How do I send a UI fix back to Claude Code with full context?

Use Vessel's 'Dispatch' surface. Any inspected element plus the comment thread on it becomes a structured prompt containing the target file path, the surrounding code, the conversation, the inspected computed styles, and any design-token deltas you applied. One click ships it to Claude Code.

What's in the prompt

When you dispatch, Vessel constructs a prompt with five clearly labelled sections:

1. Target: the file path and line range you inspected (e.g. app/checkout/Button.tsx:42-58).

2. Current code: the JSX/TSX snippet around the target, exactly as the agent will see it. No paraphrase.

3. The thread: every comment in the conversation, with author + timestamps, so the agent knows the disagreement and the resolution.

4. Inspected styles: the computed font, spacing, color, transform — the actual rendered values, useful when the conversation describes the change in visual terms ("add 4px to the bottom").

5. Token deltas: if you tuned design tokens live (e.g. --space-md from 12 → 14), the delta is in the prompt so the agent can apply the change as a token edit, not a one-off inline value.

Why structured beats free-form

Free-form prompts ("can you fix the spacing on the trust badge row in the checkout please") have to be re-grounded by the agent every time. It re-grep's the file, infers what "trust badge row" means, tries to find which spacing value the human meant. Latency and error rate both go up.

Vessel's dispatched prompt is already grounded: file is named, code is quoted, conversation is included, target style is measured. Claude Code's job is straightforward: apply this change to this code. Round-trips drop from 3-5 to 1.

Targets other than Claude Code

The same Dispatch payload also exports as a Markdown brief that pastes into Cursor, GitHub Copilot Chat, or any text-input agent. The format is plain enough that it's also a good handoff to a human engineer when you'd rather have one in the loop.