What problem does it solve? Go services often scatter log calls across layers, producing duplicate lines for the same failure, leaking sensitive values, and mixing business rejections with real errors. This Skill enforces a convention where only the outermost boundary (Connect interceptor, worker supervisor, external intake) logs each event exactly once, with a fixed attribute vocabulary and secret redaction. ## Core Features & Use Cases - Boundary-only logging: Defines which layers log (interceptors, supervisors) and which never do (domain, repository), with the rule "if you return it, don't log it; if you swallow it, log it." - Middleware patterns: Provides a Connect logging interceptor that recovers panics, translates errors to connect.Code, derives log level from the code, and a ctx-aware slog.Handler wrapper that adds request_id and caller attributes. - Severity and redaction rules: Supplies a level table (business rejections are Info, only unexpected failures are Error), a snake_case attribute vocabulary, and a four-point redaction strategy using primitives only, JSON handler, and ReplaceAttr. - Use Case: When asked to add logging to a Go worker that expires holds one by one, the Skill places per-item failure/success logs in the worker's Run loop while cycle-level errors return to the supervisor, verified with go vet and grep checks. ## Quick Start Apply the write-logs convention to add slog logging to my Connect interceptor and worker so each RPC is logged exactly once.
