Skip to main content
Version: 2.0

Getting Started

This page takes you from zero to a running dashboard with your first task.

insight-flow is agent-driven

In normal use you don't run the task commands by hand. The agents — Claude Code slash commands like /task-analyze, /taskmaster, /task-implement — create tasks and drive the whole lifecycle with you. The CLI commands shown below are what the agents call under the hood; you're welcome to run them directly, but you rarely need to. Think of insight-flow as the engine the agents drive, plus a dashboard for you to watch.

1. Initialize in your project​

npx insight-flow init

init scaffolds the agent role files, slash commands, and a taskflow.config.json with zero technology assumptions — no package manager, language toolchain, or git host is baked in. You add your own stack-specific commands via the agents.extend mechanism.

Optionally install the CLI globally so insight-flow is on your PATH everywhere:

npm install -g insight-flow

2. Create your first task​

There are two ways to create a task — the agent way (normal) and the manual CLI way.

In Claude Code, just describe what you want and let the agents do it. Typically you start with the strategist, which challenges the brief and then hands off to the taskmaster to write the spec:

/task-analyze   # discuss the idea; it proposes options
/taskmaster     # writes the task spec (TASK.md + CHECKLIST.md)

You don't type any create command — /taskmaster calls the CLI for you and the new task appears on the dashboard. This is how you'll create tasks day to day.

The manual way (optional)​

If you want to create a task directly, the CLI is always there:

insight-flow create \
  --title "Add user auth" \
  --type feat \
  --priority high \
  --tags auth,api

Either way, each task gets a unique Nxx ID and a folder under insightFlow/workTasks/ containing a TASK.md spec and a CHECKLIST.md, and starts in the ready state of the default flow.

3. See what to work on​

insight-flow current    # the active task
insight-flow next       # the next actionable task, by priority
insight-flow list       # everything
insight-flow stats      # aggregate statistics

These are mostly how the agents decide what to do next — /task-implement runs next, /task-review runs next-review, and so on. Run them yourself any time you want to peek at state, but you don't have to; the dashboard shows the same thing visually.

4. Launch the dashboard​

insight-flow            # dashboard at http://localhost:6006
# or
insight-flow ui         # dashboard + multi-project master overview (port 6100)

The dashboard is a live Kanban board, timeline, and per-task detail panel that updates as the CLI (or the agents) change task state.

5. Run the lifecycle with agents​

In Claude Code, the work is driven by agents — slash commands that move a task through the flow:

/task-analyze   → challenge the brief, propose options
/taskmaster     → write the task spec
/task-implement → implement it
/task-review    → AI review
/task-git       → branch, push, PR, merge

Prefer to drive it yourself? Every transition also has a plain CLI command — see the CLI reference.

Next steps​

  • Understand the Default Flow your tasks move through.
  • Meet the Agents and the module model behind them.
  • Browse the full CLI, grouped by purpose.