claude-configlisted

# Claude Configuration Primitives Three distinct purposes in `.claude/`. Each has a clear job -- never conflate them. Agents are their own file type. Commands and skills are both `SKILL.md` files, told apart by who is allowed to invoke them. ## Agents = Roles An agent defines **who you are**. It shapes focus, identity, and constraints for a subagent. - Role identity ("You are a senior backend reviewer") - Tool and permission constraints - What the agent cares about (evaluation criteria, output format, confidence thresholds) - Runs in an isolated subagent context **Agents are not workflows.** They define a lens for approaching work, not steps to follow. A backend-reviewer says "you care about async correctness and session handling" -- not "Step 1: read the diff." **Good:** Role identity, focus areas, evaluation criteria, output format, what to read for context. **Bad:** Step-by-step procedures, bash scripts, workflow orchestration. ## Skills = Context and Knowledge A skill defines **what you know**. It provides conventions, recommendations, how-tos, and reference material. - Reference knowledge ("Here's how migrations work in this project") - Conventions ("Module files follow this structure") - Recommendations ("When writing docs, prefer tables over prose") - Templates and examples when they're the core value **Skills are not workflows.** They provide knowledge for good decision-making. A skill MAY include a sequence when it's critical reference -- but as illustrati