workstation
The machine layer: a reproducible Apple Silicon macOS setup built from manifests rather than migrated from the last laptop, with per-project runtimes and security boundaries for working across several client environments at once.
Setup
How this machine is put together, laid out the way a Unix system lays out its own parts. The arrangement is not decoration: each directory answers a different question, and keeping them separate is what stops the answers from drifting into each other.
Tree
- bin/ The commands I actually type, and the few I refuse to make easy.
- etc/ How the machine is configured, and the reasoning behind each choice.
- lib/ The context, rules, and skills that steer how I work with AI tools.
- opt/ The applications, and what each one earns its place for.
- sbin/ Security posture — where keys live and how secrets stay out of git.
Source
The boundary between the two is deliberate: llm-context owns
portable agent behaviour, a project owns its own local context, and
workstation owns machine setup and private values.
llm-context is public. workstation is private by
design — a precise description of one person's tooling and access patterns is
more useful to an attacker than to a reader, so GitHub returns a 404 unless
you are signed in with access. The pages above carry the reasoning without
the operational detail.
llm-context
The portable layer: agent rules, personal context, tool adapters, enforcement hooks, skills, and the brief-to-handoff workflow shared across Claude Code, Codex, and the repos that publish selected content.