Skip to content
BranchesTags

Repository files navigation

solopreneur

solopreneur is a family of Claude Code and Codex plugins that gives solo founders the workflows of a full engineering org: plan review, code review loops, automated PR cycles, marketing, design, and platform-specific experts. Install à la carte.

Quick Start

Claude Code:

claude plugin marketplace add hanamizuki/solopreneur
claude plugin install solopreneur@solopreneur

Codex:

codex plugin marketplace add hanamizuki/solopreneur --ref main
codex plugin add solopreneur@solopreneur

Claude Code can also install the role plugins (marketer, designer, ios-dev, android-dev, ai-engineer, neo4j-dev). Codex currently ships the verified core subset only. See the full install guide.

What's in the box

Plugin What you get
solopreneur 17 in-house skills (15 product workflows + merge and Codex-agent plumbing)
marketer marketer agent + 8 in-house skills (7 marketing workflows + agent router)
designer designer agent + 10 vendored design skills (taste-* family + impeccable)
ios-dev ios-dev agent + ios-patterns + ios-app-templates (reference apps) + 24 vendored skills (apple-design + asc-* + iphone-apps)
android-dev android-dev agent + android-patterns + 35 vendored skills (Compose + gplay-* + Android official)
ai-engineer ai-engineer agent + ai-app-templates + 1 vendored skill (senior-prompt-engineer)
neo4j-dev neo4j-dev agent + 4 vendored Neo4j skills

On Claude Code, installing any sub-plugin auto-pulls solopreneur. Requires Claude Code ≥ v2.1.110 for plugin dependency resolution.

Codex currently publishes the compatible core subset: autopilot, greenlight, handoff, merge-pr, perspective, plan-review, and post-mortem. Unsupported skills stay out of the Codex package instead of appearing and failing later.

Codex V1 is intentionally a smaller, fail-closed surface:

Invoke on Codex Supported V1 contract
$solopreneur:autopilot One natural PR, run now, from a clean and up-to-date main. No multi-PR waves or scheduling.
$solopreneur:greenlight external Open PRs sized S or M, with an authenticated Claude CLI as the independent final gate. No uncommitted, post-commit, or L-size runs.
$solopreneur:handoff Document printed inline, plus a best-effort /tmp file. A read-only sandbox blocks the file; the printed document is the deliverable.
$solopreneur:merge-pr Merge-only flow with exact-head checks; no pre-merge plan or worktree mutation.
$solopreneur:perspective Same ten perspectives. Name the one you want up front in codex exec — the menu needs a conversational surface.
$solopreneur:plan-review internal Technical and lean findings only; no outside review, adjudication, or write-back.
$solopreneur:post-mortem Same git archaeology and report. In codex exec, put the bug description in the prompt; the run stops at the printed report.

The skill descriptions below document the full Claude Code surface. Use the contracts above when running the seven published skills on Codex.

Migrating from a previous version? See MIGRATION.md. For per-release, per-plugin notes, see CHANGELOG.md.

Repository layout

  • skills/ is the single source for all 106 skills, grouped by plugin so each skill is directly browsable.
  • src/ holds manifests, agents, shared helpers, vendor metadata, and maintainer scripts.
  • plugins/claude/ and plugins/codex/ are generated install packages. Do not edit them directly.

Plugins

Each plugin section describes the bundled skills (in-house and vendored) plus its Requirements: external CLIs, plugins, MCPs, or GitHub apps the plugin's skills and agent integrate with. Hard requirements are called out explicitly; everything else is recommended and degrades gracefully if absent.

solopreneur (core)

The foundation. Every other plugin depends on this one. No agent, just 17 skills: 15 product workflows plus merge and Codex-agent plumbing.

