A Kanban Board for Claude Code: See Every Agent Session as a Card

A Claude Code session is not just a process, it is a work item with state. At any moment it has a task it was given, a branch it is writing to, a status (planning, coding, waiting on a permission prompt, finished, failed), and a growing diff that will eventually need your review. One session holds all of that in your head fine. Two, barely. At three or four, the state outgrows memory, and the tool you are using to view it, a terminal, is actively working against you.

Terminals hide state. A terminal shows you one session's most recent output, and only for the tab you happen to have focused. Everything that matters for managing parallel work, which session is stalled, which finished twenty minutes ago, which diff is getting dangerously large, is invisible until you go tab-by-tab and reconstruct it. A kanban board is the opposite trade: it shows less of any one session's output and all of every session's state, at a glance, in columns that literally are the status. This is why boards won for human teams tracking work items, and the argument transfers directly, agent sessions are work items that change state faster.

To be clear about scale: two occasional sessions do not need this, two tabs are fine, and we say so in the FAQ below. The board is the answer for when parallel sessions are routine. The broader escalation path from tabs to tmux to worktrees to a board is mapped in managing multiple Claude Code sessions, and the general case for agent-native boards, beyond Claude Code specifically, is in AI kanban for developers.

A board is only as useful as its cards. A card that just shows a title is a to-do list wearing columns. For agent sessions, the card has to mirror the session's real state, which means tracking six things:

Notice that half of these, spec, branch, review state, are not things the CLI shows you even in a focused tab. They live in your head or in git. Moving them onto a card is the actual value of the board; the columns are just the index.

AIDEN did not invent this category, and pretending otherwise would be silly. Vibe Kanban defined it: an open-source board whose cards were coding-agent tasks, running Claude Code and other CLIs, and it proved the model well enough that "kanban for coding agents" became a phrase people search for. Vibe Kanban shut down in April 2026. A community fork keeps the code alive for people who want to self-maintain, and our Vibe Kanban alternative page walks through the migration options in detail.

Nimbalyst fills the free-and-open-source niche today: MIT-licensed, cross-platform, a kanban view over agent sessions. If your requirement is "open source or nothing," it is the honest recommendation, and we compare it fairly on the desktop apps for coding agents roundup. AIDEN's bet is different: not just a board over sessions, but a process around them, a spec gate before any agent codes, automated worktrees, and mixed Claude Code and Codex fleets, which is the next section.

AIDEN's board runs Stories → Spec Review → In Progress → Review → Done. The shape is deliberately boring, it mirrors how a story actually flows through an agent, and the load-bearing part is the gate between the first two columns and the rest: no agent starts coding until you have approved a written spec on the card.

The board is the visible part; the orchestration underneath, how stories are queued, how branches become PRs, is covered in Claude Code orchestration.

1. 1

   Install AIDEN next to your CLI

   AIDEN is a macOS desktop app (macOS 12+, Apple Silicon and Intel) that runs on top of the Claude Code you already have, your CLI, your login, your usage limits. Codex CLI is optional but supported the same way.
2. 2

   Point it at a repo and add stories

   Create a project from a local git repository, then write one card per unit of work. Specific stories, "Add rate limiting to /api/checkout", produce specs an agent can actually be held to.
3. 3

   Approve specs, let cards flow

   Each story gets a drafted spec in Spec Review. Edit, approve, and the card moves to In Progress on its own worktree and branch. Run one card or several, the board is useful either way.
4. 4

   Review diffs from the Review column

   Finished cards land in Review with the full diff and the spec side by side. Read, send back with a comment, or open a PR in one click. Merged cards move to Done and their worktrees are cleaned up.