ops-browserlisted

# Ops Browser ## Overview Operate browser pages and collect evidence without conflating browser surfaces. The ChatGPT desktop built-in browser, ChatGPT cloud/agent browser, controlled Chrome, and isolated managed automation have different state, login, download, visibility, and background guarantees. Select from capabilities proven in the active environment; leave frontend code changes to `dev-frontend`. ## Workflow 1. Identify the target hostname, path, environment, account/session, and task goal. 2. Preflight only task-required capabilities and return the Capability Snapshot from `references/browser-operation-protocol.md`. Set unselected availability fields to `unknown` and explain `not assessed: outside selected preflight scope` in `gaps.reason`; expand the matrix only for authenticated, state-changing, transfer, or delegated review work. 3. Enumerate browser sessions and existing tabs only when the available tool exposes them; never invent missing tab/window identity. Imported bookmarks, history, and saved credentials may accelerate target discovery or user login, but do not prove an active session, account/workspace identity, conversation ownership, authorization, or operation state. 4. When called by `ask-chatgpt`, validate the Handoff Request fields, reuse or refresh the named Capability Snapshot, and return a Handoff Result with the same `operation_id`; do not reconstruct bridge policy locally. 5. Choose the surface mode and evidence plan based on capabil