Your Virtual Product Team

Skill What it does
/mvp PM. Drives the full new-product flow end-to-end: brainstorming → PRD visual confirmation → template lookup (auto-discovers *-app-templates in installed plugins) → plan → execution. Use when starting from scratch
/plan-review Tech Lead. Vets a spec, plan, or design doc in three stages — latest official docs + platform best practices, a leanness pass that cuts what the plan doesn't need, and an independent outside reviewer challenging it across 5 dimensions — then walks you through every finding. internal runs the first two stages only and reports findings without adjudication or write-back
/worktree-handoff Coworker. Creates an isolated git worktree with a CONTEXT.md so the next session picks up exactly where you left off
/handoff Scribe. Packages the current session into a self-contained markdown context doc, printed inline so you can copy and paste it into any other agent (Codex, ChatGPT, a fresh Claude session, an agent on another machine). Also saved under /tmp/handoff/. No worktree
/preview Presenter. Turns any proposal / plan / idea into interactive HTML and publishes a private Library (path-scoped config, catalog sidebar, provenance footer, in-page comments). Single-item Share requests deploy an isolated preview of one entry without changing the Library production URL. Legacy per-page deploy remains for compatibility
/specialist-review Code Reviewer. Detects your tech stack, dispatches matching expert agents, and reviews against best-practice skill indices
/post-mortem SRE. Traces a bug through git history, finds the root cause commit, produces a structured post-mortem report
/session-retro Coach. Reviews the current conversation for mistakes, traces root causes, proposes durable process improvements
/perspective Thinking Partner. Switch between thinker perspectives (Musk, Feynman, Munger, Naval, Jobs, Taleb, …) to analyze problems from a different angle

Backlog Management

This repo dogfoods the backlog workflow with public task files under todos/.

Skill What it does
/todos-review Backlog Reviewer. Deep-reviews a single todo/backlog item for feasibility and priority — is it worth building? Dispatches platform-specific expert agents and outputs a readiness rating. Specs and implementation plans go to /plan-review
/todos-cleanup Backlog Janitor. Batch-scans backlog, matches against git history, moves completed/partial items to done/ or doing/

Automation Pipelines

Start them and walk away. They loop until the job is done.

Skill What it does
/autopilot Auto Build. Splits a large feature into multiple PRs and orchestrates unattended implementation, review, and merge. Supports scheduling for off-hours execution
/greenlight Code Review Loop. Triggers external reviewers (Codex bot + CLI, the Gemini bot when active on the repo, CodeRabbit), fixes issues, re-triggers. Review depth scales with PR risk (S/M/L sizing). Loops until the PR is clean
/todos-babysit Backlog Monitor. Scans backlog and in-progress todos, cross-references PR status, reviews new items, and maintains worktrees. Interactive mode: presents a confirmation checkpoint before acting. Loop mode (/loop 24h /todos-babysit): auto-executes safe operations and auto-implements bug fixes that pass the readiness gate. Notifies only for items that need human judgment
/merge-pr Merge Gate. Verifies reviews, checks, branch state, and mergeability before merging the current pull request and cleaning up its worktree

Discovery and platform plumbing

Skill What it does
/rebuild-skill-index Generates per-platform extended indexes of every relevant skill installed on this machine. Feeds the ios-dev, android-dev, designer, marketer, and neo4j-dev agents' extended discovery. Run after installing/removing platform skills.
codex-agents-bootstrap Installs or refreshes managed solopreneur-family custom agents for Codex, while preserving hand-authored agents and reporting inactive or orphaned managed copies without deleting them.

