fix(ai): 处理 rebase 后 Codex 新增的三项审阅意见

- catalog CHANGELOG 删除 rebase 重放进已发布 [16.1.4] 段落的重复 Claude 4.6 条目，使该段落与上游 main 完全一致（已发布段不可变）
- Duo Agent finally 清理在最终 idle timeout（重试已耗尽）时也发送 stop PATCH，避免代理/LB 持续断连场景下服务端工作流残留
- auth-broker login 仅对 pasteCodeFlow provider 传入 onManualCodeInput，普通 loopback provider 不再让 readline 提示与 HTTP 回调竞争导致终端残留

packages/ai/src/providers/gitlab-duo-workflow.ts

@@ -1229,15 +1229,16 @@ async function runGitLabDuoWorkflow(
 		// The socket loop can exit several ways that leave the remote workflow running
 		// and `active` referencing a dead socket: a user abort; `runGitLabDuoWorkflowSocket`
 		// rejecting (e.g. `ws.onerror`) so the settle block never ran (`settledNormally`
-		// stays false); or the socket closing before any terminal status arrived
-		// (`lastSocketResult === "closed"` — a proxy/server drop). In all of these the
+		// stays false); or the socket reached a half-open/stuck terminal state with no
+		// real completion — `lastSocketResult === "closed"` (proxy/server drop) or
+		// `"timeout"` (idle deadline, retry already exhausted). In all of these the
 		// local stream is finalized but the server workflow has no explicit stop, so drop
 		// the resumable session and stop it with a FRESH signal (the request's own signal
 		// may be aborted, which would cancel the PATCH before it is sent). The happy path
 		// that intentionally keeps `active` for an `action`/`pause` resume reaches a real
-		// terminal status, never "closed", so it is not affected.
+		// terminal status, never "closed"/"timeout", so it is not affected.
 		const aborted = options.signal?.aborted ?? false;
-		if (aborted || !settledNormally || lastSocketResult === "closed") {
+		if (aborted || !settledNormally || lastSocketResult === "closed" || lastSocketResult === "timeout") {
 			if (providerSessionState) {
 				providerSessionState.active = undefined;
 			}

packages/catalog/CHANGELOG.md

@@ -98,7 +98,6 @@
 
 - Fixed Claude 4.6 routing on the `google-antigravity` (and `google-gemini-cli`) Cloud Code Assist providers, whose backend exposes the models asymmetrically: `claude-sonnet-4-6` has no `-thinking` twin and `claude-opus-4-6` has only the `-thinking` twin. The shared `thinkingPair` family was routing thinking efforts on `claude-sonnet-4-6` to a non-existent `claude-sonnet-4-6-thinking` wire id (404 `Requested entity was not found`); replaced both 4.6 entries with bespoke single-wire families that declare the dead ids as `retiredMembers` so `reconcileRetiredRouting` re-points stale bundled-catalog and SQLite-cache rows away from the 404 wire id. Refreshed the bundled `models.json` Sonnet 4.6 entry whose stored `effortRouting` still targeted the dead `-thinking` id. Added `claude-sonnet-4-6` and `claude-opus-4-6-thinking` entries to `ANTIGRAVITY_MODEL_WIRE_PROFILES` capped at the backend's 64000-output-token limit (over-cap requests 400'd with `Request contains an invalid argument`); `modelEnum` is now optional on `AntigravityModelWireProfile` since the Claude wire ids are accepted without a captured `labels.model_enum`. ([#3067](https://github.com/can1357/oh-my-pi/issues/3067))
 
-- Fixed Claude 4.6 routing on the `google-antigravity` (and `google-gemini-cli`) Cloud Code Assist providers, whose backend exposes the models asymmetrically: `claude-sonnet-4-6` has no `-thinking` twin and `claude-opus-4-6` has only the `-thinking` twin. The shared `thinkingPair` family was routing thinking efforts on `claude-sonnet-4-6` to a non-existent `claude-sonnet-4-6-thinking` wire id (404 `Requested entity was not found`); replaced both 4.6 entries with bespoke single-wire families so every effort and off resolve to the live wire id. Added `claude-sonnet-4-6` and `claude-opus-4-6-thinking` entries to `ANTIGRAVITY_MODEL_WIRE_PROFILES` capped at the backend's 64000-output-token limit (over-cap requests 400'd with `Request contains an invalid argument`); `modelEnum` is now optional on `AntigravityModelWireProfile` since the Claude wire ids are accepted without a captured `labels.model_enum`. ([#3067](https://github.com/can1357/oh-my-pi/issues/3067))
 ## [16.1.3] - 2026-06-19
 
 ### Fixed

packages/coding-agent/src/cli/auth-broker-cli.ts

@@ -28,6 +28,7 @@ import {
 	type OAuthCredential,
 	type OAuthProvider,
 	type OAuthProviderInfo,
+	PASTE_CODE_LOGIN_PROVIDERS,
 	PROVIDER_REGISTRY,
 	SqliteAuthCredentialStore,
 } from "@oh-my-pi/pi-ai";
@@ -211,6 +212,12 @@ async function runLocalLogin(provider: OAuthProvider): Promise<void> {
 	const storage = new AuthStorage(store);
 	await storage.reload();
 	try {
+		// Only paste-code providers (fixed non-loopback redirect, e.g. GitLab Duo
+		// Agent's vscode:// URI) get the manual paste fallback. For normal loopback
+		// providers `onManualCodeInput` would make OAuthCallbackFlow race a readline
+		// prompt against the HTTP callback; if the callback wins, the outstanding
+		// prompt is never cancelled and leaves the terminal in a dirty/blocked state.
+		const usesManualInput = PASTE_CODE_LOGIN_PROVIDERS.has(provider);
 		await storage.login(provider, {
 			onAuth({ url, instructions }) {
 				process.stdout.write(`\nOpen this URL in your browser:\n${url}\n`);
@@ -223,12 +230,13 @@ async function runLocalLogin(provider: OAuthProvider): Promise<void> {
 			onPrompt(p) {
 				return ask(`${p.message}${p.placeholder ? ` (${p.placeholder})` : ""}:`);
 			},
-			onManualCodeInput() {
-				// Providers with a fixed non-loopback redirect (e.g. GitLab Duo Agent's
-				// vscode:// URI) never hit the local callback server, so offer the same
-				// paste-the-redirect fallback the interactive TUI sign-in uses.
-				return ask("Paste the authorization code (or full redirect URL):");
-			},
+			...(usesManualInput
+				? {
+						onManualCodeInput() {
+							return ask("Paste the authorization code (or full redirect URL):");
+						},
+					}
+				: undefined),
 		});
 		process.stdout.write(`\nCredentials saved to ${getAgentDbPath()}\n`);
 	} finally {