Telemetry that asks, and telemetry that doesn't
Two telemetries with opposite consent defaults on one OTLP layer: consent-gated analytics for a CLI, implied-consent observability for a service.
Tags
23 pages
Architecture
One graph, not micro-stacks
Why an infrastructure repo stays one OpenTofu graph rather than splitting into per-concern stacks: the engine enforces ordering for free.
Why I hand-rolled every module
Why security-critical infrastructure got raw resources rather than community wrapper modules, and what that actually cost in lines and in review.
Supporting a provider, or actually using it
A lowest-common-denominator AI interface throws away what each provider is good at. When to route through an abstraction and when to drop below it.
Two kinds of feature flag
Runtime feature flags decide which commands are reachable on a given run. Cargo features decide what is compiled in at all. Do not confuse them.
Reloading config without a restart
Hot-reloading configuration: watch the file, re-read it, swap it in atomically and notify observers, without restarting a long-running service.
Registering commands without life before main
Rust has no init or pre-main phase, so self-registering commands need another mechanism. Using a distributed slice to get the same shape.
What survives a port, and what doesn't
Porting a Go framework to Rust separates design from idiom. Which decisions survived the move, which were habits, and how to tell them apart.
An AI interface that fits on one screen
You do not need LangChain in a command-line tool. A four-method chat interface can hide five providers, tool calling and structured output.
Middleware for CLI commands, not just web servers
Timing, auth, recovery and logging are cross-cutting in a CLI too. Wrapping Cobra commands in web-style middleware rather than PreRun hooks.
