Problem

When using a remote CDP profile (e.g., Browserless), ensureTabAvailable() throws "tab not found" after the persistent Playwright connection drops and reconnects.

Reproduction scenario

1. OpenClaw connects to a remote Browserless instance via connectOverCDP() → Chrome fix: add @lid format support and allowFrom wildcard handling #1 is created
2. A tab is opened → lastTargetId = "abc123" is stored in profile state
3. The persistent CDP WebSocket drops (network issue, Browserless session cleanup, etc.)
4. onDisconnected fires → cached = null → Chrome fix: add @lid format support and allowFrom wildcard handling #1 is killed by Browserless
5. Next browser request → connectBrowser() → fresh connectOverCDP() → Chrome Login fails with 'WebSocket Error (socket hang up)' ECONNRESET #2 (new process, no previous tabs)
6. The AI agent passes targetId = "abc123" from the previous session
7. resolveById("abc123") returns null — Chrome Login fails with 'WebSocket Error (socket hang up)' ECONNRESET #2 has no tab with that ID
8. Error: "tab not found"

Root cause

In server-context.ts, ensureTabAvailable() has a graceful fallback for the extension driver when a stale targetId doesn't match and only one tab is available:

let chosen = targetId ? resolveById(targetId) : pickDefault();
if (!chosen && profile.driver === "extension" && candidates.length === 1) {
  // Recover by using the single attached tab
  chosen = candidates[0] ?? null;
}

This fallback does not apply to remote CDP profiles (!profile.cdpIsLoopback), even though the same stale-targetId scenario is common with remote browsers like Browserless where each new WebSocket connection spawns a fresh Chrome process.

Proposed fix

Extend the single-tab fallback to also cover remote CDP profiles:

if (!chosen && (profile.driver === "extension" || !profile.cdpIsLoopback) && candidates.length === 1) {
  chosen = candidates[0] ?? null;
}

The reasoning is the same as for extension relay: if an agent passes a stale/foreign targetId but we only have a single page available, recover by using that page instead of failing hard.

Environment

• Remote CDP via Browserless v2 (Chrome image) on Fly.io
• Persistent connectOverCDP() connection with TIMEOUT=-1
• Connection drops trigger fresh Chrome creation (no session persistence in Browserless v2 + Playwright)

Related issues

• [Bug]: Isolated OpenClaw browser (Brave) frequently becomes unreachable during automation (“browser control service timed out”), sometimes leaving CDP port stuck #10994 — AbortSignal fix for stuck CDP connections (helps with reconnection, but doesn't address stale targetId)
• Browser extension relay: tab not found when multiple tabs have same URL #1998 — tab not found with extension relay (same findPageByTargetId code path)
• Browser extension relay returns stale tab cache after CDP connections die #6175 — stale tab cache after CDP connections die