← ClaudeAtlas

knowledge-registrylisted

Use when an investigation is wasted by a misread data model, a confusing legacy behavior, or a non-obvious gotcha - and when writing code near one. Maintains a living registry of the traps that have burned past investigations (easy-to-misread models, misleading names, load-bearing quirks) so the same mistake never costs time twice. Read the relevant entry before diving into a known-treacherous area; add an entry whenever a surprise costs real time. Triggers on "this burned me before", confusing or legacy behavior, a field/status that doesn't mean what it says, repeated investigations into the same area.
scott-garvin/claude-code-guardrails · ★ 0 · AI & Automation · score 70
Install: claude install-skill scott-garvin/claude-code-guardrails
# knowledge-registry Every codebase has a handful of places that mean something other than what they look like. The cost isn't the first investigation. It's the fifth person, or the same agent next week, re-deriving the same gotcha from scratch. ## Rule Keep a registry of the traps. Each entry records three things: what's misleading, what it actually does, and how to work with it safely. Before working in a known-treacherous area, read its entry. When a surprise costs you real time, write the entry so it never costs anyone that time again. ## What earns an entry - A model, field, or status that doesn't mean what its name implies. - A "catch-all" default that quietly masks several distinct states. - Behavior that's correct but counterintuitive, and therefore easy to "fix" into a bug. - Anything that has burned an investigation more than once. ## Navigate by name, not line number Reference code by symbol or method name. Line numbers drift and turn a registry stale within a sprint; the meaning it captures outlives the layout. This is an error-registry discipline. It's what lets an agent, or a new engineer, skip the traps you already paid for instead of paying for them again.