A lot of the team's day-to-day operational and development work — investigations, routine changes, repetitive validation — was manual and handled differently by every engineer, which made it slow and inconsistent to onboard new team members onto.
I built a suite of AI agent skills that automate this operational and development work using spec-driven development flows: instead of an engineer improvising a one-off script or process each time, a written spec drives the agent through a repeatable, reviewable workflow. This turned fuzzy, tribal-knowledge tasks into something consistent and auditable.
Building the tools was only half the problem — the other half was making sure they didn't stay a personal habit. I led knowledge-sharing sessions to teach the rest of the team how to use and extend these skills, which turned them into everyday tools the team now relies on rather than something only I used.