gmickel claude marketplace

Plugins that make AI agents actually work.

🤖 New: Ralph mode — ship features while you sleep. Multi-model review gates that actually block on quality.

🧠 New in v0.3.17: Memory system — agents learn from NEEDS_WORK feedback. Stop repeating the same mistakes.

---

The Problem

Process failures, not model failures:

• Starting to code before understanding the codebase
• Reinventing patterns already there
• Forgetting the plan mid-implementation
• Skipping edge cases obvious in hindsight

This marketplace ships plugins that fix these problems.

---

Plugins

Plugin	What It Does
flow-next	Plan-first workflow with .flow/ task tracking. Zero deps. Multi-user safe. Recommended.
flow	Full-featured plan+work with optional Beads integration

Pick a plugin

• flow-next: .flow/ + bundled flowctl + Ralph mode (autonomous overnight loop) via /flow-next:ralph-init
• flow: original behavior (plan files, optional external tracker)

---

Flow-Next

Experimental. Give it a spin. Report issues.

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

Plan first, work second. Zero external dependencies.

# 1. Install
/plugin marketplace add https://github.com/gmickel/gmickel-claude-marketplace
/plugin install flow-next

# 2. Setup (recommended - adds CLI access + project docs)
/flow-next:setup

# 3. Use
/flow-next:plan Add a contact form with validation
/flow-next:work fn-1

Setup is technically optional but recommended - it adds CLI access via flowctl and project docs that help other AI tools understand your flow.

Agents that finish what they start.

flowchart TD
  A[Idea or short spec<br/>prompt or doc] --> B{Need deeper spec?}
  B -- yes --> C[Optional: /flow-next:interview fn-N or spec.md<br/>40+ deep questions to refine spec]
  C --> D[Refined spec]
  B -- no --> D
  D --> E[/flow-next:plan idea or fn-N/]
  E --> F[Parallel subagents: repo patterns + online docs + best practices]
  F --> G[flow-gap-analyst: edge cases + missing reqs]
  G --> H[Writes .flow/ epic + tasks + deps]
  H --> I{Plan review? rp-cli only}
  I -- yes --> J[/flow-next:plan-review fn-N/]
  J --> K{Plan passes review?}
  K -- no --> L[Re-anchor + fix plan]
  L --> J
  K -- yes --> M[/flow-next:work fn-N/]
  I -- no --> M
  M --> N[Re-anchor before EVERY task]
  N --> O[Implement]
  O --> P[Test + verify acceptance]
  P --> Q[flowctl done: write done summary + evidence]
  Q --> R{Impl review? rp-cli only}
  R -- yes --> S[/flow-next:impl-review/]
  S --> T{Next ready task?}
  R -- no --> T
  T -- yes --> N
  T -- no --> U[flowctl epic close fn-N]
  classDef optional stroke-dasharray: 6 4,stroke:#999;
  class C,J,S optional;

Loading

📖 Full guide (CLI, workflow, .flow layout): plugins/flow-next/README.md

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

Why It Works

You control the granularity:

• /flow-next:work fn-1.1 — one task at a time with full review cycles
• /flow-next:work fn-1 — throw the whole epic at it, walk away

Either way you get the same guarantees: re-anchoring, evidence, cross-model review.

No context length worries:

• Planning ensures every task fits one work iteration
• Re-anchoring after each task (and after compaction) prevents drift
• Fresh context window every iteration in Ralph mode

Reviewer as safety net:

• If drift happens despite re-anchoring, a different model catches it
• Reviews block until SHIP verdict — no "LGTM with nits" that get ignored

Bundles everything in a single Python script. No npm. No daemons. No config edits. Try it in 30 seconds. Uninstall by deleting .flow/ (and scripts/ralph/ if enabled).

Ralph (Autonomous Mode)

⚠️ Warning: Ralph defaults to YOLO=1 (skips permission prompts). Start with ralph_once.sh to observe a single iteration. Consider running in a Docker sandbox for isolation.

