Module tabs 

Named-tab orchestration: SQLite + the engine-agnostic TabBackend working together.

The registry layer (crate::registry::tabs) is pure SQL CRUD. This module adds the logic the agent contract actually requires:

• Get-or-create: tab_open(_, name, url) returns an existing live tab under that name, or creates a fresh about:blank (or the given url) and registers it.
• Navigate-on-mismatch: if the existing tab’s last_url doesn’t match the requested url, run navigate first.
• Sweep-on-read: stale rows whose target_id no longer exists in the live browser are dropped before they’re returned to the caller.
• Budget pressure: when daemon-created rows exceed HARD_CAP, the LRU is closed and recycled.
• Cute names: agents who pass no --name get tab-<cute-word> from the same generator that names browsers.

The same code path serves CDP (Chromium) and BiDi (Firefox) browsers via TabBackend — CDP targetId and BiDi context are both opaque ids stored in the tabs.target_id column. The registry doesn’t care which engine produced the id.

Constants

HARD_CAP

Hard cap on daemon_created rows per browser. Hitting this triggers LRU close+recreate of the oldest daemon-created tab. Chromium itself starts to struggle around a few hundred tabs depending on RAM; 50 is comfortably under that ceiling and large enough that even busy scraping agents won’t routinely hit it.

Functions

resolve_tab

Resolve <browser>/<name> to a live tab row for cross-command routing (eval, fetch, etc.). Returns Ok(None) if no row matches or the row’s target_id no longer exists in the browser (TabNotFound at the call site).

tab_list

tab list <browser> backend with sweep-on-read.

tab_open

Open a named tab (get-or-create), navigating if url differs from the existing tab’s last_url. Returns the up-to-date row.

with_named_tab_recovery

Run op against the named tab <browser>/<name> with one round of recover-and-retry on tab failures. Mirrors crate::session::with_scratch_recovery but for the agent-owned named-tab path.