Requirements

  • git, gh (GitHub CLI), jq: required CLIs. Used across /greenlight, /autopilot, /post-mortem, /todos-babysit, and scripts/sync-vendored.sh; jq is also required by codex-agents-bootstrap.
  • Codex CLI: required for codex-agents-bootstrap and /greenlight uncommitted mode (the only path on main with uncommitted changes). Also used by /plan-review (stage 3, the external reviewer), /greenlight PR mode (one reviewer option), and /naming (multi-model candidate generation). The custom-agent bootstrap and delegation pilot is validated against Codex CLI 0.147.0.
  • Python 3.9+: required by codex-agents-bootstrap to validate agent TOML before installation. Its fixed Tomli 2.4.1 parser is bundled, so no pip install or network access is required. Other workflows may also invoke python3, but do not establish this bootstrap-specific minimum version.
  • superpowers plugin: strongly recommended. /greenlight and /specialist-review use superpowers:requesting-code-review and receiving-code-review for the review framework. Graceful fallback if absent.
  • Ponytail plugin: optional. /greenlight uses ponytail:ponytail-review for over-engineering, dead-code, YAGNI, and simplification findings. The current runtime skips it when unavailable.
  • context7 MCP: strongly recommended. Used by /plan-review, /specialist-review, and every stack agent (ios-dev, android-dev, ai-engineer, neo4j-dev, designer) for current official docs. Graceful skip if absent.
  • gstack review: optional. /greenlight uses it when the current session can resolve the real gstack review capability; a similarly named bundled command is not treated as gstack without matching provenance.
  • Codex GitHub bot: optional. It is the preferred GitHub final gate for Claude-hosted /greenlight PR runs, with Codex CLI as the local fallback.
  • Cursor Bugbot: optional, and not currently usable as a final gate — the registry holds no verified login for it, and a reviewer with no login cannot close a round. It can still be triggered as an additional reviewer.
  • Claude CLI: optional on Claude Code and required on Codex for Greenlight's independent final gate. Cross-host calls use the Claude profile matching the active Codex config and never guess another profile.
  • Gemini Code Assist: optional. /greenlight PR-mode reviewer (/gemini review), offered only when activity detection finds it acting on the repo. Consumer Code Assist stopped GitHub code review on 2026-07-17; enterprise is unaffected. For post-commit review, /greenlight uses the Antigravity CLI (agy) as its Gemini-family reviewer.
  • CodeRabbit: optional. /greenlight can trigger it with @coderabbitai review and also collects its automatic PR review activity.
  • Vercel CLI: optional. /preview publishes a private Library (and optional per-item Share deployments) to a Vercel project when present; first-run setup provisions a single private target. Gracefully degrades to a local open of the HTML when absent.

marketer

Brand, GTM, content, and writing work. Ships the marketer agent plus 8 in-house skills: 7 domain workflows and one agent router.

Skill What it does
/gtm Strategist. Generates a complete Go-To-Market strategy. Analyzes the codebase, interviews you across multiple sessions, and produces 4 strategy docs (brand, market landscape, messaging, channel playbook). Supports weekly incremental updates
/naming Brand Namer. Generates product or company names through structured brief, multi-model candidate generation (Claude + optional Codex / Gemini), and two-layer evaluation. Supports greenfield and rebrand modes. Grounded in Lexicon / Interbrand / Siegel+Gale methodology plus processing fluency, sound symbolism, and iconicity research. Auto-reuses docs/gtm/ if present
/humanly Editor. Removes AI writing patterns from text with English and Traditional Chinese pattern catalogs, generated prewrite briefs, 3-tier word tables, and severity-based rewrite/review audits. A fidelity layer keeps prices, quotes, names and commitments verbatim and forbids inventing facts or the author's memories; a Taiwan localization layer catches mainland vocabulary and half-width punctuation
/x-writing Writing Coach. X/Twitter writing coach for single tweets, threads, and long-form posts. Generates hooks, suggests topics, reviews drafts, and explains craft principles grounded in Aesthetic Writing, RARE hooks, and the algorithmic reality of X
/x-growth X Growth Consultant. Diagnoses X/Twitter profiles, co-creates personalized 12-week growth plans. Covers algorithm mechanics, content strategy, engagement tactics, monetization, and Dream 100 outreach. Integrates with GTM docs
/linkedin-growth LinkedIn Growth Consultant. Diagnoses LinkedIn profiles, co-creates personalized 90-day growth plans. Covers algorithm mechanics, content pillars, engagement engine, audience strategy, and KPI tracking. Integrates with GTM docs
/slide-design Presentation Designer. Wraps frontend-slides or revealjs with a brand setup phase. Bakes brand colors, typography, and assets in from slide 1. Includes projection-optimized typography scale, Phosphor SVG icon sprite, layered backdrop system, keyboard-driven reveal patterns, fade-in/out background music, 13 reusable layout components, and AI-slop review via /humanly (English + Chinese)
using-marketer Routes explicit marketer requests and cross-concern marketing work to the marketer agent when available, with a bounded inline fallback.

