flow-next

---

Active development. Changelog | Report issues

🌐 Prefer a visual overview? See the Flow-Next app page for diagrams and examples.

New: Codex Review Backend. Cross-model reviews now work on Linux/Windows via OpenAI Codex CLI. Same Carmack-level criteria as RepoPrompt. See Cross-Model Reviews for setup.

---

Table of Contents

• What Is This?
• Why It Works
• Quick Start — Install, setup, use
• When to Use What — Interview vs Plan vs Work
• Agent Readiness Assessment — /flow-next:prime
• Troubleshooting
• Ralph (Autonomous Mode) — Run overnight
• Features — Re-anchoring, multi-user, reviews, dependencies
• Commands — All slash commands + flags
  • Command Reference — Detailed input docs for each command
• The Workflow — Planning and work phases
• .flow/ Directory — File structure
• flowctl CLI — Direct CLI usage

---

What Is This?

Flow-Next is a Claude Code plugin for plan-first orchestration. Bundled task tracking, dependency graphs, re-anchoring, and cross-model reviews.

Everything lives in your repo. No external services. No global config. Uninstall: delete .flow/ (and scripts/ralph/ if enabled).

Planning: dependency-ordered tasks	Execution: fixes, evidence, review

---

Epic-first task model

Flow-Next does not support standalone tasks.

Every unit of work belongs to an epic fn-N (even if it's a single task).

Tasks are always fn-N.M and inherit context from the epic spec.

Flow-Next always creates an epic container (even for one-offs) so every task has a durable home for context, re-anchoring, and automation. You never have to think about it.

Rationale: keeps the system simple, improves re-anchoring, makes automation (Ralph) reliable.

"One-off request" -> epic with one task.

---

Why It Works

You Control the Granularity

Work task-by-task with full review cycles for maximum control. Or throw the whole epic at it and let Flow-Next handle everything. Same guarantees either way.

# One task at a time (review after each)
/flow-next:work fn-1.1

# Entire epic (review after all tasks complete)
/flow-next:work fn-1

Both get: re-anchoring before each task, evidence recording, cross-model review (if rp-cli available).

Review timing: The RepoPrompt review runs once at the end of the work package—after a single task if you specified fn-N.M, or after all tasks if you specified fn-N. For tighter review loops on large epics, work task-by-task.

No Context Length Worries

• Tasks sized at planning: Every task is scoped to fit one work iteration
• Re-anchor every task: Fresh context from .flow/ specs before each task
• Survives compaction: Re-anchors after conversation summarization too
• Fresh context in Ralph: Each iteration starts with a clean context window

Never worry about 200K token limits again.

Reviewer as Safety Net

If drift happens despite re-anchoring, a different model catches it before it compounds:

1. Claude implements task
2. GPT reviews via RepoPrompt (sees full files, not diffs)
3. Reviews block until SHIP verdict
4. Fix → re-review cycles continue until approved

Two models catch what one misses.

---

Zero Friction

• Works in 30 seconds. Install the plugin, run a command. No setup.
• Non-invasive. No CLAUDE.md edits. No daemons. (Ralph uses plugin hooks for enforcement.)
• Clean uninstall. Delete .flow/ (and scripts/ralph/ if enabled).
• Multi-user safe. Teams work parallel branches without coordination servers.

---

Quick Start

1. Install

# Add marketplace
/plugin marketplace add https://github.com/gmickel/gmickel-claude-marketplace

# Install flow-next
/plugin install flow-next