Flow

Overview

Flow packages a battle-tested workflow for shipping software with Claude Code. It wires up twelve slash commands, a focused reviewer agent on top of Claude Code's built-in agents, and a handful of hooks so every session starts, runs, and ships consistently — while staying stack-agnostic by reading project specifics from a per-repo config.md.

Key features

1. 01

   Thirteen slash commands covering the full session lifecycle: init and uninstall, continue (cold-starts or resumes), status (a read-only where-am-I), pause (including a land mode that ships), plus autonomous autopilot

2. 02

   One focused reviewer agent, layered on Claude Code's built-in Explore and general-purpose agents plus the writing-plans skill, coordinated by a main orchestrator

3. 03

   Autopilot: an autonomous multi-sprint loop that spawns a non-overlapping agent team, deep-reviews, ships, and self-generates work via audit

4. 04

   Project-agnostic: dev/lint/test/build commands come from your repo's .flow/config.md

5. 05

   Deterministic shell helpers + a pause-time verification gate (config key `stop_check`: ask, always, or never) that runs your build/test commands at deliberate push points

6. 06

   Issue tracking that adapts to your backend (GitHub Issues, Linear, or local files)

7. 07

   Deep-review mode that fans out parallel reviewers and loops fixes until clean

8. 08

   Secret guards on reads, writes, and shell; auto-saved cross-session memory; doc drift-check at pause

Use cases

• — Solo developers who want a repeatable AI workflow across many repos
• — Small teams standardizing how they ship features with Claude Code
• — Codebases that want consistent reviews, issue hygiene, and CI checks before push
• — Anyone tired of re-explaining their stack to Claude on every session