Multi-Agent Parallel Workflow

Use this skill when work should be parallelized through Trellis multi-agent pipeline.

Goal

• Keep orchestrator work in the main repo.
• Run code implementation inside linked worktrees.
• Track each task with task metadata and registry records.

Quick Start

python3 ./.trellis/scripts/get_context.py
python3 ./.trellis/scripts/multi_agent/plan.py --name "<feature>" --type "<backend|frontend|fullstack>" --requirement "<requirement>"
python3 ./.trellis/scripts/multi_agent/start.py "<task-dir>" --platform <claude|opencode|codex>
python3 ./.trellis/scripts/multi_agent/status.py

Platform Notes

• claude and opencode: background CLI agent is started automatically.
• codex: manual mode. The script prepares worktree and context, then continue inside Codex in that worktree.

Completion

• Monitor progress with status.py.
• Cleanup worktrees with cleanup.py after tasks are merged or archived.