Luisa Lima

Agentic engineering

The craft of building and operating with agents: the stack I actually run, the method for growing it, the tooling that keeps it honest, and the task layer that keeps work in the diff. Engineering practice, not theory.

Ongoing · 4 chapters

Working well with agents is its own engineering discipline. It is quickly becoming a new craft: how you set up, grow, and operate the system day to day so it augments you, instead of increasing your technical debt.

Everything here is something I run, and a ledger of my agentic engineering practice as I go.

I’ll be adding more as the practice evolves.

  1. 01. My current "secure" agentic coding stack

    I value diversity of agents and isolation, I live in the terminal.

  2. 02. Earn your abstractions

    I built a publishing system by hand with the agent first. Once it was battle-tested, I extracted the hooks and skills. Like in code, you can't abstract what you haven't done by hand at least twice.

  3. 03. My agent policy kit

    I was maintaining the same operating rules in a different config for every coding agent I run. So I built one shared policy and skill set that installs across all of them.

  4. 04. I switched from beads to plaintext tasks that live in the diff, managed by a skill

    I wrote opentasks-skill to teach my coding agents to manage tasks without external dependencies, with a git-tracked audit system.