Plugins, skills, and coding resources by Noah Rasheta -- hosted at shipfast.cc
I'm a podcast host with a bunch of side businesses and projects. I build Claude Code plugins, skills, and agents to automate the things I do repeatedly. This repo is where all of that lives -- open source, for anyone to use, fork, or build on.
The stuff here spans many domains because my work does too. Data infrastructure, automations, prompt engineering, content workflows, and whatever I'm tinkering with next. The common thread is Claude Code and agentic AI.
Add the marketplace and install any plugin:
/plugin marketplace add noahrasheta/shipfast
/plugin install <plugin-name>@shipfast
| Plugin | What it does | Command |
|---|---|---|
| create-prd | AI-optimized PRD generation through guided brainstorming | /create-prd |
| create-image | AI image generation with multi-agent pipeline | /create-image |
| human-writing | Write text that reads as human, not AI | /human-writing |
| dc-due-diligence | Data center site analysis across 9 domains | /due-diligence <folder> |
Turn a product idea into a comprehensive, AI-agent-optimized Product Requirements Document through guided conversation. Invoke /create-prd and the skill acts as a brainstorming partner -- asking one question at a time, adapting to your technical level, and producing a PRD that any AI coding tool can build from.
What it does:
- Detects your technical level and adapts questioning accordingly (plain language for non-technical founders, direct stack questions for developers)
- Runs a 7-phase conversation: Vision & Context, Technical Assessment, Features & Scope, User Experience, Business Context (optional), Generate PRD, Refine
- Produces a structured PRD with: executive summary, problem statement, user stories with testable acceptance criteria, MVP feature scope, user flows, tech stack, data models, UI/UX requirements, non-functional requirements, explicit out-of-scope list, risks, implementation phases, and environment setup
- Validates the PRD against a 6-point quality checklist before delivering
- Includes bundled reference material on AI-optimized PRD best practices and a complete sample PRD
Works with: Claude Code, Cursor, Windsurf, Lovable, GSD (/gsd:new-project --auto @PRD.md), Director, Conductor, or any LLM-based builder.
Usage:
/create-prd
The skill will start a guided conversation to understand your product idea and generate the PRD.
Status: v1.0.0 -- initial release. Conversational PRD generation with adaptive technical assessment.
AI image generation using the PaperBanana agentic framework with Nano Banana Pro (Gemini 3 Pro). Invoke /create-image and a team of specialized agents handles the rest -- from analyzing reference images to generating 5 variants and ranking them.
What it does:
- Asks structured clarifying questions about what you want (subject, style, mood, aspect ratio, resolution, reference images, etc.)
- Spawns 4 agents in a sequential pipeline: Research Agent, Prompt Architect, Generator Agent, Critic Agent
- Generates 5 image variants (Faithful, Enhanced, Alt Composition, Style Variation, Bold/Creative)
- Critiques each image on 4 dimensions from the PaperBanana paper (Faithfulness, Readability, Conciseness, Aesthetics)
- Presents ranked results with a top recommendation and offers iteration
Based on: PaperBanana (Google Cloud AI Research & Peking University) -- a multi-agent framework that outperforms single-shot image generation by +17% overall.
Prerequisites:
pip install google-genai Pillow python-dotenvAdd your Gemini API key to a .env file in the project where you'll run the plugin:
GEMINI_API_KEY=your-key-here
Get an API key from Google AI Studio.
Usage:
/create-image
Or with an initial description:
/create-image an infographic about global coffee consumption in the style of Visual Capitalist
Generated images are saved to a shipfast-images/ folder in your project directory for easy access.
Status: v0.1.0 -- initial release. Core pipeline functional.
Write text that reads as human-written, not AI-generated. Based on Wikipedia's "Signs of AI writing" field guide and a comprehensive anti-AI writing guide, this skill knows exactly which vocabulary, structures, and formatting patterns mark text as machine-generated -- and avoids all of them.
What it does:
- Rewrites existing text to remove AI tells (banned vocabulary, structural formulas, formatting patterns)
- Writes new text from scratch that passes as human-authored
- Reviews text and flags specific AI detection signals with explanations
- Adapts tone by context: casual for emails, direct for technical docs, personality-forward for blog posts
What it catches:
- 60+ banned words and phrases (nestled, vibrant, delve, testament, pivotal, showcase, etc.)
