Rafael (Ralph) Peña Claim this profile

Rafael (Ralph) Peña

Marine vet. CEO and co-founder of WeCheck AI (AI-driven due diligence). Miami.

I build with AI agents and I don't trust claims without evidence. Including my own.

The receipts system:

• rules-with-receipts — a Claude Code quality pack that ships with its own eval evidence, including the tests where it does nothing
• rulebench — point it at any rules file, get honest behavior deltas (its first published run declined to confirm our own headline, and its six-pack study reported one of its own rows as grader noise, with proof; that's the point)
• agent-failure-modes — the AFM Index: numbered failure modes of AI coding agents, with detection traps and evidence grades

• About to let an agent into an unknown repo? → pipx install agent-zero-trust, then azt scan . (agent-zero-trust).
• Want the operating discipline? → Install rules-with-receipts (60-second bootstrap in the README).
• Want to test your own rules file? → pipx install rulebench (rulebench).
• Want the failure taxonomy? → Read the AFM Index (start with "the five you'll see this week").

Collapse