Staff Software Engineer at Google building private clouds. I started in programming, spent years in networking & security, and now work across both — a developer with a deep respect for infrastructure.
Lately I build agent-native developer tools: command-line tools designed to be driven by LLM agents — and the standard that defines what "agent-native" actually means.
aclig.dev — a living, versioned standard for CLIs an LLM agent can discover, drive, and trust: read-only by default, structured & self-describing, token-bounded, injection-fenced. Every rule is pinned to the assumption about agent capability behind it, so the standard evolves as the models do.
→ The invariants · Conformance · Why
Each one is read-only by default, speaks clean JSON, self-describes via schema, and ships an
embedded agent guide — so an agent can drive it safely with no hand-holding.
| Tool | What it does | Install |
|---|---|---|
| rivr | Read-only Amazon shopping for agents — one schema over 4 backends | brew install rnwolfe/tap/rivr |
| gfly | Google Flights search, JSON-first, no API key | uvx gfly |
| knit | Agent-safe CLI for Threads — posting gated in the binary | brew install rnwolfe/tap/knit |
| vabc | Virginia ABC catalog & live store stock | brew install rnwolfe/tap/vabc |
| nxstate | Read-only Cisco Nexus (NX-OS) state as clean JSON | uvx nxstate |
All conform to the Agent CLI Guidelines (Full) — <tool> schema declares the exact spec
version it targets, machine-verifiable at runtime.
- 🏠 A residential homelab (Linux + a Mac mini) running self-hosted everything.
- 🤖 Tooling for a household AI agent.
- ⚙️ Agent-native design patterns — non-interactive flows, stable exit codes, bounded output.
- The standard → aclig.dev
- Tools → pinned repos below