{{ t.f1t }}
{{ t.f1b }}
Guides
A bubble is a sub-agent. An edge is a handoff. Vibisual reads the graph you draw and launches the matching Claude Code workflow.
Updated 2026-08-18
Double-click empty canvas to place a bubble, then fill in its panel. Every field maps to something the CLI already understands, so a graph is a visual spelling of a configuration you could have written by hand.
| model | opus / sonnet / haiku, per agent |
| permission mode | default · acceptEdits · auto · dontAsk · plan · bypassPermissions |
| tools | allow-list, e.g. Read, Bash, Grep |
| isolation | shared tree · own git worktree |
| max turns | hard stop for runaway loops |
| rules | free text appended to that agent’s prompt |
Draw an edge from A to B to say that B starts when A hands off. An agent with no inbound edge starts with the run; an agent with an unresolved inbound edge stays idle and is drawn with a dashed line.
Edges are not documentation. Vibisual holds the downstream agent until the upstream one closes, which is what keeps a doc-writer from documenting a build that failed.
Give each writing agent its own worktree when their file sets overlap. Read-only agents can share the main tree safely — a test-runner restricted to read-only src/** cannot undo the work it is checking.
Isolation is enforced by the harness, not by the model. It holds even if an agent decides it would rather edit something else.
{{ t.nameLine }}
{{ t.sub }}
{{ t.canvasCaption }}
{{ t.vdeLead }}
{{ t.f1b }}
{{ t.f2b }}
{{ t.f3b }}
{{ t.instLead }}
{{ t.instSrcBody }}
Contributing →{{ t.statusBody }}
{{ t.secBody }}
{{ t.ctaSub }}
{{ ui.blogLead }}
{{ art.excerpt }}
{{ pa.v }}
{{ doc.lead }}
{{ pa.v }}
{{ s.note }}
{{ t.ideAnnoLead }}
{{ an.b }}
{{ an.k }}