⚡ skillpack

One command replaces an afternoon of clicking

AI-powered terminal skills that turn multi-click professional workflows into a single command.

Extract data from PDFs · Wrangle CSVs · Scaffold code · Generate infrastructure configs · Process content · Batch-rename files · Parse invoices · Manage resumes

Works with Claude Code, OpenCode, Codex CLI, and any MCP-compatible AI agent.

curl -fsSL https://raw.githubusercontent.com/muhammad-saadd/skillpack/main/install.sh | bash

Features · Install · Quickstart · Skills · How it Works · Contributing

---

🎯 What it does

skillpack is a curated collection of 8 battle-tested terminal skills that automate the boring, repetitive, click-heavy tasks that fill a professional's day. Each skill is a tiny POSIX shell script — no Node.js, no Python, no Docker — just awk, sed, jq, and a handful of system tools.

Every skill ships as two files:

File	Purpose
SKILL.md	A machine-readable context file your AI CLI reads to understand what the skill does, when to trigger it, and how to call it.
index.sh	A standalone POSIX shell script that does the actual work — file I/O, parsing, transformation, output. Runs with or without an AI agent.

This means the same skill works three ways:

1. Standalone — skillpack doc-extract invoice.pdf from any terminal.
2. AI-driven — Claude Code / OpenCode / Codex reads the SKILL.md, understands intent, and calls index.sh for you.
3. Piped — cat invoice.pdf | skillpack doc-extract --format json from a shell pipeline.

---

💡 The problem

Every professional task requires context switching between 5+ apps. A single invoice processing workflow might look like:

Adobe Acrobat → Excel → Google Sheets → email → Slack

That's 45 minutes of clicking for something a terminal can do in 5 seconds.

Task	Old way	With skillpack
Parse 50 invoices	Open each PDF → copy data → paste into Excel (2 hours)	skillpack finance-extract invoice invoices/ --batch
Convert CSV to JSON	Open in Excel → Save As → fix formatting (15 min)	skillpack data-wrangle data.csv --to json
Generate K8s from Compose	Google "docker-compose to k8s" → copy-paste YAML (1 hour)	skillpack infra-gen k8s docker-compose.yml
Rename 200 photos	Right-click each → Rename → type new name (45 min)	skillpack file-ops rename photos/ --pattern "{date}-{seq}-{name}"
Parse resumes for hiring	Open each PDF → copy info → build spreadsheet (3 hours)	skillpack people-ops resume resumes/ --batch
Generate favicon set from logo	Open Sketch/Figma → export 6 sizes (20 min)	skillpack file-ops favicon logo.svg --out public/
Summarize a 50-page paper	Read the whole thing (90 min)	skillpack content-craft summarize paper.pdf --style executive
Create a job description	Open Google Docs → write from scratch (30 min)	skillpack people-ops jd --role "Senior Engineer"

---

📦 Install in 30 seconds

Option 1 — curl installer (recommended)

The curl installer auto-detects your package manager and installs both skillpack and its system dependencies:

curl -fsSL https://raw.githubusercontent.com/muhammad-saadd/skillpack/main/install.sh | bash

It handles: apt (Debian/Ubuntu) · dnf/yum (Fedora/RHEL) · pacman (Arch) · brew (macOS) · apk (Alpine).

Option 2 — clone + symlink

If you prefer to install the system tools yourself first:

# 1. Install system dependencies (pick your platform)
sudo apt install -y poppler-utils jq imagemagick docx2txt   # Debian/Ubuntu
brew install poppler jq imagemagick pandoc                 # macOS

# 2. Clone and symlink
git clone https://github.com/muhammad-saadd/skillpack.git ~/.local/share/skillpack
ln -s ~/.local/share/skillpack/bin/skillpack ~/.local/bin/skillpack

# 3. Make sure ~/.local/bin is on your PATH
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc   # or ~/.bashrc

Option 3 — install via npm

npm install -g @muhammad-saadd/skillpack

Verify

skillpack --version          # → skillpack v1.0.3
skillpack --help             # → lists all 8 terminal skills

---

🧩 Install Greyboard and Orangebox

The same two workflow skills have portable versions for Codex and OpenCode. Install both globally with one command:

npx skills add muhammad-saadd/skillpack --skill orangebox greyboard --global --agent codex opencode --yes

This installs only these workflows, not the eight terminal skills. Codex can invoke them as $orangebox and $greyboard; OpenCode can invoke them as /orangebox and /greyboard or let the agent select them automatically. Orangebox uses each host's own subagent tool for implementation and independent reviews. If separate agent contexts are unavailable, it marks those review gates blocked instead of claiming they passed. Codex and OpenCode do not run Claude's plugin hooks, so the portable instructions keep file ownership explicit and require the coordinator to inspect changes before integration.

