Sign in

15-minute setup

Your agent structures GitHub, the graph runs the fleet

You describe a feature in plain language. Your agent (Claude Code, Grok, or Codex) proposes an epic, breaks down sub-tasks, sets blocked-by links and parent/child relations. Roxabi Live reads all of it and computesready, blocked, and done.

1 Pick your agent and install the kit

The roxabi-issues plugin bundles the issue-triage skill. Once installed, the agent activates it on its own when you talk about features, epics, or issue dependencies.

From a terminal, in any repo:

claude plugin marketplace add Roxabi/roxabi-live
claude plugin install roxabi-issues

The roxabi-live-marketplace catalog references the plugin and skill. You can invoke /issue-triage explicitly, but in practice the agent triggers it when context calls for it.

Grok is Claude Code compatible with zero extra config: it automatically reads Claude marketplaces, plugins, and skills (xAI docs).

Two equivalent paths:

  1. Install via Claude Code (commands above), then launch Grok in the same repo.
  2. Or open the Grok TUI Marketplace tab (/plugins) and install roxabi-issues from Roxabi/roxabi-live.

The skill appears as a slash command. Natural triggers: "create an epic", "block this issue", "break this into sub-tasks".

In the Codex CLI or Codex app:

codex plugin marketplace add Roxabi/roxabi-live
codex plugin install roxabi-issues

You can also browse Plugins in the Codex app or run /plugins in the CLI, then install from the marketplace (OpenAI docs).

Codex loads the skill implicitly when your request matches its description, or explicitly via @issue-triage.

2 You describe the feature, the agent structures GitHub

No commands to memorize. Describe the intent; the agent proposes a plan, you approve, it runs the skill in the background.

You
I want a feature with a user registration form.
Agent (Claude · Grok · Codex)
I suggest structuring this as an epic with 4 sequential sub-tasks. Frontend and API work only start once prerequisites are done.
  • #421 · DB schema + migration · size S · ready on creation
  • #422 · POST /signup API · blocked-by #421
  • #423 · Registration form + validation · blocked-by #422
  • #424 · E2E signup tests · blocked-by #423
Should I create the epic and these issues on GitHub?
You
Yes, go ahead.
Agent
Done. Epic #420 created, 4 children attached,blocked-by dependencies set. Only #421 is ready for now.
What the agent runs in the background (you don't have to)
  • Activates the issue-triage skill installed from the marketplace
  • Creates the epic with epic type, size, and priority
  • Creates each sub-issue as a child of the epic (native GitHub parent/child)
  • Sets blocked-by links to sequence the work
  • Applies size, priority, and lane labels for the cockpit
Remember: you steer intent ("I want this feature"). The agent handles GitHub mechanics (epic, children, blocked-by, labels). The Roxabi Live graph updates and shows what else you can launch in parallel.

3 What the graph understands

Two native GitHub relation types are enough. No external board, no Projects V2.

blocked-by: who waits for what

While a blocker is open, the issue stays blocked. When the blocker closes, it becomes ready. That's how you avoid launching agents on incompatible tasks.

Parent / child: epics

The epic groups sub-tasks. The cockpit filters by parent and shows hierarchy in the graph, like the registration example above.

Labels: size, priority, lane

The agent sets them via the skill. You use them in cockpit filters (size:M, P1-high, graph:lane/b, etc.).

4 Connect the Roxabi Live cockpit

  1. Sign in at app.live.roxabi.dev
  2. Install the GitHub App on your organization
  3. Webhooks + reconciliation keep the graph in sync in real time

Open the graph view: issues that are ready with no open blockers are the ones you can hand to an agent right away.

Resources

Open cockpitVersion française