Agent Orchestration

Claude Code: /plugin marketplace add wshobson/agents → /plugin install agent-orchestration@claude-code-workflows

Two commands and a context-manager agent handle the part of multi-agent work that usually stays informal. The improvement command starts by pulling thirty days of history — completion rate, correction count, tool-call efficiency, token spend — so a prompt rewrite can be compared against something rather than argued about. Failures get sorted into named categories such as instruction misunderstanding, context loss and tool misuse, which is what makes the next edit targeted instead of a full rewrite. The optimization command works the other direction, profiling database, application and frontend layers in parallel and treating the context window itself as a budget to compress. Production safety is explicit: changes ship with a rollback path.