Requirements

  • frontend-slides plugin: single-HTML, animation-rich slide engine. Used by /slide-design (recommended).
  • revealjs-skill plugin: reveal.js scaffolding (fragments, vertical stacks, speaker notes, Chart.js). Alternative engine for /slide-design. Wraps the underlying reveal.js library.
  • Gemini CLI: optional. Used by /naming for parallel multi-model candidate generation. Without it, /naming runs Claude + (optional) Codex CLI only.

designer

The designer agent for UI/UX work that spans web, iOS, and Android. Ships 10 vendored design skills.

Bundled skills

Requirements

  • frontend-design plugin: creative, polished frontend code generation that avoids generic AI aesthetics. Auto-classified into the design extended index by /rebuild-skill-index.
  • ui-ux-pro-max-skill plugin: UI/UX intelligence library: 50+ styles, 161 color palettes, font pairings, product types, UX guidelines, chart types across 10 stacks. Auto-classified.
  • Pencil MCP: optional. If mcp__pencil__* tools are available, the designer agent prefers them for .pen design file access (layout, variables, guidelines, export).

ios-dev

The ios-dev agent (SwiftUI / SwiftData / concurrency / testing / App Store) plus the in-house ios-patterns skill and 24 vendored skills covering Apple design principles, App Store Connect CLI workflows, and full iPhone app build/debug/ship.

Bundled skills

Requirements

  • asc CLI: required by every asc-* skill. Without it, those 22 skills won't run. Install from rorkai/App-Store-Connect-CLI.
  • Axiom plugin: 200+ skills covering SwiftUI, SwiftData, concurrency, testing, App Store, camera, AI, graphics. After install, run /rebuild-skill-index once and the iOS-relevant skills appear in the ios-dev agent's extended index. Install: claude plugin marketplace add CharlesWiltgen/Axiom then claude plugin install axiom@axiom-marketplace.

android-dev

The android-dev agent (Jetpack Compose / Kotlin / Play Console / build / performance) plus the in-house android-patterns skill and 37 vendored skills from 5 different upstream repos.

Bundled skills

Requirements

  • gplay CLI: required by every gplay-* skill. Without it, those 16 skills won't run. Install from tamtom/play-console-cli.

ai-engineer

The ai-engineer agent for LangGraph / LangChain / streaming / tool calling / structured output, plus 2 in-house skills and 1 vendored skill.

Bundled skills

  • ai-app-templates (in-house): reference implementations for common AI backend shapes (currently simple-llm-api: minimal FastAPI service with one POST /chat endpoint, provider chosen at scaffold time — Anthropic / Gemini / OpenRouter).
  • senior-prompt-engineer: vendored from alirezarezvani/claude-skills. Advanced prompt patterns + LLM eval frameworks + agent orchestration.

(No external requirements.)

neo4j-dev

The neo4j-dev agent for modern Cypher (QPP, CALL subqueries), graph data modelling, schema design, drivers, and query plan tuning. Ships 4 vendored skills.

Bundled skills

Requirements

  • neo4j-admin, cypher-shell, aura-cli: Neo4j first-party CLIs. Install per the neo4j-cli-tools skill's guidance, depending on which workflow (admin / query / cloud) applies.

How it works

solopreneur wraps the full development lifecycle — from idea to shipped code — with skills that plan, build, review, and iterate. Six specialist agents (ios-dev, android-dev, ai-engineer, designer, marketer, neo4j-dev) are dispatched automatically by pipeline skills or invoked manually when you need domain expertise.

Idea
 │
 ├─ /mvp ────────────── Brainstorm → PRD → template → plan → execute
 │                       dispatches: ios-dev · android-dev · ai-engineer
 │
 ├─ /plan-review ────── Vet the spec: official docs · leanness · outside reviewer
 │
 ├─ /worktree-handoff ─ Isolate the work in a git worktree
 ├─ /autopilot ──────── Split into PRs, auto-implement, review, merge
 │   │                   dispatches: ios-dev · android-dev · ai-engineer
 │   │                               designer · marketer
 │   │
 │   ├─ /specialist-review ── Expert review per tech stack
 │   │   dispatches: ios-dev · android-dev · neo4j-dev · designer · marketer
 │   │
 │   └─ /greenlight ────────── External review loop
 │       triggers: Codex bot · Gemini bot (when active) · CodeRabbit
 │
 ├─ /preview ──────────── Deploy interactive HTML for human review
 │
 ├─ /post-mortem ──────── Trace the root cause when something breaks
 ├─ /session-retro ────── Capture lessons from this session
 │
 └─ Backlog
     ├─ /todos-review ──── Deep-review before implementing
     ├─ /todos-cleanup ─── Batch-triage against git history
     └─ /todos-babysit ─── Periodic: review → implement → merge

