model-routing plugins

by Spotify

Shunts I/O-heavy work to AiKA modes to save tokens. Hooks block large file reads and redirect to bulk-reader/code-writer scripts backed by the Portal CLI.

by Seyed Yahya Shirazi

Complete project lifecycle toolkit: read-only startup preflight, bounded native PR review panels, tracked cross-agent project memory, initialization, cross-agent user-instruction setup, tiered model routing, epic/sprint workflow, onboarding, planning, debugging, engineering loop, agent fan-out, CI/CD, Docker, security, and document processing

by Alberto Leal

Install and maintain a "model routing" section in a project's CLAUDE.md that teaches Claude Code which model to use for which work, including delegating to GPT models via the Codex CLI. Use when adding model-picking guidance to CLAUDE.md, setting up Codex/GPT as a delegation target or fallback, or routing subagent and workflow tasks across models by cost, intelligence, and taste.

by Melodic Software

Workflow and model-routing guidance for multi-agent work: assess (workflow, subagent or single context), route (the role map a workflow script reads from args, keeping fan-out stages off a frontier model), audit-defaults (rechecks bundled defaults against upstream, or sweeps the repo's model and workflow guidance, via a read-only drift-audit workflow), check (whether node resolves and the fetch gate is registered), and setup (writes a user, team or local layer after a preview and a yes).

by Matt Corbett

Cross-tool AI-coding model guidance — three strategist agents that help you reason about model selection in the non-Claude coding ecosystems: GitHub Copilot's model picker (completions / chat / coding agent / cloud agent + org model rules), OpenAI Codex (CLI + cloud model lineup, reasoning levels, /model), and xAI Grok (Grok 4.x lineup + the coding-tool surface). Teaches the DURABLE reasoning — task-complexity → reasoning-level → cost-tier → surface — over a single citation-grounded, dated lineup that is the marketplace's freshness anchor for non-Claude model facts. Volatile specifics (prices, context windows, picker availability) carry retrieval dates + a re-verify-at-use rider, and the agents fetch live before quoting. Seam: when the answer is a CLAUDE model or a Claude-app build, escalate to claude-app-engineering/claude-solution-architect — this plugin owns the non-Claude tools. Requires ravenclaude-core@>=0.7.0.

by TemMax

Dual-host development orchestration for Claude Code and Codex: super-plan creates wave-ready tasks with machine-checkable contracts, multi-model executes isolated supervised waves through the host-native adapter, and ship carries approved plans through reviewed delivery while leaving merge authority with the user. Exact Claude, GPT-5.6, and GPT-6 Astra/Sol/Luna profiles provide evidence-scoped guidance with a conservative generic fallback.

by anshss

Self-improving multi-agent harness that grinds your ticket backlog across every repo in your product — guarded auto-merge on green CI, and every resolved ticket writes a lesson into your git-tracked CLAUDE.md. Every ticket starts on a cheap tier and escalates once, on measured failure, instead of being predicted expensive up front; deterministic fixes skip the model entirely; and the lessons file evicts rather than accumulates, so always-on context stays a fixed budget.

by YPY

Claude is the architect: it writes specs and verdicts and picks a reasoning effort per task. The relay model (follows your codex config) on an OpenAI-compatible relay does all the implementation — agentically through the Codex CLI by default, or as chat completions applied verbatim as the fallback — and a clean-context Claude advisor reviews every deliverable before it ships.

by Ivan Gudak

Shared foundation for the dev-workflows plugin family: the addressing grammar and specs-repo git entry points, phase handoff, model routing, escalation and finding-triage rules, cost/feedback/follow-up emission, and the reference loader skill that lets a dependent plugin read them. Also carries the family-meta utility commands — /feedback, /prompt, /prompt-brainstorm, /prompt-grill-me, /statusline — and /frames for indexing design frame sets.

by TokenLab

TokenLab model selection, API integration, native endpoints, cost routing, and OpenAI-compatible migration skills with optional TokenLab MCP servers.

by Andre Paat

Multi-model delegation for browser testing — Kimi K2.5 (via opencode) for execution, Opus for analysis

by claude-skills

Routes work from an expensive lead model to tool-scoped worker agents -- explorer and researcher on Haiku, editor and test-runner on Sonnet, refuter and investigator on Opus -- and makes the lead re-verify what they report instead of trusting it. Adds fabflows:build, a deterministic loop in which an Opus builder implements a spec and a fresh reviewer returns ACCEPT or REWORK. Ships an ACTIVE guard hook that blocks package installs, commits and pushes to a default branch, destructive shell commands, and reads or writes of credential files; the guard fails open, so a bug degrades to no guard rather than a blocked session.

by Alexander Degtyarev

Planner, Coder, Reviewer pipeline with a different model per role. Route implementation to a cheap model and review to a strong one.

by dd106-lang

Keep the expensive lead model on planning and review; route routine work to pinned cheaper workers (Haiku, Sonnet, Opus) and audit which models actually ran.

by Roy Rojas

Orquestación 'arquitecto' para Claude Code: el modelo principal clarifica, diseña y revisa; subagentes con el modelo que vos configurás implementan. Compuertas (hooks) en PowerShell 7 que hacen cumplir el protocolo. Integra graphify + Obsidian.

by agentmc15

Model routing, relative-burn estimation, usage analysis, and verified execution workflows for Codex.