
       if (Date.now() >= deadline) return "private tmux server was not reaped";
       await this.#sleep();
     }
   }

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/harness/live-e2e/kiro-tui.ts` around lines 463 - 471, Update
`#reapPrivateServer` to distinguish an absent private server from other
list-sessions failures: retain the null return for failures matching
absentPrivateServer(listed), but immediately return the command’s stderr for any
other command failure instead of polling until the deadline. Preserve the
existing timeout message for a successful command that never confirms reaping.

tests/harness/live-e2e/journey.ts (1)

136-163: 📐 Maintainability & Code Quality | 🟠 Major | ⚡ Quick win

createClaudeTuiJourney との完全重複が残っています。共通ファクトリへ抽出してください。

行 143-147 の 5 項判定と行 155-157 の digest 式は行 107-134 と同一です。差分は id、prompt、timeoutMs、診断文、evidence.kind のみです。判定条件を片方だけに追加するドリフトが発生します。

♻️ 提案するリファクタ

+interface TuiAnchorJourneySpec {
+  readonly id: string;
+  readonly prompt: string;
+  readonly timeoutMs: number;
+  readonly evidenceKind: string;
+  readonly passedDiagnostic: string;
+  readonly failedDiagnostic: string;
+}
+
+function createTuiAnchorJourney(spec: TuiAnchorJourneySpec): LiveJourney {
+  return {
+    id: spec.id,
+    prompt: spec.prompt,
+    timeoutMs: spec.timeoutMs,
+    retryPolicy: { maxAttempts: 1 },
+    assert: (execution) => {
+      const passed = execution.exitCode === 0 &&
+        execution.structured?.anchorVerified === true &&
+        execution.structured.inputCount === 1 &&
+        typeof execution.structured.paneDigest === "string" &&
+        typeof execution.structured.sessionDigest === "string";
+      return {
+        passed,
+        diagnostic: passed ? spec.passedDiagnostic : spec.failedDiagnostic,
+        evidence: [{
+          kind: spec.evidenceKind,
+          value: digest(
+            `${execution.exitCode}:${execution.structured?.anchorVerified}:${execution.structured?.inputCount}:${execution.structured?.paneDigest}`,
+          ),
+          source: "assertion",
+        }],
+      };
+    },
+  };
+}
+
 export function createKiroTuiJourney(): LiveJourney {
-  return {
-    id: "kiro-tui-anchor-v1",
-    prompt: KIRO_TUI_PROMPT,
-    timeoutMs: 180_000,
-    retryPolicy: { maxAttempts: 1 },
-    assert: (execution) => {
-      const passed = execution.exitCode === 0 &&
-        execution.structured?.anchorVerified === true &&
-        execution.structured.inputCount === 1 &&
-        typeof execution.structured.paneDigest === "string" &&
-        typeof execution.structured.sessionDigest === "string";
-      return {
-        passed,
-        diagnostic: passed
-          ? "private Kiro TUI session, current-run file anchor, and bounded pane evidence passed"
-          : "Kiro TUI anchor mismatch",
-        evidence: [{
-          kind: "kiro-tui-anchor",
-          value: digest(
-            `${execution.exitCode}:${execution.structured?.anchorVerified}:${execution.structured?.inputCount}:${execution.structured?.paneDigest}`,
-          ),
-          source: "assertion",
-        }],
-      };
-    },
-  };
+  return createTuiAnchorJourney({
+    id: "kiro-tui-anchor-v1",
+    prompt: KIRO_TUI_PROMPT,
+    timeoutMs: 180_000,
+    evidenceKind: "kiro-tui-anchor",
+    passedDiagnostic:
+      "private Kiro TUI session, current-run file anchor, and bounded pane evidence passed",
+    failedDiagnostic: "Kiro TUI anchor mismatch",
+  });
 }

createClaudeTuiJourney も同じファクトリへ載せてください。

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/harness/live-e2e/journey.ts` around lines 136 - 163, Extract the
duplicated assertion and evidence logic from createClaudeTuiJourney and
createKiroTuiJourney into a shared journey factory. Parameterize only the
differing id, prompt, timeoutMs, diagnostic text, and evidence kind, then have
both creators use that factory so their five-part validation and digest
construction remain identical.

tests/integration/t-live-e2e-kiro-tui.integration.test.ts (2)

185-189: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

100 はマジックナンバーです。MAX_SOCKET_PATH_BYTES を参照してください。

tests/harness/live-e2e/kiro-tui.ts が同じ値を MAX_SOCKET_PATH_BYTES として持ちます。テストが独立した定数を持つと、片方の変更でドリフトします。kiro-tui.ts から export して import してください。

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/integration/t-live-e2e-kiro-tui.integration.test.ts` around lines 185 -
189,
テストのソケットパス長検証にあるマジックナンバー100を削除し、tests/harness/live-e2e/kiro-tui.tsのMAX_SOCKET_PATH_BYTESをexportして、この統合テストへimportし参照してください。既存のserverCalls検証条件は維持してください。

---

302-313: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

このテストは readiness deadline への到達を証明していません。

AMADEUS_LIVE_E2E:FAIL:EXECUTION_FAILED は他の失敗経路でも返ります。tmux コマンド失敗、pane 上限超過、anchor 不一致でも同じコードです。診断文は stderrDigest としてハッシュ化されるので、受領証から readiness 失敗を区別できません。

tmux.calls を検証してください。trust クリアの Enter が送られ、prompt の send-keys -l が送られていないことを示せば、readiness 段階で停止したと確定できます。

💚 提案する検証の追加

   test("a trust picker that never clears reaches the readiness deadline", async () => {
     const item = fixture();
+    const tmux = new FakePrivateTmux({ stuckOnTrustPrompt: true });
     try {
-      const result = await runFixture(item, new FakePrivateTmux({ stuckOnTrustPrompt: true }), 0);
+      const result = await runFixture(item, tmux, 0);
       expect(result).toMatchObject({
         ok: true,
         value: { kind: "recorded", outcome: { code: "AMADEUS_LIVE_E2E:FAIL:EXECUTION_FAILED" } },
       });
+      // Readiness never completed, so the prompt was never sent.
+      expect(tmux.calls.some((call) => call.args[2] === "send-keys")).toBe(true);
+      expect(tmux.calls.some((call) => call.args.includes("-l"))).toBe(false);
     } finally {

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/integration/t-live-e2e-kiro-tui.integration.test.ts` around lines 302 -
313, Strengthen the test “a trust picker that never clears reaches the readiness
deadline” by inspecting the FakePrivateTmux instance’s `calls` after
`runFixture` completes. Assert that the trust-clearing Enter was sent and that
no prompt `send-keys -l` command was issued, while preserving the existing
recorded failure-code assertion and cleanup.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes