Output is currently rendered as plain text via `src/shared/formatter.ts`. This works for clipboard-paste workflows, but programmatic consumers (CI tools, coding agents, extensions) need structured formats.
Proposed formats:
- JSON — machine-readable capture payload. Element selector, component info, annotations as coordinate arrays, transcription segments with timestamps. Direct input for tool integrations.
- Markdown — formatted for pasting into GitHub issues, PR descriptions, or docs. Annotations rendered as inline images (base64 data URIs of the canvas snapshot).
- MCP-compatible — structured as an MCP resource or tool response, so coding tools that speak MCP can consume captures directly.
Design:
The formatter should become a strategy pattern:
```typescript
interface OutputFormatter {
format(session: CaptureSession): string;
mimeType: string;
label: string; // shown in sidepanel dropdown
}
```
The sidepanel gets a format selector dropdown. Default remains plain text for backward compatibility.
Implementation notes:
- JSON schema should be versioned (`"version": 1`) so consumers can handle format evolution
- Markdown format needs to handle missing fields gracefully (no component info? skip that section)
- MCP format depends on the MCP spec stabilizing — may want to defer that variant
Related to NLnet milestone M6 (output formats and integrations).
Output is currently rendered as plain text via `src/shared/formatter.ts`. This works for clipboard-paste workflows, but programmatic consumers (CI tools, coding agents, extensions) need structured formats.
Proposed formats:
Design:
The formatter should become a strategy pattern:
```typescript
interface OutputFormatter {
format(session: CaptureSession): string;
mimeType: string;
label: string; // shown in sidepanel dropdown
}
```
The sidepanel gets a format selector dropdown. Default remains plain text for backward compatibility.
Implementation notes:
Related to NLnet milestone M6 (output formats and integrations).