- Structural formulas (challenges-and-future, significance inflation, negative parallelisms, ritual conclusions)
- Formatting tells (title-case headings, inline-header bullet lists, em-dash overuse, excessive boldface)
- Copulative avoidance ("serves as" instead of "is"), elegant variation (synonym cycling), vague attributions
Usage:
/human-writing
Or just add "make it sound human" or "write this as a human would" to any writing request.
Status: v1.0.0 -- initial release. Full coverage of documented AI detection patterns.
Automated due diligence for data center site opportunities. Point it at a folder of broker documents (PDFs, spreadsheets, Word docs, PowerPoint decks, images) and it runs a full analysis across 9 domains, synthesizes cross-domain risks, and produces a scored executive summary with a Pursue / Proceed with Caution / Pass verdict.
What it does:
- Converts all documents in the folder to markdown (PDF, Excel, Word, PowerPoint, images via vision API)
- Spawns 9 domain research agents in parallel, each analyzing the converted documents and conducting independent web research to verify broker claims:
- Power -- utility interconnection, grid capacity, substation, redundancy design
- Connectivity -- fiber carriers, route diversity, carrier neutrality
- Water & Cooling -- water supply, cooling system design, scarcity risk
- Land, Zoning & Entitlements -- zoning compliance, permits, building readiness
- Ownership & Control -- property ownership verification, middleman detection, litigation
- Environmental -- natural hazards, contamination, Phase I ESA status
- Commercials -- deal terms, land cost, power rates, lease structure
- Natural Gas -- pipeline access, on-site generation feasibility, permitting
- Market Comparables -- comparable transactions, market rates, competition
- Runs a Risk Assessment agent that reads all 9 domain reports and identifies cross-cutting risks, deal-breakers, and compound risks
- Generates an Executive Summary that scores each category (High / Medium / Low), applies a tiered verdict system, and delivers a stakeholder-ready report
Output:
<folder>/EXECUTIVE_SUMMARY.md-- scored summary with verdict, strengths, concerns, deal-breakers, and next steps<folder>/research/*.md-- 10 detailed research reports (one per domain + risk assessment)
Step 1: Add the marketplace and install the plugin
/plugin marketplace add noahrasheta/shipfast
/plugin install dc-due-diligence@shipfast
Python 3.11+ is required on your machine. The plugin automatically sets up its Python virtual environment the first time you run /due-diligence -- no manual setup step needed.
Step 2 (only if your folder has images or scanned PDFs): Set ANTHROPIC_API_KEY
The document converter uses the Anthropic API directly to extract text from images and scanned PDFs via Claude's vision capability. This runs as a Python subprocess, separate from Claude Code itself, so it needs its own API key.
Add to your shell profile (~/.zshrc or ~/.bashrc):
export ANTHROPIC_API_KEY=your-key-hereThen restart Claude Code so it picks up the new variable. Environment variables are inherited at launch time -- adding the key while Claude Code is running won't take effect until you restart it.
If your folder only contains native PDFs, Word docs, Excel files, and PowerPoint decks, you do not need this -- those formats are converted using local Python libraries with no API calls.
The 12 research agents verify broker claims using web research. There are two layers, and the baseline works with zero configuration:
Baseline (always available): Agents use Claude Code's built-in WebSearch and WebFetch tools. These work automatically -- no API keys or setup needed.
Enhanced (optional): If you have Tavily, Exa, or Firecrawl configured as MCP servers in Claude Code, agents will automatically detect and use them via ToolSearch for deeper search capabilities. This is a nice-to-have, not a requirement.
/due-diligence ./path/to/opportunity-folder
The folder should contain the broker-provided documents for the opportunity. The plugin handles conversion, analysis, and reporting end to end.
Status: v0.3.1 -- switch PDF rendering from markdown-pdf to WeasyPrint for proper page breaks, page numbers, and no table background bleed.
- Marketplace: shipfast.cc
- Personal site: noahrasheta.com
- GitHub: github.com/noahrasheta/shipfast
MIT -- see LICENSE for details.