zoo-browser-verification

name	zoo-browser-verification
description	Verify browser-visible Zoo behavior and capture focused UI evidence. Use only when explicitly requested.

Zoo Browser Verification

Read and follow .zoo/zoo.md if it exists.

Use this when verifying browser-impact Zoo subtasks.

Goal: prove browser-visible behavior works and collect actionable UI evidence.

Read and follow .zoo/browser.md if it exists.

Browser Method

• Run after implementation for browser-impact subtasks.
• Use the repo's browser/app harness mode when it exists and covers the needed flow.
• Record the actual mode, scenario, port, URL, viewport, and user/auth state.
• Use the harness in-page browser tool for navigation, interaction, inspection, screenshots, and evidence capture. In Codex this is Codex Browser Use; in Claude Code this is the Chrome extension MCP (mcp__Claude_in_Chrome__*, see _ai/browser-testing.md).
• Use the OS-level automation fallback only when the in-page browser tool cannot exercise the required browser-visible behavior, e.g. native browser alerts, print dialogs, permission prompts, or other OS-level UI that blocks the in-page tool. To an agent, these might look like the browser session freezing. In Codex the fallback is Codex Computer Use; in Claude Code use whatever OS-level automation is available to the harness.
• Avoid broad global keystrokes that could affect the harness conversation instead of the browser. Prefer visible dialog actions or the safest specific dismissal sequence. Codex only: sending Esc to Codex will stop yourself.
• Codex only: if using the macOS-level fallback, verify Codex has Screen & System Audio Recording plus Accessibility permissions before relying on screenshots or keyboard events.
• Codex only: in the Codex desktop app, the in-app browser and conversation share one app window; avoid global Escape because it can abort the Codex turn.
• For dropdown/popover/disclosure/select changes, prove open/dismiss/select/submit parity with screenshots per impacted flow.
• Save screenshot evidence for each impacted flow or acceptance checkpoint, including before/after states when the transition matters.
• Record video evidence when screenshots cannot prove behavior, such as animation, transient states, or timing-sensitive interactions.
• Record important browser-testing learnings immediately in the relevant durable notes before continuing; browser sessions can be interrupted or compacted, and exact recovery steps are easy to lose.
• Keep artifacts focused and named clearly.
• If verification cannot be completed, stop, mark the result failed, and explain exactly what blocked the flow and what you tried.

Report Handoff

Include Final-state validation in the browser-verification report:

• browser checks run against the current implemented state
• result on that exact state: passed, failed, or not run
• whether any repo file changed after those checks
• if browser validation did not pass, the exact flow or command the orchestrator should rerun after fixes