What problem does it solve? Coordinating multiple AI agents (Claude, Codex) on shared work requires a durable communication layer: spawning peer workers, routing messages, reviewing each other's output, and debugging stalled sessions. This Skill teaches agents how to use the trellis channel CLI correctly so multi-agent collaboration happens through a persistent event log instead of ad-hoc copy-paste. ## Core Features & Use Cases - Worker orchestration: Spawn Claude or Codex workers with agent cards, inject context files and JSONL manifests, then dispatch tasks and wait on done/turn_finished system events. - Forum channels: Create durable --type forum boards for issue tracking, design feedback, and internal changelogs with threads, labels, statuses, and context entries. - Progress debugging: Diagnose stalled workers using raw event inspection, worker logs, PID sidecar files, and the OOM guard, avoiding truncated pretty output. - Use Case: A dispatcher agent spawns two parallel check reviewers (one Claude, one Codex) on a task's PRD and design docs, waits for both with --all, then records the consolidated findings as a thread on a global forum channel. ## Quick Start Ask the agent to brainstorm a design decision with a Codex peer over a Trellis channel, spawning a worker and running multiple pressure-test rounds before summarizing the result.
