Programmer building software systems that humans and agents can understand, change, and verify.
A curious life in software systems: architecture, tools, notes, and experiments.
AI coding does not make architecture less important. It changes what architecture must control.
When editing becomes cheap, the expensive parts become understanding, validation, review, rollout, and recovery.
| Project | Role | What it explores |
|---|---|---|
| Agent-operable Architecture | Architecture | How codebases can expose truth, boundaries, proofs, and safe change paths for humans and agents |
| Agent Skill Control Theory | Agent control | A framework for designing, evaluating, and evolving LLM agent skills |
| DBX Skills | Practice | Claude/Codex skills for planning, reviewing, architecture health, and engineering workflows |
| db.oneir.xyz | Writing | Notes on AI coding, architecture, programming foundations, and software judgment |
- Make source of truth explicit.
- Keep semantic changes locally understandable.
- Bind success claims to executable evidence.
- Treat repository context as part of the architecture.
- Report architecture findings by future failure mode, not by taste alone.
Codebase side Agent-operable Architecture
Agent side Agent Skill Control Theory
Workflow side DBX Skills
Writing side db.oneir.xyz
TypeScript 8 hrs 35 mins ██████████░░░░░░░░░░░░░░░ 40.03 %
Markdown 6 hrs 11 mins ███████▒░░░░░░░░░░░░░░░░░ 28.83 %
Other 1 hr 36 mins ██░░░░░░░░░░░░░░░░░░░░░░░ 07.48 %
OCaml 1 hr 16 mins █▒░░░░░░░░░░░░░░░░░░░░░░░ 05.92 %
JavaScript 1 hr 9 mins █▒░░░░░░░░░░░░░░░░░░░░░░░ 05.42 %