nine local-first tools that govern AI agents — decide, enforce, detect, consolidate, observe
an allow / deny / ask policy engine for agent actions, written from scratch in Rust
runtime enforcement for MCP tool calls — a transparent stdio proxy
a clinical record backend where authorization is the feature
evidence-backed autopsy reports and revival plans for stalled repos
an orientation layer for unfamiliar codebases
a local tripwire for capability drift in a diff
content-pinning and drift detection for terraform & opentofu modules
a documentation drift scanner that checks doc claims against the actual repo
install-time hash verification for CPAN dependencies
elder-care coordination for families doing it themselves
a family command center — calendars, tasks, and a daily digest
workplace-violence compliance for California employers (SB 553)
one owned queue for everything stuck between Shopify and ShipStation
break-compliance auditing for warehouse teams
an operations desk for SoCal event-market vendors
a reef of its own, surfacing soon
numbers, but playable — surfacing soon
explainable academic workload, before it turns into burnout
n-of-1 nutrition experiments that show their work
staged return-to-play evidence, with the named decision behind it
a client-intelligence layer for personal trainers
a client-side recovery advisor for your Apple Health export
The coding interview reframed as preregistration — judging how a candidate bounds ambiguity before the first token, once AI is allowed in the room.
Vibe coding, enabled coding, and the line between owning a system and merely accepting it — and why the agent's gravity runs toward the cage.
Caring for an aging parent is really a coordination problem. What research shows about shared goals, shared knowledge, and safer handoffs in care.
Family caregiving is more than errands. Research shows caregivers manage ADLs, IADLs, medical tasks, finances, appointments, and constant coordination.
What I learned coordinating care for family — and why it turned into software.