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
Markdown 3 hrs 27 mins ███████▓░░░░░░░░░░░░░░░░░ 30.52 %
OCaml 2 hrs 55 mins ██████▒░░░░░░░░░░░░░░░░░░ 25.86 %
JSON 1 hr 10 mins ██▓░░░░░░░░░░░░░░░░░░░░░░ 10.45 %
TypeScript 1 hr 8 mins ██▓░░░░░░░░░░░░░░░░░░░░░░ 10.12 %
Other 50 mins ██░░░░░░░░░░░░░░░░░░░░░░░ 07.47 %