What problem does it solve? Teams running multiple coding agents (Claude Code, Codex) struggle to keep shared doctrine, per-agent instruction files, and plugin adapters consistent, and often misunderstand when a running agent picks up a new or edited skill. ## Core Features & Use Cases - Cross-agent mental model: Explains the author-once-render-per-agent doctrine layer, where shared doctrine plus per-agent specifics produce one instruction file per agent (CLAUDE.md, AGENTS.md). - Skill reload lifecycle: Clarifies that SKILL.md edits in watched directories are picked up live mid-session, while new top-level skill directories and non-SKILL plugin components need a restart or /reload-plugins. - Background agent handling: Documents why claude --resume <id> --bg branches a new session and how claude attach plus /reload-plugins reloads a background agent under the same session id. - Use Case: When building a plugin with Circus, use this skill to decide where knowledge belongs (personal source vs. distributed artifacts) and to run circus plugins build, check, and versions --check before committing. ## Quick Start Ask the agent to explain how Circus briefs multiple coding agents from one doctrine source and whether a newly edited skill requires a session restart.
