ops-browser-verification

name	ops-browser-verification
description	Use when verifying ops-agent changes that affect console runs, terminal initialization, asset terminal context, asset command execution, approvals, SSE rendering, or plan mode behavior in the browser.

Ops Browser Verification

Overview

Verify Ops Agent in a real browser against the local app. Type checks and builds do not prove console, terminal, asset execution, approval, SSE, or plan-mode behavior works.

Required Startup

Identify the OS before starting from the repository root:

OS	Command
Windows	scripts\\run.bat
macOS/Linux	./scripts/run.sh

Confirm backend 127.0.0.1:8000 and frontend http://localhost:5173 are reachable. Never use ./scripts/run.sh on Windows.

Required Browser Flows

1. Terminal initialization multi-turn

Verify both paths when available:

Path	Required observations
Normal terminal	Open/create a terminal, start an assistant conversation, ask an initial terminal-related prompt, then at least one follow-up. Confirm the follow-up keeps the correct terminal context and streamed messages render in order.
Asset terminal	Select an asset, open its terminal, start a conversation, then ask at least one follow-up. Confirm asset identity and terminal session context persist across turns.

If local data has no usable asset, state that explicitly and verify the normal terminal path.

2. Asset command execution from conversation

From an asset-scoped conversation, request a safe read-only command that requires backend asset execution. Confirm:

• approval UI appears when required
• approval triggers backend execution, not chat-only pretending
• command card status changes are visible
• output or failure reason renders in the conversation
• browser Console and Network show no unexpected errors

Do not bypass the human approval chain.

3. Plan mode conversation

Enable plan mode and ask for a small operational plan. Confirm:

• plan content renders instead of command execution
• commands mentioned in the plan are not run automatically
• plan events render correctly
• a follow-up stays in plan mode, or the UI clearly shows the mode change

Report Evidence

Report only observed evidence:

Area	Evidence
Startup	OS and command used
Terminal multi-turn	normal terminal result and asset terminal result
Asset execution	command requested, approval behavior, final status
Plan mode	prompt used and whether execution was avoided
Browser health	Console/Network errors, if any

Common Mistakes

Mistake	Correction
Running only pyright or pnpm build	Still verify all three browser flows.
Using ./scripts/run.sh on Windows	Use scripts\\run.bat.
Testing only normal terminal	Also test asset terminal when an asset is available.
Asking for dangerous commands	Use safe read-only commands and preserve approval.
Trusting chat text as execution proof	Verify command card, approval, backend request, and result.
Treating plan mode as normal chat	Confirm plan mode does not execute commands.