Claude Code plugin marketplace

This repo is also a Claude Code plugin marketplace with two workflow plugins:

Plugin	What it does
greyboard	A senior staff engineer that interrogates the problem, challenges assumptions and shows the design as an HTML view before any code is written. Trigger: /greyboard
orangebox	A multi-agent workflow: grill the problem → diagram → PRD + task plan → build on a local branch → independent QA + requirements-fidelity review. Includes grill, architect, lead, engineer, qa, final-review and five eng-* subagents. Trigger: /orangebox <task>

/plugin marketplace add muhammad-saadd/skillpack
/plugin install greyboard@skillpack
/plugin install orangebox@skillpack

The Claude plugins retain their native agents, hooks, and Artifact integration. They need git and node on PATH; Orangebox diagrams publish only where Claude Artifacts are available. Greyboard uses other installed skills when available and falls back to its built-in lightweight process.

---

⚡ 60-second quickstart

# 1. Extract text from a PDF
skillpack doc-extract report.pdf --format markdown

# 2. Clean and deduplicate a CSV, then convert to JSON
skillpack data-wrangle sales.csv --clean --dedup --to json

# 3. Generate a favicon set from a logo
skillpack file-ops favicon logo.svg --out public/

# 4. Turn a docker-compose.yml into a Kubernetes manifest
skillpack infra-gen k8s docker-compose.yml --namespace production

# 5. Summarize a long paper
skillpack content-craft summarize paper.pdf --style executive

---

🧰 The 8 skills

1. doc-extract — Document data extraction

What it does, step by step:

1. Detects the input file type (PDF, DOCX, PNG, JPG, TIFF) from extension + magic bytes.
2. Routes to the right parser (pdftotext for PDFs, docx2txt/pandoc for DOCX, OCR for images).
3. If --ocr is set and the file is a scanned image, runs OCR text extraction.
4. If --schema is specified (e.g. invoice, resume, contract), applies a structured template.
5. If --extract is specified, returns only the requested elements (tables, clauses, metadata).
6. Formats output as markdown, csv, json, or txt.
7. Writes to --out <dir> or prints to stdout.

skillpack doc-extract invoice.pdf --format json --schema invoice
skillpack doc-extract resume.pdf --format json --schema resume
skillpack doc-extract *.pdf --format csv --out ./extracted/
skillpack doc-extract scanned.pdf --ocr --format markdown

Output:

{ "vendor": "Acme Corp", "invoice_number": "INV-2024-001", "total": 542.50 }

---

2. data-wrangle — CSV / JSON / XLSX automation

What it does, step by step:

1. Detects input format (.csv, .tsv, .json, .xlsx).
2. Normalizes encoding (fixes UTF-8 mojibake, BOM, line endings).
3. Trims whitespace, normalizes date formats, infers column types.
4. Applies requested transforms: --clean → trim/normalize · --dedup → drop duplicates · --merge → join on key · --validate → schema check.
5. Optionally converts to a different format (--to json|csv|tsv|sql).
6. Optionally generates SQL from a natural-language description (--sql "total revenue by region").
7. Writes to --out or stdout.

skillpack data-wrangle sales.csv --clean --fix-encoding --dedup
skillpack data-wrangle jan.csv feb.csv --merge --key "order_id"
skillpack data-wrangle "total revenue by region from sales.csv" --sql

Output:

SELECT SUM(revenue) FROM sales GROUP BY region;

---

3. code-scaffold — Project & code generation

What it does, step by step:

1. Parses the subcommand: --new <name> (project), <file> (test gen), or template (README, .env.example, etc.).
2. For --new: detects/uses --stack flag (e.g. node express typescript) and generates the full directory tree.
3. For test generation: reads the source file, detects function signatures, emits a test file in the chosen framework.
4. For template generation: writes a best-practice README.md, .env.example, API docs, CHANGELOG.md, or TypeScript types from JSON.
5. Skips files that already exist (unless --force is set).
6. Prints a summary of what was created.

skillpack code-scaffold --new my-api --stack "node express typescript"
skillpack code-scaffold tests src/utils.ts --framework vitest
skillpack code-scaffold types api-response.json --out src/types/api.ts

Output:

Created directory: my-api/src
Created file: my-api/package.json
Created file: my-api/tsconfig.json
Created file: my-api/src/index.ts

---

4. infra-gen — Infrastructure as code

What it does, step by step:

1. Detects the subcommand: k8s, ci, nginx, apache, terraform.
2. For k8s <compose-file>: parses docker-compose.yml, maps services → Deployments, volumes → PVCs, ports → Services, env → ConfigMaps.
3. For ci <dir>: scans the project root for language markers (package.json, requirements.txt, go.mod, etc.) and generates a working CI pipeline.
4. For terraform "<description>": emits a Terraform skeleton for the described architecture.
5. Applies the requested namespace, environment, or provider flags.
6. Writes one or more files to --out and prints what was generated.

skillpack infra-gen k8s docker-compose.yml --namespace production
skillpack infra-gen ci . --provider github-actions
skillpack infra-gen terraform "AWS: VPC + ECS + RDS + S3"

Output:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: api
  namespace: production

---

5. content-craft — Content processing

What it does, step by step:

1. Accepts a file, a directory of files, or stdin.
2. Routes to the requested action: proofread, summarize, translate, repurpose, actions, reply.
3. For summarization, applies the chosen --style (executive, bullet, eli5, tldr).
4. For proofing, optionally applies fixes in place (--fix) or returns a diff.
5. For action-item extraction, scans meeting notes and emits a checkbox-style task list.
6. Writes the result to --out or stdout.

skillpack content-craft proofread docs/ --fix
skillpack content-craft summarize paper.pdf --style executive
skillpack content-craft actions meeting-notes.md --out actions.md

Output:

## Executive Summary

This paper presents a novel approach to container orchestration…

---

6. file-ops — Batch file operations

What it does, step by step:

1. Accepts a subcommand: rename, images, favicon, tokens.
2. For rename <dir> --pattern <tmpl>: walks the directory, supports pattern tokens ({date}, {seq}, {name}, {ext}, counter padding).
3. For images <file>: converts to requested formats (webp, avif, jpg, png) and generates responsive sizes (1x, 2x, 3x).
4. For favicon <logo.svg>: emits a full favicon set (16, 32, 48, 64, 128, 192, 512) plus manifest.json and apple-touch-icon.
5. For tokens <tokens.json>: converts a design-token JSON into a ready-to-import CSS :root { --… } block.
6. Shows a dry-run by default; --apply to commit changes.

skillpack file-ops rename photos/ --pattern "{date}-{seq}-{name}"
skillpack file-ops images hero.png --to webp,avif --sizes "1x,2x,3x"
skillpack file-ops tokens design-tokens.json --out src/styles/tokens.css

Output:

:root {
  --color-primary: #3B82F6;
  --spacing-md: 16px;
  --font-size-base: 16px;
}

---

7. finance-extract — Financial data extraction

What it does, step by step:

1. Detects the financial document type: bank statement, invoice, or receipt.
2. Optionally uses --bank (chase, wells, hsbc, etc.) for parser-tuned extraction.
3. Extracts transactions, line items, totals, and metadata.
4. Applies --categorize to label each transaction (meals, travel, office, software, etc.).
5. Aggregates by --period (2024-Q4, 2025-01, last-month, etc.) when requested.
6. Emits a CSV, JSON, or human-readable report.

skillpack finance-extract statement bank.pdf --bank chase --categorize
skillpack finance-extract invoice invoices/ --batch --out data.csv
skillpack finance-extract report receipts/ --period "2024-Q4"

Output:

date,description,amount,category
01/15/2024,STARBUCKS,4.95,meals
01/16/2024,AMAZON.COM,89.99,office

---

8. people-ops — Resume & HR automation

What it does, step by step:

1. Accepts a subcommand: resume, jd, onboard.
2. For resume <file>: parses the PDF/DOCX into a structured JSON (name, email, experience, education, skills).
3. Supports batch mode: resume resumes/*.pdf builds a candidate spreadsheet.
4. For jd --role "X" --level Y: generates a complete job description with responsibilities, requirements, and interview questions.
5. For onboard --role "X" --start YYYY-MM-DD: builds a role-specific week-by-week onboarding checklist.
6. Writes JSON / Markdown to --out or stdout.

skillpack people-ops resume candidate.pdf --format json
skillpack people-ops jd --role "Senior Engineer" --level senior
skillpack people-ops onboard --role "Frontend Dev" --start "2025-02-01"

Output:

# Onboarding Checklist — Frontend Dev

**Start Date:** 2025-02-01

## Week 1: Getting Started

### Day 1 — First Day
- [ ] Welcome meeting with manager
- [ ] Team introductions

---

🤝 Works with your AI CLI tool

Tool	Support	Notes
Claude Code	✅ Full	SKILL.md + MCP native — drop skills/ into your Claude config
OpenCode	✅ Full	Drop skills/ folder into your OpenCode config path
Codex CLI (OpenAI)	✅ Full	index.sh scripts run standalone
Aider	✅ Partial	Use as standalone scripts
Cursor	✅ Partial	Run from the integrated terminal panel
Any terminal	✅ Always	curl / clone install works everywhere

---

🏗️ How it works

Each skill has two parts: a machine-readable context file (SKILL.md) and a standalone shell script (index.sh).

        You:  "Extract the invoice total from this PDF"
                                    │
                                    ▼
                  ┌──────────────────────────────────┐
                  │  AI CLI (Claude / OpenCode / …)  │
                  │  reads SKILL.md for context      │
                  └─────────────┬────────────────────┘
                                │
                                ▼
                  ┌──────────────────────────────────┐
                  │  AI calls: skillpack doc-extract │
                  │  invoice.pdf --format json       │
                  └─────────────┬────────────────────┘
                                │
                                ▼
                  ┌──────────────────────────────────┐
                  │  index.sh executes               │
                  │  (pdftotext → awk → sed → jq)    │
                  └─────────────┬────────────────────┘
                                │
                                ▼
                  ┌──────────────────────────────────┐
                  │  Structured output to stdout /   │
                  │  --out directory                 │
                  └──────────────────────────────────┘

The whole pipeline is local. No cloud calls, no telemetry, no API keys required for any skill.

---

🪶 Zero dependencies

• No node_modules. No pip install. No Docker.
• Pure POSIX shell — runs on any Linux, macOS, or WSL system out of the box.
• Optional system tools for full feature coverage: pdftotext, jq, imagemagick, docx2txt/pandoc. The curl installer adds them automatically.
• No API keys needed for any skill. AI CLI features (smart replies, better summaries) work better with an AI tool, but everything runs standalone.

---

❓ FAQ

Does this send my files to the cloud? No. Every skill runs 100% locally. No data leaves your machine unless you wire up an external API yourself.

Does it require an API key? No. All eight terminal skills work standalone. They become smarter when an AI CLI is connected (e.g. Claude Code reads SKILL.md for context), but none of them require it.

Can I use it without an AI CLI tool? Yes. Every index.sh is a standalone POSIX shell script. You can install via curl and use only the CLI, or clone the repo and run any skill directly.

How do I add my own skill? See CONTRIBUTING.md. The short version: create a folder under skills/ with three files — SKILL.md, index.sh, README.md — following the existing patterns. Most skills take ~30 minutes to add.

Can I install only specific skills? Yes. Each skill is independent. Clone the repo and copy only the skill folders you need into your own skills/ directory, or install globally and call only the ones you want.

Does it work on Windows? Yes via WSL. Native Windows support is on the roadmap — the scripts are POSIX-clean and most of them work under Git Bash, but WSL is the recommended path.

How is this different from a Claude skill, an OpenCode skill, etc.? Most AI CLI skills are prompts — they tell the model what to do. skillpack skills are prompts + execution — the index.sh does the real work, so results are deterministic, fast, and reproducible. The AI only decides what to do; the shell does how.

---

🗺️ Roadmap

• Native Windows support (PowerShell wrappers)
• More skills: git-ops, docker-debug, log-analyze, sql-migrate
• Pluggable schema registry for doc-extract
• Interactive skillpack init wizard
• Skill marketplace (community-contributed skills)

Have an idea? Open an issue.

---

🤝 Contributing

We love contributions! Adding a new skill takes about 30 minutes if you follow the template.

1. Read CONTRIBUTING.md.
2. Create skills/<your-skill>/ with SKILL.md, index.sh, README.md.
3. Make sure bash -n skills/<your-skill>/index.sh passes (no syntax errors).
4. Open a PR.

See existing skills (e.g. skills/doc-extract/) for the canonical structure.

---

📝 License

MIT © 2026 Muhammad Saad

---

⭐ Show your support

If skillpack saved you an afternoon of clicking, consider giving it a ⭐ on GitHub — it helps other developers find it.

Built with ⚡ by Muhammad Saad — because life is too short for multi-click workflows.