The skills and workflows I use to give agents context, keep changes focused, and get a second review.
My role
Hands-on development, AI orchestration, and workflow design
When
Jul 2026–present
Status
Used on client work and shared across development teams
Codex
Claude Code
BMAD
Independent review
The problem
A change can touch application code, data, and infrastructure in separate repositories. Coding agents need to understand those connections while staying focused on the task.
My work
I use coding agents in projects I help implement, leading and orchestrating AI use alongside hands-on software development. I built on BMAD and added custom skills for preparing features, planning, implementation, and independent review.
I also lead a biweekly developer meeting focused on using AI effectively and keeping the team current. We share practical experiences and discuss new models, agent harnesses, capabilities, and use cases, learning from one another about what could help our work.
How the pieces fit togetherA simplified view of my workflow. The internal tools stay separate from client code, and review starts in a fresh agent context.
Internal tools & skills
The reusable preparation, planning, and review tools live here.
Application · Data · Infrastructure
Agents can see the relevant codebases without copying internal tools into client projects.
A focused change
An agent implements a defined task with the context it needs.
A fresh review
Another agent context checks the change. I review the findings and decide what’s ready.
Engineering decisions
Share context without copying internal tools
I set up a separate internal repository that gives agents access to related codebases. Our reusable skills stay there instead of being copied into client projects.
Review from a fresh context
I ask a separate agent session, sometimes using a different model, to question assumptions made during implementation. I check its findings alongside tests and my own review.
Technical details
One example: a cloud-to-local migration
The project needed to move a group of cloud applications and supporting services onto local physical hardware because of security and internet-connectivity constraints. My contribution was scoping and development-method guidance around an approximately six-week turnaround requirement. The team carried out the migration.
Where things stand
I apply these workflows in my own implementation work and lead AI use on those projects. Independent reviews have caught issues missed in earlier pull-request reviews.
The biweekly developer meeting provides a regular place to exchange practical experience and stay current with changes in AI tools, models, and capabilities.
What I’m still working on
As models improve, I’m trying simpler instructions and checking whether the extra planning helps catch problems.