Faultkeep: an error tracker for teams shipping web apps
A paste-in snippet reports uncaught errors, grouped by stack trace into live issues with charts, triage and email alerts.
Agent specs
- Agent
- Claude Code, Opus 5.5 Medium
- Time
- 32 min
- Cost
- $8.16*
Agents work faster and more accurately on Elements.
* Cost at API rates, September 2026.
Run it yourself
Paste this prompt into your AI agent. You get the finished app on your own machine, ready to change however you like.
Install elements: curl -fsSL 'https://install.elements.dev?os=darwin' | sh && export PATH=~/elements/bin:$PATH Create an Elements app from the elementscode/demo-faultkeep scaffold: cd ~/elements/projects elements create faultkeep -scaffold=elementscode/demo-faultkeep cd faultkeep Start the app.
How it's built
Faultkeep needed an ingest endpoint and a paste-in snippet, grouping by stack trace, live issue lists and charts, triage, and alert emails. Each of those is a part of Elements, so the agent spent its 32 minutes on error tracking itself.
What Elements gave the app
- An ingest endpoint and snippet. One route takes JSON error events authenticated by the project's key, and another serves a snippet any web page can include to report uncaught errors and unhandled rejections.
- Grouping in SQL. A database function fingerprints each event by its stack frames, then updates the issue's counts, affected users, tags and hourly chart in one call. An event on a resolved issue reopens it.
- Live issues and charts. Issues, events, tags and chart buckets are LiveTables, and a database trigger broadcasts every ingest, so the issue list, each issue's chart and its latest events update as errors arrive.
- Triage. Resolving, ignoring and assigning save through the issues LiveTable, which checks the user and the assignee are members of the project.
- Alert emails from a job. A new or reopened issue schedules a job inside the ingest transaction, and it emails every project member from a template.
- Data from SQL files. Migrations define the schema and seed two users, two projects, 15 issues and about 1,860 events over the past week, some resolved, some ignored and one reopened. The project server applied each one as soon as it was saved.
What the project server gave the agent
The project server runs alongside the agent and answers as soon as a file is saved: it type-checks the templates, TypeScript and SQL, applies migrations and reruns the tests, so every question came back right away and the agent kept building.
What shipped
The app type-checks with zero errors and all 24 tests pass. Every page works on desktop and phone, and live updates arrive as errors come in, such as an uncaught error on a page with the snippet raising its issue's count on an open issue list.
The code
Build it from scratch
To watch your agent build this app from an empty project, use the prompt we used instead. It takes longer than starting from the finished app above.
Install elements: curl -fsSL 'https://install.elements.dev?os=darwin' | sh && export PATH=~/elements/bin:$PATH Create an Elements app: cd ~/elements/projects elements create faultkeep cd faultkeep Start the app, then build this: Build an error tracker named faultkeep for web apps. - Accounts and projects. Each project has a DSN key. - An ingest endpoint that accepts JSON error events (message, stack trace, url, browser, release, user) authenticated by the key, and a small snippet to paste into a web page that reports uncaught errors to it. - Group events into issues by stack trace. Issue list: count, users affected, first and last seen, sorted by last seen or frequency. - Issue detail: the stack trace, a chart of events over time, tags, and the latest events. - Resolve, ignore, or assign an issue. A resolved issue that happens again reopens. - Email when a new issue appears or a resolved one comes back. Seed two users, two projects, and a week of events across about fifteen issues. Show the seeded logins on the sign-in page. New events and issues appear in real time.
Comments· 0