A durable, MCP-native job relay: agents dispatch work, and a Claude Code session is woken (via a channel) when it finishes — no tmux.
Agent
Part of mcp-agent-relay
Use for reasoning-heavy phases — architecture decisions, debugging complex/subtle issues, algorithm design, evaluating trade-offs between approaches. Think thoroughly, but return a concise conclusion the orchestrator can act on directly, not a transcript of the reasoning.
Agent
Part of mcp-agent-relay
A second, independent reasoning lens — run in parallel with deep-reasoner (and codex, when available) on plan reviews and high-stakes decisions, never as a solo replacement for deep-reasoner. The value is a different model family scrutinizing the same problem, not a faster or cheaper deep-reasoner.
Agent
Part of mcp-agent-relay
Use for mechanical, well-specified work — boilerplate, repetitive edits across files, writing tests for already-decided behavior, formatting, straightforward refactors. Not for anything requiring judgment calls or architecture decisions; that goes to deep-reasoner instead.
Agent
Part of mcp-agent-relay
Independent read-only Opus max reviewer for the dual-r panel; challenge invariants and architecture.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: