Guides

Problem-oriented, task-focused recipes for getting specific things done with insight-flow. These assume you've already been through Getting Started and want to solve a concrete problem.

Recipes

Extend the agents for your stack

• Wire up PR creation for your git host — configure agents.extend.task-git so /task-git opens PRs/MRs via gh / glab (or a prefill URL).
• Add your stack's quality gates — make /task-implement and /task-review-fix run your typecheck / lint / test before marking work done, via agents.extend.

Customize the composition model

• Author a custom module — write a custom:<slug> module (section / include / mcp-server / hook / skill / bundle) in user space.
• Compose a custom agent — build an agent from module ids and generate its role file with prompt-build --compose.
• Define a custom flow — declare a lifecycle (agents, edges, statuses, entry agents) and bind task types to it.
• Install & uninstall modules — run a target's install plan, fill ${VAR} inputs, and roll changes back via the manifest.

Operate insight-flow

• Run the multi-project master — track several insight-flow projects from one overview server on port 6100.
• Upgrade from 1.x to 2.0 — migrate the layout and the composition model.

Troubleshooting & operations

• Troubleshooting & FAQ — problem → fix entries: ports, stale dashboards, the old workTasks/ layout, skipped agent steps, suppressing the browser, and recovering a stuck master.
• Using insight-flow with Cursor — scaffold the Cursor skills, hooks, and AGENTS.md, read the cursor provider badge, and understand the cloud-agent / approval-gate caveats.
• Set up observability (Langfuse) — opt into Langfuse tracing, resolve credentials config-first then env, and see what gets traced.

In the meantime

• Configuration documents every taskflow.config.json key, including agents.extend and flows.
• The CLI reference covers each command's flags.
• Agents describes each slash command's role.