rl0ve/work-router
Routes substantial agent work by task shape, ambiguity, risk, latency, and token use, with tables for Claude Code and Codex. Includes capability discovery, skills ecosystem audits, plugin distribution audits, and scoped deduplication of installed skills.
Inspect the active Codex or Claude Code host, router components, model catalog, and workspace constraints. Run only when the user explicitly asks to check or troubleshoot the router setup.
Audit and consolidate duplicate or overlapping installed skills and plugins across Codex, including standalone-to-plugin consolidation. Distinguish copies, variants, useful specialists and inactive remnants before scoped changes.
Find missing agent capabilities or review how existing plugins and skills are distributed and updated. Compare maintained upstream packages, preserve plugin preferences and integrations, and avoid duplicate installs.
Choose model, effort and delegation for agent work. Use when picking or changing a model or effort level, deciding whether to delegate or run parallel agents, planning long autonomous or multi-agent work, or when the user asks to save tokens or time. Ordinary single-session tasks stay in the parent without this skill. Do not delegate trivial work.
Review the user's bookmarks, GitHub starred repositories, and current discussions against their full skills and plugins ecosystem to recommend worthwhile additions or changes. Use for ecosystem improvement audits rather than a single skill edit or installation.