Bug Description

The wait --url command does not support glob patterns, despite the documentation and help text showing glob examples. This is a regression introduced during the Native Rust rewrite (f341c0a).

Current Behavior

wait --url uses JavaScript String.includes() for matching, which treats glob characters (*, **) as literal strings. This means any pattern containing wildcards will never match.

agent-browser open https://github.com/settings/profile --headed
# Browser redirects to login page, user logs in, page navigates back

agent-browser wait --url "**/settings/profile"
# BUG: never resolves, times out

The built-in help text (agent-browser wait --help) shows glob examples that don't actually work:

  agent-browser wait --url "**/dashboard"

Expected Behavior

Glob patterns should work as they did in the previous TypeScript implementation, which used Playwright's page.waitForURL() with native glob support:

// Old implementation (src/actions.ts)
await page.waitForURL(command.url, { timeout: command.timeout });

Playwright's matching semantics:

• No wildcards → exact match
• * → matches any characters except /
• ** → matches any characters including /
• {a,b} → alternation

Root Cause

In the Rust rewrite, wait_for_url (cli/src/native/actions.rs) was implemented as:

let check_fn = format!(
    "location.href.includes({})",
    serde_json::to_string(pattern).unwrap_or_default()
);

This passes the raw glob pattern to JS String.includes(), which does literal substring matching only.

The same issue also affects:

• Route matching (simplified glob logic that only handles a single * split into prefix/suffix)
• response-body URL matching (uses String::contains())

Environment

• agent-browser version: 0.23.0
• Introduced in: f341c0a (Native Rust rewrite of agent-browser daemon)
• Previously worked in: v0.15.x and earlier (TypeScript daemon using Playwright)