PlaythroughFor reviewing what your agents build
The agent says it’s done. You decide if it’s good.
Playthrough records your agent’s change the way a person would use it, before and after, including what it didn’t mention. Click what’s wrong, or take over and try it yourself.
Show me this change with npx playthroughPaste it to Claude Code, Cursor or Codex while it works. Open source, on your machine, with nothing to sign up for.
Invoices
$2,760.00 overdue
| Invoice | Customer | Due | Reminded | Amount | Status |
|---|---|---|---|---|---|
| INV-1041 | Harbor & Finch | Jul 14, 2026 | — | $1,840.00 | Paid |
| INV-1042 | Northwind Studio | Jul 30, 2026 | — | $3,650.00 | Paid |
| INV-1043 | Alder Dental | Aug 12, 2026 | — | $920.00 | 79 days late |
| INV-1044 | Kestrel Logistics | Aug 29, 2026 | — | $5,100.00 | Paid |
| INV-1045 | Harbor & Finch | Sep 15, 2026 | Oct 2, 2026 | $1,840.00 | 45 days late |
| INV-1046 | Juniper Books | Oct 3, 2026 | — | $460.00 | Open |
| INV-1047 | Northwind Studio | Oct 28, 2026 | — | $2,975.00 | Open |
| INV-1048 | Alder Dental | Nov 9, 2026 | — | $920.00 | Open |
Send a reminder
0:00 / 0:09
Send a reminder
0:00
A real review, played back. Leave a note, hold Compare to see the round before, or take over: from that moment on, it’s the app.
All of the judging, none of the setup.
Checking an agent’s change today
- 1Read the agent’s summary, then the diff
- 2Start the app on its branch
- 3Sign in and get the data right
- 4Click through to each change
- 5Hunt for what it changed without saying
- 6Describe what’s wrong, and check again next time
With Playthrough
- 1Watch the flows it touched
- 2Click what’s wrong, or take over and try it
- 3Judge the next try, beside the last
What the agent changed without saying is marked, so it can’t hide in a long diff.
Try it on your agent’s next change.
Open source, and it runs on your machine, with any agent that has a shell: Claude Code, Cursor, Codex and others. Reviews stay in your repo; nothing is uploaded.
Show me this change with npx playthrough