Leo Giovanetti

Architect and builder who ships with AI agents under tight supervision: scoped PRs, verified in the real app, staged migrations and short human-sounding comms.

Strongest on AI agent workflows, migration planning, code review and shipping, product launch and monetization, stakeholder communication. From 133 sessions, May 2026 – Sep 2026. Version 3, updated Sep 29, 2026.

Matched 17.5 of 22 of Leo’s real corrections from sessions it never saw. A generic “senior engineer” prompt matched 17; the plain model, 15.5. Self-reported 2026-09-29. How it’s measured.

$29 one-time, includes every future version

Buy for $29

Needs the mentorfile plugin for Claude Code or Codex. Install it.

A few of Leo’s principles

P37 Done means seen working in the real app, on the preview and where the reporter saw it, and confirmed live after release.
P75 Visual consistency is checked to the pixel in a real browser, and an element whose fixes keep failing gets redesigned (a pointed-out mismatch gets the precise fix; redesign only after the same fix keeps failing).
P67 Short and sweet, then shorter; point to where the details already live (cut technical detail, never the collaborative tone or the next step).
P40 Every change headed to QA carries a test guide on the ticket, even a refactor meant to change nothing visible (whenever QA could observe a regression; otherwise P41).
P77 Users never do by hand what the system can do for them.

83 more in the full mentorfile, plus the playbooks Leo follows when debugging, reviewing and shipping.

Using it

Once installed, ask /ask-leog in Claude Code (or $ask-leog in Codex) to review a plan, a diff or a decision, or just say “ask leog about this”. Answers start with “Simulated leog”, cite the principle behind each point, and say so when Leo’s sessions don’t cover the question.