claude-skills

21 skills converted from Quickwork's internal skill set to the standard Agent Skills format (SKILL.md + scripts/references), for use with claude.ai or any environment that reads this format.

Layout

claude-skills/
├── install.sh           # macOS / Linux (Ubuntu) — installs .venv/ + node_modules/
├── install.ps1          # native Windows (PowerShell) — same, .venv\ + node_modules\
├── requirements.txt     # Python deps (pip)
├── package.json         # Node deps (npm)
├── skills/               # source — one folder per skill, edit here
└── packaged/             # ready-to-upload .skill files, one per skill

Setup

Installs everything locally to this folder (.venv, node_modules) — nothing is installed globally on the system. Safe to re-run on any OS.

macOS / Linux (Ubuntu, Debian, etc.):

./install.sh

Prerequisites if missing — the script tells you which, but for reference:

• macOS: brew install python3 node
• Ubuntu/Debian: sudo apt update && sudo apt install -y python3 python3-venv python3-pip nodejs npm

Windows (PowerShell):

Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass -Force   # only needed if scripts are blocked
.\install.ps1

Prerequisites: Python from https://www.python.org/downloads/ (check "Add python.exe to PATH" during setup) and Node.js from https://nodejs.org/.

Windows via WSL/Git Bash: use install.sh instead of install.ps1 — it runs the same as on Linux.

install.ps1 has been written carefully but not executed on a real Windows machine (none available here) — if it errors, the failing command and output will point at what's missing; report back and it can be fixed.

Uploading to claude.ai

Each file in packaged/ is a self-contained skill: Settings → Capabilities → Skills → "+ Create skill" → "Upload a skill", one .skill file at a time.

If you edit anything under skills/<name>/, re-package before uploading.

macOS / Linux:

cd skills/<name>
find . -name ".DS_Store" -delete
zip -r ../../packaged/<name>.skill . -x "*.DS_Store" -x "__MACOSX/*"

Windows (PowerShell):

cd skills\<name>
Compress-Archive -Path * -DestinationPath ..\..\packaged\<name>.zip -Force
Rename-Item ..\..\packaged\<name>.zip <name>.skill

Skill Index

Skill	What it does	Extra setup
plan-mode	Progress tracking via a script-backed markdown checklist	—
document-conversion	Convert between Markdown/HTML/PDF/DOCX	Python deps
html-design	Design system + scaffold for self-contained HTML widgets	—
engram-builder	Build a writing-style note from pasted message history	—
deep-analysis	Structured multi-track research with cited deliverable	needs plan-mode + document-conversion
word-documents	Create (.docx via docx-js) / edit (via python-docx) Word docs	Python + Node deps
pdf-documents	Create PDFs via ReportLab	Python deps
powerpoint-presentations	Create (pptxgenjs) / edit (patch engine) PowerPoint decks	Python + Node deps
excel-spreadsheets	Create/edit Excel workbooks, charts, formulas	Python deps
highcharts-charts	Highcharts API reference (40+ chart types)	none (CDN-loaded)
markdown-documents	Rich Markdown docs with embedded SVG/PNG charts	Python deps (optional, for charts)
knowledge-graph	File-backed nodes/edges knowledge graph	—
memory-notes	File-backed durable notes about the user (no auto cross-chat recall)	—
batch-processing	Sequential batch processing with tracked, structured results	optional pydantic
scheduled-monitoring	Generates real cron/launchd/Task Scheduler config + result log	user installs the generated job themselves
cli-agent-delegate	Shell out to a local CLI coding agent (Claude Code, aider, etc.)	needs the target agent CLI on PATH
web-fetch	Fetch/parse static web pages (no JS, no login)	Python deps
desktop-control	Mouse/keyboard/screenshot control	Python deps + requires a real attached display
skill-authoring-guide	Draft + package a new skill for claude.ai	—
skill-improvement-guide	Revise an existing skill from execution feedback	—
product-self-guide-builder	Build a "what can you do" self-guide for a specific product	—

Known Limitations (by design, not bugs)

• memory-notes and knowledge-graph don't persist across separate claude.ai conversations automatically — the user has to keep the JSON file and re-attach it.
• web-fetch can't read JavaScript-rendered pages (no real browser).
• desktop-control does nothing in a headless/hosted sandbox — it needs to run on a machine with a real screen attached.
• scheduled-monitoring and cli-agent-delegate generate config/commands but can't install a cron job or find/run an agent CLI that isn't actually present on the target machine.