Agents can also be called directly — ask the designer to critique a UI, the marketer to draft copy, or the ios-dev to review SwiftUI patterns.

Use Cases

Full auto-pilot development

Give /autopilot a spec. It splits the work into PRs, dispatches the right agents to implement each one, runs /greenlight review loops until clean, and merges. You come back to merged code.

Build an app from zero

/mvp walks you through brainstorming, PRD, template selection (discovers ios-app-templates, ai-app-templates, etc. from installed plugins), plan, and implementation. Dispatches ios-dev or android-dev for platform work.

Call the designer to polish a page

Ask the designer agent to critique or redesign a frontend page. It uses impeccable for interface polish and the taste-* family to override generic AI aesthetics with a specific design archetype. Works across web, iOS, and Android.

Review code until it's clean

/greenlight triggers external reviewers (Codex bot, the Gemini bot when active on the repo, CodeRabbit), fixes every finding, re-triggers. Repeats until no new suggestions remain. Also works in uncommitted mode on main for quick pre-commit cleanup, and in post-commit mode where it adds the Antigravity CLI (agy) as a Gemini-family reviewer alongside Codex.

Brand & GTM strategy

/gtm analyzes your codebase and interviews you to produce 4 strategy docs. Follow up with /naming for product names, /x-writing and /linkedin-growth for content strategy, and /slide-design for investor or launch decks.

Backlog on auto-pilot

Run /loop 24h /todos-babysit. It scans your backlog, auto-implements bug fixes that pass the readiness gate, and notifies you only for items that need human judgment.

Vet the plan before building

/plan-review runs three stages over a spec, plan, or design doc: it checks the approach against the latest official docs, hunts what the plan doesn't need, and sends it to an independent reviewer that challenges it across 5 dimensions. Then it walks you through the findings one by one. Add internal to run the first two stages only and get the findings reported back without adjudication or write-back.

Install

Claude Code

Add this repo as a marketplace source once, then install the pieces you need:

# Add the marketplace
claude plugin marketplace add hanamizuki/solopreneur

# Everyone: install the core plugin
claude plugin install solopreneur@solopreneur

# Then install whichever roles apply
claude plugin install marketer@solopreneur
claude plugin install designer@solopreneur
claude plugin install ios-dev@solopreneur
claude plugin install android-dev@solopreneur
claude plugin install ai-engineer@solopreneur
claude plugin install neo4j-dev@solopreneur

To update later, then start a new session:

# Refresh the marketplace catalog
claude plugin marketplace update solopreneur

# Update core
claude plugin update solopreneur@solopreneur

# Repeat for every installed role, for example:
claude plugin update marketer@solopreneur

Codex

Codex publishes the core plugin only:

codex plugin marketplace add hanamizuki/solopreneur --ref main
codex plugin add solopreneur@solopreneur

To refresh the marketplace and installed plugin later, then start a new session:

codex plugin marketplace upgrade solopreneur

Quick-start by role

If you build / do… Install
Run a one-person engineering org on auto-pilot solopreneur alone
GTM / brand / writing / slides solopreneur + marketer
Pure design / UI / UX work solopreneur + designer
iOS / macOS SwiftUI apps solopreneur + ios-dev
Android / Kotlin apps solopreneur + android-dev
LangGraph / AI agents / LLM apps solopreneur + ai-engineer
Neo4j / graph database work solopreneur + neo4j-dev

License

MIT

About

Skills and agents for solopreneurs — ship, review, debug, and think through problems with AI.

Resources

Readme
MIT license
Activity

Stars

137 stars

Watchers

0 watching

Forks

8 forks
Report repository

Releases

Packages

Contributors

Languages