Kick off a cloud run — AI testers audit your URL and report back in minutes.
⚡ Full testone-shot · ~minutes
⊙ Smart-diff monitorbaseline + recurring
Standard1 cr / page
Deep3 cr / page
Superwide1.5 cr / page · wide
Super-deep4 cr / page · flows
Complete5 cr / page · wide+deep · hours
Dailyrecommended · ~0.2cr/quiet day
Hourlypower-user · higher cost
Weeklyrecommended · 1× per 7 days
Every tickaggressive · matches recurrence
▸ Baseline is captured immediately on submit (one-time free). Each subsequent tick reports only the meaningful changes vs the baseline — AI-classified structural / content / timely-noise, persona-judged. Daily is the sweet spot. dashboard →
Crawls ~every page one hop from your URL and exercises the actions on each. Default 25.
Runs the top business-critical user flows and executes each this many steps deep. Default 15.
From the team that found jank at scale for Google Chrome
Removes the jank from anything you build with AI.
Code, docs, or a live URL — 10 AI jank-finders, ~3 min. Works in Claude Code, Claude Cowork, and OpenAI Codex.
Works in Claude Code, Claude Cowork, OpenAI Codex, Antigravity & Cursor.
Install in Claude Code
Adds /jank, /jank_light, /jank_test, /jank_explore, /jank_cloud, /risk, and /risk_auto as typed slash commands.
2 — Add it in Codex → Settings → MCP servers → “Connect to a custom MCP” (STDIO)
Command to launch: node
Argument: ~/.jank/mcp/jank-mcp-server.js
3 — Use it
Ask Codex: “jank this repo” or “quality-check my changes.” It calls the jank tools and runs the same triage → report.
Tip: add a one-line note to ~/.codex/AGENTS.md — “for quality checks, call jank_get_playbook and follow it” — so plain “jank this” auto-triggers the full flow.
Kick all cloud runs from inside your dev console /jank_cloud
Perform deep, parallel runs in the cloud of the code you're building locally — results in minutes while you keep coding, then Claude auto-fixes every issue it finds.