---
id: "11020ca4-868e-4e58-8142-b216053d389d"
name: "Existing Tab Workflow"
micro_summary: "Owns tab targeting and the boundary between default isolation and human-authorized interaction."
status: "resolved"
see_also: []
scope: "Choose and operate browser tabs while preserving the user's work by default and honoring explicit task-scoped human overrides."
warnings: "Call `list_tabs()` first and verify target identity before acting. Protect the active tab by default; honor clear, scoped human authorization using this node's procedure. Tab authorization does not authorize unrelated actions. Distinguish instruction defaults from observed tool limits."
---
## Existing Tab Workflow

This procedure governs tabs the agent drives by hand. The sibling named Decision-Model Fast Path owns the loop's separate hidden-tab requirements.

1. Call `list_tabs()`. Identify the requested site, matching targets and whether the intended target is active or visible. Use target-scoped reads to inspect state without changing it.
2. By default, reuse a suitable non-active matching tab. If only the active tab matches, use a dedicated tab and leave the active tab unchanged. Avoid replacing an existing authenticated matching page with unnecessary navigation.
3. The human may override this tab workflow, including authorizing interaction with an active tab. Bind the override to the identified target, requested actions and stated duration; absent a longer duration, it lasts only for the current task. If those are clear from the conversation, acknowledge the scope and proceed without asking again. Clarify only missing or ambiguous scope. Do not silently make an override permanent or substitute a copy for an explicitly authorized existing target.
4. Attach with `switch_tab(target, activate=False)` or target-scoped tooling. Verify the target before mutations; if the target or page changes outside the authorized scope, stop affected actions and re-establish the target.
5. Verify the requested effect and report what actually changed. End the task-scoped override at completion and return to default isolation for subsequent work.

An override changes workflow defaults, not higher-priority safety requirements, permission for unrelated actions or actual tool limits. If a tool rejects an authorized action, report the observed limit and use another authorized route when available. Do not describe a prompt-level default as a technical prohibition.

Saved instructions are not proof of fresh-agent compliance; behavioral verification is separate.