/supervibe-ui

/supervibe-ui

Starts a local browser control plane for Supervibe's native work-item graph. The UI is IDE-neutral: every host can open the same localhost URL, and IDE webviews can wrap it without changing the canonical JSON graph.

Invocation

/supervibe-ui --file .supervibe/memory/work-items/<epic-id>/graph.json
/supervibe-ui --port 3057
/supervibe-ui --daemon
/supervibe-ui --foreground

Equivalent local command:

npm run supervibe:ui -- --file .supervibe/memory/work-items/<epic-id>/graph.json --daemon
npm run supervibe:ui -- --foreground --file .supervibe/memory/work-items/<epic-id>/graph.json
npm run supervibe:ide-bridge -- --file .supervibe/memory/work-items/<epic-id>/graph.json --out .supervibe/ide-bridge.json

--daemon is the normal local control-plane launch: it starts a detached hidden process, returns the URL and PID, and writes logs under .supervibe/servers/. --foreground keeps output attached for debugging.

What It Shows

• Epic/work-item status groups: ready, blocked, claimed, deferred, review, done.
• Graph tree for epic, task, and subtask hierarchy with rollup status.
• Ready queue, blocker list, stale claims, and completion blockers derived from the same native graph used by loop and status commands.
• Workflow phase rail derived from real graph/run state: plan, atomize, execute, verify, close, archive. It marks completed phases, current phase, and blocked execution/verification instead of using decorative static labels.
• Kanban board for epics, tasks, subtasks, active agent claims, blockers, verification counts, and work movement across ready/claimed/blocked/review/done.
• Individual work items with quick selection.
• Context Pack preview for the selected work item.
• Loop state.json summary with current wave, gates, tasks, reports, and dashboard model.
• RAG, project memory, and CodeGraph health with separate visual tabs.
• SLA report preview from the active work-item graph.
• Work-item GC preview.
• Raw JSON output for debugging and IDE webview adapters.
• IDE bridge descriptor for wrapping the same localhost URL in VS Code, Cursor, Zed, JetBrains, or any host that can render a local webview.

Safe Actions

The UI can perform local-only graph actions:

• claim
• defer
• close
• reopen
• skip
• cancel
• create
• edit
• split
• reparent
• dep-add
• dep-remove
• delete

Every action supports preview first. Real mutations require an explicit confirm=apply-local apply request after preview. The server binds to 127.0.0.1 only and does not call provider CLIs, MCPs, network trackers, deployment targets, or external APIs.

Safety Boundaries

• Bind to 127.0.0.1 only.
• Preview before local graph mutation; apply requires confirm=apply-local.
• Do not call provider CLIs, MCPs, network trackers, deployments, billing, account, DNS, or external APIs.
• Store server logs and bridge descriptors under .supervibe/.

Local JSON Endpoints

• GET /api/graph?file=<graph.json>
  • Includes graphTree plus panels.readyQueue, panels.blockers, panels.staleClaims, and panels.completion.blockers so UI clients can show hierarchy, next work, stalled claims, and close blockers without guessing from raw columns.
  • Includes kanban.graphSummary, kanban.epics, kanban.agents, and kanban.columns[] so IDEs can show task movement and epic ownership.
  • Includes flow.steps[], flow.activeId, flow.status, and flow.metrics so IDEs can show phase progress from graph status, task status, gates, and archive markers.
• GET /api/index-status
• GET /api/context-pack?file=<graph.json>&item=<item-id>
  • Includes workflowSignal and flow so the selected task's context pack carries the same phase, epic, gate, claim, and next-action signal that the UI shows.
• GET /api/run?file=<state.json>
  • Includes the same flow model derived from loop status, waves, gates, reports, and task completion.
• GET /api/report?file=<graph.json>&type=sla
• GET /api/gc
• POST /api/action with preview-first body; apply uses confirm=apply-local

Output Contract

SUPERVIBE_UI
URL: http://127.0.0.1:<port>/
BIND: 127.0.0.1
AUTH: localhost-only
IDE_WIDGET: npm run supervibe:ide-bridge -- --port <port> --file <graph.json> --out .supervibe/ide-bridge.json

Confidence: N/A Rubric: read-only-research

Agent Orchestration Contract

This command must load its executable profile from scripts/lib/command-agent-orchestration-contract.mjs and follow rules/command-agent-orchestration.md. The profile is the source of truth for ownerAgentId, agentPlan, requiredAgentIds, dynamic specialist selection, default real-agents mode, and agent-required-blocked behavior.

Before durable work or completion claims, run node <resolved-supervibe-plugin-root>/scripts/command-agent-plan.mjs --command /supervibe-ui and follow the printed SUPERVIBE_COMMAND_AGENT_PLAN. The plan must show host dispatch support, proof source, AGENT_SELECTION_MODE, required agents, REQUIRED_AGENT_SOURCES, CALLABLE_AGENT_SOURCES, CALLABLE_AGENTS_READY, SCOPED_RECEIPT_GATE, MISSING_CALLABLE_AGENTS, and durable-write permission before any agent-owned artifact is produced. Role sources must distinguish definition availability from host-callable availability: REQUIRED_AGENT_SOURCES may include plugin-only, but CALLABLE_AGENT_SOURCES, CALLABLE_AGENTS_READY, and MISSING_CALLABLE_AGENTS decide whether the selected host can actually invoke the role. Plugin-only definitions are not enough for a real-agent completion claim.

Invoke the real host agents named by the plan and issue runtime receipts with hostInvocation.source and hostInvocation.invocationId. For active workflows, build the plan with --active --slug <slug> --handoff-id <handoff-id>; SCOPED_RECEIPT_GATE must be trusted for the current run before durable agent-owned outputs are allowed. Old global receipts are diagnostic only and do not unlock a new command/handoff. In Codex, invoke with spawn_agent using the CODEX_SPAWN_PAYLOAD_RULES and CODEX_SPAWN_PAYLOADS printed by command-agent-plan.mjs: forked payloads must set fork_context=true, must omit agent_type, model, and reasoning_effort, and must encode the Supervibe logical role in message instead of Codex agent_type. Record each returned Codex agent id with node <resolved-supervibe-plugin-root>/scripts/agent-invocation.mjs log ... before receipts are issued. inline is diagnostic/dry-run only. Do not emulate specialist agents, and do not let command or skill receipts substitute for agent, worker, or reviewer output.

Workflow Invocation Receipts

Any claim that this command invoked another Supervibe command, skill, agent, reviewer, worker, validator, or external tool must be backed by a runtime-issued workflow receipt created with node <resolved-supervibe-plugin-root>/scripts/workflow-receipt.mjs issue .... Hand-written receipts are untrusted. Agent, worker, and reviewer receipts must include hostInvocation.source and hostInvocation.invocationId from a real host dispatch; command or skill receipts must not substitute for specialist output. Scoped current-run receipts are required for active command/handoff claims; old global receipts are diagnostic only and cannot authorize a new agent-owned output. Durable artifacts produced by this command must stay linked through .supervibe/memory/workflow-invocation-ledger.jsonl and artifact-links.json; run npm run validate:workflow-receipts and npm run validate:agent-producer-receipts before claiming the command, delegated stage, or produced artifact is complete.