Your kanban board is now a one-click AI development pipeline.
Every developer using Claude Code has the same problem: it lives in the terminal, tasks live in the project board, and context lives in your head. You context-switch constantly — copying task titles into prompts, re-explaining checklists you already wrote, manually marking things done afterward. Claude Coder eliminates every one of those steps. A real embedded terminal sits right next to your board. Click Ask Claude Code on any task, and Navique generates the full structured prompt, navigates to your project directory, and submits it automatically. Claude starts working immediately — with your complete task context on the first message.
Claude Coder is not a chat window or a terminal simulation. It is a real PTY (pseudo-terminal) session powered by the same technology as macOS Terminal. Navique spawns your login shell (zsh, bash, or whatever you use) in login mode — your complete environment is intact: $PATH, nvm, rbenv, pyenv, $ANTHROPIC_API_KEY, every alias, every shell function. If claude works in your terminal, it works here. Zero configuration. The full Claude Code TUI runs inside the app: streaming responses, file diffs, interactive prompts, slash commands (/clear, /exit, /init), and tool confirmation dialogs.
Click Ask Claude Code on any kanban task. A composer sheet appears pre-filled with your task title, description, and every checklist item formatted as a structured work order. Review the prompt, add any session-specific context ("use the existing auth pattern in AuthService.swift", "don't touch the database schema"), and click Launch Claude. Navique opens the terminal, navigates to your project directory, and submits the full prompt automatically. Claude starts coding before you've put your hands back on the keyboard.
Checklist items in a task are not just notes — they become Claude's literal work order. Every item is included in the generated prompt in [ ] / [x] format so Claude knows exactly what still needs doing. Items already checked off appear struck-through in the composer preview. The specificity of a well-built checklist — "add auth middleware", "write unit tests", "update API docs", "handle error state" — produces dramatically better, faster first responses than a vague prompt.
Run up to four Claude Code sessions simultaneously, each in its own tab. All sessions stay live in the background — nothing pauses when you switch tabs because all terminal NSViews remain mounted simultaneously. Named tabs show the task title (task-linked sessions) or a custom name (manual sessions) so you always know what is running where. Spin up a blank interactive session any time for exploratory work or one-off questions not tied to a specific task.
Claude Code's --name and --resume flags let you maintain persistent conversation context across sessions. Navique manages this entirely automatically. The first session on a project starts with claude --name your-project. Every session after that automatically detects the existing named session and starts with claude --resume your-project — Claude remembers the codebase, the patterns established, and the decisions made in previous conversations. Each project gets its own isolated named session. Working on Project A's auth system never bleeds into Project B's refactor.
Before Claude Code can be truly useful on a project, it needs CLAUDE.md — the context file that tells Claude about your architecture, conventions, and commands. The first time you launch a session on a project, Navique shows a progress modal and runs claude --init-only --name your-project in the project directory. Claude scans the codebase, generates CLAUDE.md, and exits cleanly before the interactive session starts. This runs exactly once per project. Every session after the first skips initialization entirely and goes straight to the task.
When you type /exit inside Claude Code or close a task-linked session, Navique presents a structured wrap-up card instead of a dead terminal. The card shows the task title, lists every checklist item with toggles (pre-selecting all incomplete items since Claude probably just finished them), and offers two primary actions: Mark Done (moves the task to the Done column) or Move to Review (moves to Review for a human check). All changes write directly to the database — the kanban board updates instantly. Your project board reflects reality automatically, with no separate ticket-updating step.
You do not have to close a session to start the next task. Open the task picker while a session is running: choose the next task, wrap up the current one (Done, Review, or Skip), then confirm the auto-generated prompt for the new task. Navique injects the full prompt directly into the live terminal's input and submits it — exactly as if you typed it and pressed Enter. Claude immediately starts the next task with no re-launching, no re-navigating, and no re-establishing codebase context. Chain tasks back-to-back through a single long-running session with proper board tracking at every handoff.
A persistent right-hand panel shows all running sessions at a glance even when a terminal fills the main content area. Each session card shows the session name, the project directory path (~-shortened), and how long ago the session started. Click any card to bring that terminal to the foreground. A session count badge shows current and maximum slots (e.g. 2/4) so you always know how many are available. Per-session close buttons trigger wrap-up for task-linked sessions before closing.
Claude Coder is a first-class citizen in the Quick Launcher (⌘K). Type claude to get "Ask Claude Code — Open interactive session" as the top result. Type claude fix the auth bug and the result becomes "Ask Claude Code — fix the auth bug" — select it to open a session with that text pre-loaded as the initial prompt. Claude starts answering before you have navigated anywhere. Works from anywhere in the app.
The Prompt Composer shows a "Will resume session project-name" indicator at the bottom when an existing named session is detected, so you always know whether you are continuing prior context or starting fresh. A Start fresh link resets the project's session state — prompting for confirmation with a clear explanation — which also re-triggers CLAUDE.md initialization on the next launch. Useful when a project's structure has changed significantly and you want a clean slate.