Setup (one-time, inside Claude):

/flow-next:ralph-init

Or from terminal without entering Claude:

claude -p "/flow-next:ralph-init"

Run (outside Claude):

scripts/ralph/ralph.sh

How Ralph differs from other autonomous agents:

Most agents gate by tests alone. Ralph adds production-grade quality gates:

• Multi-model reviews: Uses RepoPrompt to send code to a different model. Two models catch what one misses.
• Review loops until SHIP: Reviews block progress until <verdict>SHIP</verdict>. Fix → re-review cycles continue until approved.
• Receipt-based gating: Every review must produce a receipt JSON proving it ran. No receipt = no progress. At-least-once delivery with idempotent retry—treats agent as untrusted actor; receipts are proof-of-work.

📸 Ralph in action (click to expand)

📖 Ralph deep dive

Features

Re-anchoring	Before EVERY task, re-reads epic/task specs + git state. No drift.
Multi-user safe	Scan-based IDs. Soft claims via assignee. Actor auto-detect.
Zero deps	Bundled flowctl.py. No external CLI. Just Python 3.
Non-invasive	No daemons or CLAUDE.md edits. Delete .flow/ (and scripts/ralph/ if enabled) to uninstall.
CI-ready	flowctl validate --all exits 1 on errors. Drop into pre-commit or GitHub Actions.
One file per task	Merge-friendly. Conflict surface is minimal.
Automated reviews	Require RepoPrompt (rp-cli). Without it, reviews are skipped.
Dependency graphs	Tasks declare blockers. Nothing starts until dependencies resolve.

Commands

Command	What It Does
/flow-next:plan	Research, create epic + tasks in .flow/
/flow-next:work	Execute epic end-to-end, task by task
/flow-next:interview	Deep interview to flesh out a spec
/flow-next:plan-review	Carmack-level plan review via rp-cli
/flow-next:impl-review	Carmack-level impl review (current branch)
/flow-next:ralph-init	Scaffold autonomous loop in scripts/ralph/
/flow-next:setup	Install flowctl locally + add project docs
/flow-next:uninstall	Remove flow-next from project

Autonomous Flags

All commands accept flags to bypass interactive questions:

# Interactive (asks questions)
/flow-next:plan Add caching
/flow-next:work fn-1

# Autonomous (flags)
/flow-next:plan Add caching --research=grep --no-review
/flow-next:work fn-1 --branch=current --no-review

# Autonomous (natural language)
/flow-next:plan Add caching, use context-scout, skip review
/flow-next:work fn-1 current branch, no review

📖 Full documentation

---

Install

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

# Install plugin
/plugin install flow-next    # Recommended: zero deps, simpler
# or: /plugin install flow   # If you use Beads or want plan files

# Setup (recommended for flow-next)
/flow-next:setup             # Adds CLI access + project docs

---

Flow

Requires RepoPrompt v1.5.62+ for review features. Using older RepoPrompt? Downgrade: /plugin install flow@0.8.0

Plan first, work second. The original, with optional Beads integration.

/plugin install flow

/flow:plan Add OAuth login for users
/flow:work plans/add-oauth-login.md

How It Works

Failure Mode	How Flow Fixes It
Weak research	Parallel agents gather context before coding starts
Ignoring existing code	Explicit pattern reuse from your repo
Drifting from plan	Plan re-read between every task
Shallow self-review	Cross-model review via RepoPrompt

Commands

Command	What It Does
/flow:plan	Research + produce plans/<slug>.md
/flow:work	Execute plan end-to-end with task tracking
/flow:interview	Deep interview to flesh out spec/bead
/flow:plan-review	Carmack-level plan review via rp-cli
/flow:impl-review	Carmack-level impl review (current branch)

Integrations

• RepoPrompt for token-efficient codebase exploration + cross-model reviews
• Beads for dependency-aware issue tracking (auto-detected)

📖 Full documentation · Changelog

---

Also Check Out

GNO — Local hybrid search for your notes, docs, and code. Give Claude Code long-term memory over your files via MCP.

bun install -g @gmickel/gno && gno mcp install --target claude-code

---

Contributing

1. Create plugins/<name>/ with .claude-plugin/plugin.json
2. Add commands/agents/skills under that plugin root
3. Update .claude-plugin/marketplace.json
4. Validate: jq . .claude-plugin/marketplace.json

---

Made by Gordon Mickel · @gmickel · gordon@mickel.tech