Launching today

Agent Suite Builder
Plan, build, test and gate AI agents in Claude Code
1 follower
Plan, build, test and gate AI agents in Claude Code
1 follower
SKILL.md files for Claude Code, Cursor and Codex CLI. Describe an agent system in one line. Your assistant plans it, builds it, tests it and runs it on your own accounts. Nothing pays, sends or goes live until a named person says yes.







Payment Required
Launch Team

AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
Hi Product Hunt, I'm Vlad, and I made this.
This is the procedure I wanted a coding assistant to follow when it builds an agent that can send or pay: plan first, put the approval check in code, then try to break it.
That procedure is 48 skills. The plan comes first: a blueprint with 12 decisions, where every outside fact cites a URL marked CONFIRMED or UNVERIFIED. Then come code with tests the assistant runs itself, testers who try to break the suite, a trial week and a one-command roll-back.
Five worked examples ship with 387 passing tests. One is a four-agent insurance claims suite with a planted bug: a hail claim that should pay $12,400.00 comes out at $8,000.00, and the Debug job finds the first wrong step.
Two honest limits. I built and tested it in Claude Code, and no run in Cursor or Codex is on record yet. The examples run on fictional data.
What would you want a plan like this to settle before any code exists?