Code Agent Security & Observability

Building the Code Agents UI from scratch

There was no existing surface for this — I built the Code Agents module end to end: navigation and agent discovery, usage dashboards, threat and violation review, command-level analysis, and the onboarding flow that gets a new workspace instrumented. It's the module teams open first to see what their Cursor, Claude Code, and other AI agents have been doing.

Reviewing threats down to the exact tool call

Flagging a threat isn't useful unless a reviewer can see exactly why. I built the event detail view that shows the tool call an agent made, which policy it tripped, and the detection signals behind the score — so a security reviewer can confirm a block, or catch a false one, in seconds instead of digging through raw logs.

Catching problems before they were reported

On the legacy platform, PostHog was already integrated. During a routine check of the data, I noticed users were struggling to get through signup. The terms-and-conditions checkbox was easy to miss, leaving the submit button disabled with no visible reason.

Integrating PostHog into the new Studio platform surfaced a different kind of problem: browser ad blockers can silently prevent events from being recorded, which undermines the reliability of the analytics itself. I traced the cause and identified a Next.js proxy approach as a way to route around it.

One definition of "threat," adopted everywhere

There was no consistent definition of what counted as a threat across the platform. I drove alignment with engineering and product on a single, platform-wide definition instead of leaving it to vary by surface.