Permissions matrix, AI code review, and a daily journal pipeline
Tightened role-based permissions, wired an automated PR review workflow, and shipped a bilingual daily journal with a /daily route.
Today split cleanly across two repos — one product, one personal site — and both moved forward in ways I care about.
itop.com.br
The big one was untangling permissions on issue #403: EDITOR can now both check-in and un-check-in, the CHECK_IN role mirrors that, and EDITOR can create secret links. I shipped the fix with a role-matrix integration + e2e test suite to lock the behavior in (#414), so future regressions surface fast.
On top of that, I added a Claude Code review workflow to CI so every open PR gets an automated review pass. I also drafted a spec for a refund policy (org default with event override), and documented the WhatsApp delivery architecture — Meta Cloud API with webhook, Inngest for the queue — with a diagram attached to PR #400. Closed the day with a harness audit and an application plan for an AI-first pivot.
tgmarinho-ai-website
Reorganized the nav (About, CV, Projects, Blog, Daily, Community, Contact) and, more importantly, shipped the bilingual daily-journal pipeline with a /daily route (#97). Added architecture and privacy diagrams in mermaid so the threat model is visible, not implicit.
Also killed two mobile annoyances on the hero: the 3D parallax tilt and a heavy SVG turbulence fog that was tanking scroll perf.
The journal itself is the meta-win: proof of work, compounding.
Journal auto-generated at 23:00:00 from my work sessions with AI agents.