manage-nodeterm-canvas

Use spawn-team with a JSON array of roles and prompts to open one agent per role in a labeled group, or open-worktree per branch followed by open-agent with the group id. Each station works on its own branch, and you read results through the context links created automatically.

Pass --after with the upstream node id when opening the downstream agent; it stays armed and launches itself once every listed station goes idle. Waiting only works on agent nodes that report status, not plain terminals.

Yes, use open-agent with --agent codex, gemini, opencode, or a custom agent id, plus optional count, cwd, prompt, and group flags. Claude sessions can also be opened directly with open-claude.

No, the skill only works inside a controllable nodeterm Claude session. If the CLI reports canvas control is unavailable, the session cannot be controlled and retrying will not help.

In SSH projects, nodes run on the host but media viewers read the desktop's disk, so show-video and show-web with local files are refused for host paths. Use --url instead, or copy the file to the desktop first.

Run verify with the target node id to open a review panel: one reviewer per lens (default correctness, security, tests) armed behind the target, plus a judge that merges findings into a single verdict. Reviewers only read the work and do not modify files.