2026, AI agents
A shared library of agent skills
One workspace of reusable skills for Claude Code and Codex, covering test-driven development, code review, bug diagnosis, domain modeling and research, plus a router skill that maps how they fit together.
Every project I work on reads from the same set of skills at the root of one workspace, so a good habit written down once is available everywhere.
How it is organized
- Engineering skills for daily code work: TDD, code review against the repo's own standards, a diagnosis loop for hard bugs, domain modeling with ADRs, research against primary sources
- Productivity skills for everything around the code: handoffs, grilling a plan until it holds, writing documents that agents can follow
- A router skill that knows every user-reachable skill and when to use it
- A script that links every skill into the local harness directories
The same workspace holds a Markdown second brain, so what I learn in one project is there for the next.
AI agents