Marker.io

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me the bug reporter that replaces Marker.io.

The value is not the screenshot. It is that **the report lands in the tracker the developers already live in**, complete enough that nobody has to ask a follow-up question.

STACK
- Node 20+ with Fastify
- SQLite through better-sqlite3, WAL mode
- A small capture script, vanilla JS
- Caddy in front

THE DATA MODEL
- reports: id, project_id, url, title, body, reporter, created_at, remote_ref, remote_status
- context: id, report_id, viewport, dpr, browser, os, console_json, network_json, screenshot_path, annotations_json
- projects: id, name, script_key, tracker, tracker_config_encrypted
- Store the remote id and the remote status: **a report whose fix is invisible from here gets reported again next week**

CAPTURE
- A button on the page; it screenshots, then lets the reporter draw on it — a rectangle, an arrow, a blur
- **The blur tool matters more than the arrow.** People screenshot pages containing real customer names, and giving them a way to hide it is what makes the capture safe to file
- Collect: URL, viewport, device pixel ratio, a coarse browser and OS bucket, the last console errors, and failed network requests. Never the raw user agent, never input values
- If the screenshot fails, submit anyway with everything else

FILING INTO THE TRACKER
- One adapter per tracker behind a small interface: create, update status, add comment. Three methods
- The issue body carries the context as a readable list, and the screenshot as an attachment — not a link that dies when your service does
- **Idempotency on submit**: a double-clicked button must not open two issues
- Poll or receive webhooks for status, and store it. The reporter sees their report move without an account

THE REPORTER, WHO IS NOT A DEVELOPER
- No account. A name once, remembered locally
- One field: what went wrong. Everything else is captured for them
- Confirm with the issue reference and a link to check it later. **A reporter who never hears anything stops reporting**

OPERATIONS
- .env: DATABASE_PATH, BASE_URL, ENCRYPTION_KEY, per-tracker credentials
- Screenshots on object storage with signed expiring URLs, and a retention sweeper
- Tracker credentials encrypted at rest; a failed sync retried with backoff and shown, never silently dropped

WHAT MATTERS MOST
The blur tool and storing the remote status. One makes capture safe on a page with real data; the other closes the loop so the same bug is not filed twice.