Set up SpeakCode the way you work

• Belts: hand one agent’s output to the next automatically — Build agent handoffs with belt ports and three delivery modes, then chain implementation, review, and tests across running terminals.
• Claude Code hooks for done and waiting status — Add inline Claude Code and Codex hooks, verify status in the Activity tab, and understand when quiet output is only an estimate.
• Git worktrees for parallel coding agents — Create isolated branches for parallel agents, install each worktree’s dependencies, avoid shared resource collisions, and clean up after review.
• How to run multiple Claude Code sessions — Start independent terminal sessions, isolate their files, and arrange a ten-card SpeakCode workspace with status hooks and test handoffs.
• Run Codex and Claude Code side by side — Run both CLIs in neighboring terminal cards, separate their writing assignments, and hand implementation to review with a fixed belt prompt.
• Share a workspace live: roles, waiting room, terminal grants — Share a live canvas with native terminal views, admit guests with explicit roles, and grant typing access to individual terminals.