What is insight-flow?

insight-flow is a workbench for AI-assisted task lifecycle management — a single CLI that owns your task state and serves a live dashboard. It gives any project a structured, auditable, and visual way to take work from an idea all the way to a merged change.

It was designed for Claude Code workflows, but it is not only for developers and it works standalone. If you want to design a flow of AI agents and track the progress of the tasks they work on, that is exactly what insight-flow does.

The core idea

Most "AI does the work" setups are a black box: you ask, something happens, and you hope it was right. insight-flow turns that into an explicit, inspectable process:

• Flows describe the lifecycle a task moves through, and which agent acts at each step.
• Agents are role-driven actors (each a slash command) that move a task forward — analyze, implement, review, fix, git, and more.
• Tasks are the unit of work. Each has a spec, a checklist, a review history, and an incident log — all stored as auditable JSON.
• The dashboard visualizes everything live: a Kanban board, a timeline, and a per-task detail panel.

        ┌── design a flow ──┐      ┌──── run agents along it ────┐
idea →  │  states + agents  │  →   │ analyze → implement → review │ → merged
        └───────────────────┘      └──────────────────────────────┘
                         everything tracked as JSON + shown live

Who runs what

insight-flow is agent-driven. In day-to-day use the agents create tasks and run the lifecycle with you — you don't manually type create/next/implement-start/etc. The CLI exists so the agents (and you, when you want) can drive task state; the dashboard is for watching it happen. So "you" in this documentation often means "you, via an agent."

Who it's for

• Builders using AI agents who want a repeatable pipeline instead of ad-hoc prompting.
• Teams who need an audit trail: what was specified, what was reviewed, what changed after approval, and what broke in production.
• Anyone orchestrating multi-step work — the flow and agent model isn't tied to writing code; it's a general "states + actors + tracking" engine.

What you get

Piece	What it does
CLI (insight-flow)	One binary: owns task state, runs the lifecycle, serves the dashboard. See CLI.
Default flow	A ready-to-use lifecycle (ready → … → merged) with agents bound to each transition. See Default Flow.
Agents	10 composable, role-driven slash commands. See Agents.
Dashboard	Live Kanban / timeline / detail view, plus a multi-project master overview.
Module model	"Everything is a module" — agents are composed from modules and customizable in user-space without forking.

Where to next

• Getting Started — install, create your first task, launch the dashboard.
• Default Flow — the lifecycle and its transitions.
• Agents — the actors and how they're composed.
• CLI — every command, grouped by what it does.