coding-principles

Purpose

Single source of implementation discipline for coding tasks. Loaded as active context so the agent applies the same rules every time without the user re-stating them per session.

Design philosophy

Built for AI agents authoring code, optimizing for first-pass correctness — right and simple and tested in one draft. The human "make it work, then right, then fast" sequence is deliberately rejected: it would license shipping the works-but-ugly version. Two operating rules follow: severity (must / should / could) governs triage, and mantra tier (Goals > Design > Pruning) governs conflicts. When in doubt, prefer the interpretation that yields simpler, more typed, more testable, more observable code in this draft over one that anticipates future flexibility.

What this skill is

A rulebook the agent reads before and during code changes — scope, abstraction, error handling, comments, and compatibility decisions flow through its rules. It is read-only: it does not modify files, emit a report, or run linters, formatters, or refactor tools; it shapes how the Edit/Write calls that follow happen. Adjacent concerns (post-edit cleanup, repo-local conventions, change narration) are out of scope — see "Scope boundaries" below.

File layout

This SKILL.md is the always-loaded router — the mantra/principle summaries, checklist, and anti-patterns below cover routine work. Load deeper material on demand: