What problem does it solve? Coordinating multiple coding agents across parallel tasks is chaotic: ad-hoc subprocesses get no isolated worktree, no visible state, and no way to message each other. This Skill teaches an agent to drive Rove's rove api CLI so every piece of delegated work gets its own git worktree, branch, tracked lifecycle, and a reliable messaging channel between agent sessions. ## Core Features & Use Cases - Task routing and creation: Maps user intent ("try it 3 ways", "in this workspace", "let codex take over") to the right verb — add for new isolated tasks, send --tab new for a helper in the same checkout, pane-open for splits. - Parallel attempts and comparison: add --count N or --agents claude:2,codex:1 spawns parallel attempts; collect --group gathers their outcomes for comparison. - Peer messaging between agent sessions: send is the only sanctioned channel for agent-to-agent messages, with [ROVE PEER] provenance prefixes and tab-precise reply addressing. - Lifecycle and issue tracking: rename, set-branch, land (merge), delete tasks, plus a daemon-owned kanban issue tracker and cron-style routines. - Use Case: Inside a Rove session, a user says "try three approaches to the caching bug and compare them" — the agent runs rove api add --repo "$PWD" --count 3 --prompt "...", then rove api collect --group <groupId> to report the outcomes. ## Quick Start Ask the agent to check whether $ROVE_TASK_ID is set and then use rove api to create, message, or collect Rove tasks for the work you describe.