Numbered screen context for OpenAI Codex

Codex is powerful when the task is clear. UI work fails when the model has to guess which button or panel you meant from a dense screenshot.

1. Hotkey into Casso and draw boxes on the target UI.
2. Attach notes like "match spacing to [1]" or "remove double border on [2]".
3. Capture. Paste the prompt and image reference into Codex.
4. Iterate with the same overlay when the next visual pass needs corrections.

• Built for CLI agents first: numbered regions and a paste-ready prompt, not an IDE plugin.
• Not locked to one vendor - switch between Codex and Claude Code without relearning tools.
• Structured visual handoff: numbers in the prompt match numbers on the image.
• Privacy-first: screenshots stay on your machine until you choose to paste them into your agent.
• 7-day free trial, then $29 one-time personal license - not a subscription.

Codex performs best when the task boundary is obvious. Pair Casso marks with acceptance checks: "After the change, [1] should have 16px padding and no double border." That turns a visual complaint into a verifiable request.

Prefer one issue per box. If layout relationships matter, use two boxes and say "match [2] to [1]." Lead the paste with the goal, then the map - same pattern that works well with Claude Code.

Keep sensitive production data out of the frame when you can. Casso stays local until paste; once you paste into Codex, OpenAI product terms apply to that turn.

Because Casso is not embedded in any single agent, you keep the same overlay when you hop from Codex to Claude Code the same afternoon.