Connor Holly

Library

AI Skills

Reusable patterns and workflows I give my AI agents

What These Are

AI agents start every session with zero memory. They don't remember what worked last time, what patterns your codebase follows, or what mistakes to avoid. Skills solve this. Each skill is a document that teaches an agent how to do a specific type of work: the architecture, the gotchas, the verification steps, the decisions that matter.

Think of them as procedural knowledge that persists across sessions. When I tell an agent to "run the overnight PR factory," it doesn't improvise. It loads the skill, follows the pattern, and produces consistent results. The skill encodes everything I've learned about that workflow so the agent doesn't have to rediscover it every time.

· · ·

How I Use Them

I have 30+ agents running at any given time. Some run on cron schedules, monitoring production health, syncing data, checking for anomalies. Others are reactive, triggered by events like Sentry alerts or customer support escalations. But most of my day-to-day is managing agents directly in the terminal. Spinning them up, giving them context, reviewing their output.

Skills are keyword-triggered. When I mention "evaluate" or "pairwise comparison," the evaluation skill loads automatically. When I say "menu bar app," the SwiftUI skill loads. The agent gets exactly the context it needs without me having to explain the same architecture for the twentieth time.

Every skill started as a one-off solution to a real problem. It got used 3-5 times, proved its value, then got formalized. If I find myself explaining the same thing twice to an agent, it becomes a skill.

· · ·

Why This Matters

The bottleneck with AI agents is context, not intelligence. At least for most things. The real shift is going from doing the work yourself to understanding how to scaffold and harness agents to do it. It's less about writing code and more about knowing how to give agents the right context, the right constraints, and the right verification steps. Skills are how I do that systematically instead of ad hoc.

These are primarily built for terminal-based agents. The patterns are architectural though, not tool-specific. How to structure multi-agent orchestration, how to design evaluations that actually measure what matters, how to gather context cheaply before spending on expensive model calls. They work regardless of which model or framework you're using.

· · ·

Most Used

· · ·

All Skills

Agent Orchestration

Evaluation & Quality

Developer Tools

Content & Docs

Productivity

← Back to index