Default agent flow: it runs the command, you read the diff after. Patchwork flips it: you see the command before it runs, and approve from your phone if you're away from the desk.
If you watch it and want to actually run those automations on your own computer (no cloud, full control, AI that asks before touching anything important), check out Patchwork OS. It turns exactly this kind of stuff into simple recipes your AI teammate follows — inside your editor, with safety guardrails and a kill switch.
https://t.co/HRjE6CzcBu
Worth a look if you’re serious about automating for real.
https://t.co/EnFEQ82AUr
@nduatirobert_ Depends on how you use them imo.
That’s why we created Patchwork it sits as an MCP bridge between your IDE and the model, so every tool call is a message we can pause, diff, and gate before it touches disk.
Patchwork sits as an MCP bridge between your IDE and the model, so every tool call is a message we can pause, diff, and gate before it touches disk. Approval isn't a wrapper around the agent‚ it's the transport.
@amenya_nelson Once you start chaining recipes together things start getting interesting.
You set the approval policy so that agents and tasks can’t complete risky tasks or send out risky messages/emails without sending out a phone notification for approval.
@amenya_nelson You’ve gone far in a short space of time. Great stuff. Keep it up.
Interesting how similar our projects are.. however mine is privacy oriented. Let me know what you think.
@amenya_nelson@claudeai Cool project.
If you like privacy and flexibility. You can automate almost anything without your data leaving your computer.
https://t.co/LF9OOC9S0F
We fixed a bug where rate limits on Claude subscriptions weren't properly adjusted for long context requests in Opus 4.7.
We've reset 5-hour and weekly rate limits. Enjoy Opus 4.7!
A minimal reproducible example of the onDiagnosticsError → runClaudeTask loop. Save a broken TypeScript file, Claude notices, diagnoses, and proposes a fix, no prompt typed.
https://t.co/DIaN1FGOtx