🧠 AI Agents

Claude Skills Complete Reference Guide

📄 36 pages
📅 Published 23 April 2026
✍️ SimuPro Data Solutions
View Guide Summary & Sample on SimuPro → 📋 Browse Complete Guide Index →

What This Guide Covers

Claude Skills are the difference between Claude producing a good document and producing an expert-grade one — and this guide is the complete operational blueprint for every skill in the public library. Each skill is a curated Markdown file containing domain-specific best practices, tested code patterns, library API references, and explicit pitfall lists that Claude reads on demand before starting any matching task. The result is consistent, professional output that applies verified rules rather than training-data guesswork — every time.

This guide is structured across three parts and an appendix. Part I introduces the skill system architecture and catalogs all eight public skills — PDF, DOCX, PPTX, XLSX, Frontend Design, File Reading, PDF Reading, and Product Self-Knowledge — with full capability tables, key tools, and the critical rules that prevent the most costly failure modes. Part II delivers complete technical deep-dives: library APIs, command-line toolchains, QA pipelines, and annotated code patterns for each skill. Part III presents ten expert real-world examples spanning every skill individually and in multi-skill combination, culminating in the 4-output pipeline that produces four professional deliverables from a single source document. The appendix closes with an honest architectural comparison of Claude Skills against OpenAI Custom GPTs.

Any professional who uses Claude for document production, data workflows, or frontend development will find this guide directly applicable — whether you are adopting the public skill library, extending it with company-specific rules, or building entirely custom skills for proprietary domains.

36
Pages
8
Public Skills
10
Expert Examples
3
Parts + Appendix

The Skill System Architecture

The skill system has a precisely defined three-tier storage architecture. Public skills live at /mnt/skills/public/ — maintained by Anthropic, always available, covering the core document and development stack. User skills at /mnt/skills/user/ are custom files you create or upload, and they take highest priority, overriding any conflicting public skill of the same name. Example skills at /mnt/skills/examples/ include meta-tools like the skill-creator, used for building and benchmarking new skills.

Every skill entry point is a SKILL.md file with a YAML front-matter trigger description — the natural-language field Claude matches against your request semantics. Supporting files include REFERENCE.md for advanced options, FORMS.md for PDF-specific form-filling guidance, and a scripts/ directory containing bundled Python and Node.js helper scripts Claude can execute directly. The trigger mechanism is automatic: you never need to name a skill explicitly. Claude evaluates semantic intent against all available trigger descriptions, reads every matched SKILL.md in full before generating any output, then executes using the skill's verified rules and QA steps.

How Skills Are Triggered and Composed

Trigger matching is purely semantic — the words in your request are matched against the meaning of each skill's trigger description, not against a fixed keyword list. This means natural phrasing such as "make me a Word report," "build a React dashboard," or "merge these PDFs" reliably activates the correct skill without any special syntax. Critically, multiple skills can fire simultaneously for a single request. A task that reads a PDF and produces an Excel model from its tables invokes both the PDF Reading skill and the XLSX skill, each contributing its specialist guidance independently. The guide documents five common multi-skill composition patterns and the underlying content-map architecture that keeps multi-output pipelines consistent.

The Eight Public Skills — Capability Overview

The public skill library covers the full professional document production, data engineering, and frontend development stack. Each skill is designed to address a specific class of output where general AI heuristics are insufficient — where format-specific rules, library quirks, and QA mandates make the difference between a deliverable that works and one that silently fails.

Key Insight: Skills are not hardcoded into Claude's training. They are runtime knowledge files read fresh on every relevant request — which means they can be updated, extended, or replaced instantly without any model retraining. This makes the skill system one of the most adaptable professional automation mechanisms in the Claude ecosystem.

All Eight Public Skills at a Glance

PDF Skill
Create, merge, split, watermark, OCR, form-fill, and encrypt PDFs using ReportLab, pypdf, pdfplumber, and qpdf.
DOCX Skill
Generate and edit professional Word documents with tables, TOC, tracked changes, comments, and images via docx-js.
PPTX Skill
Design and build presentations with pptxgenjs — 10 named colour palettes, varied layouts, and a mandatory visual QA pipeline.
XLSX Skill
Engineer spreadsheets and financial models with openpyxl — formula-first principle, industry colour coding, and zero-error recalc policy.
Frontend Design Skill
Build production-grade React components and HTML/CSS/JS apps with distinctive, purpose-built aesthetics — no generic AI patterns.
File Reading Skill
Universal file ingestion router — determines the correct first move for every file type, preventing context-flooding and binary garbage.
PDF Reading Skill
Adaptive extraction strategy selection — text-based, scanned, table-heavy, form documents, and mixed-content PDFs each get the right approach.
Product Self-Knowledge Skill
Verified Anthropic product facts — exact model strings, API parameters, plan comparisons, and Claude Code setup — always from current documentation.

Deep-Dive Reference, Expert Examples & Custom Skills

Part II of the guide delivers the full technical reference for every skill. The PDF chapter covers the complete ReportLab Platypus building-block inventory — SimpleDocTemplate, Paragraph with inline XML markup, Table with TableStyle, KeepTogether, HRFlowable, and the onFirstPage/onLaterPages page-template callbacks — alongside the command-line toolchain (qpdf, pdftotext, pdftoppm, pdfimages, pdfinfo) and the five-step OCR pipeline for scanned documents. The DOCX chapter details the docx-js new-document workflow, the dual-width DXA table pattern that prevents rendering inconsistencies across Word, LibreOffice, and Google Docs, and the unpack → XML edit → pack workflow for modifying existing files. The PPTX chapter documents the full pptxgenjs object API, the ten named colour palettes with design rationale, and the five-step mandatory QA pipeline including soffice conversion and visual slide inspection. The XLSX chapter enforces the formula-first principle — every calculated value must be an Excel formula string, never a Python-computed result — with the industry colour-coding standard and mandatory recalc.py zero-error verification. The Frontend Design chapter defines the four-question aesthetic decision framework and explicitly names the anti-patterns Claude must never default to.

Part III's ten expert examples cover the full range from single-skill tasks to the ultimate multi-skill composition. Highlights include the DCF financial model with live formulas and a 2D sensitivity table, the OCR pipeline from scanned invoices to a formula-driven Excel workbook, the interactive React analytics dashboard with custom recharts components and animated KPI cards, and the 4-output pipeline that extracts once into a shared content map and simultaneously produces a Word executive summary, an Excel data workbook, a PowerPoint management presentation, and a four-page branded PDF briefing. Each example walks every step from skill-file reading to final QA validation, making them directly reusable as workflow templates.

The custom skill section, supported by the skill-creator meta-skill, provides a complete five-step framework for building institutional knowledge files — from writing a precise trigger description to benchmarking accuracy against representative requests. The Appendix rounds out the guide with a detailed comparison of Claude Skills and OpenAI Custom GPTs across eight architectural dimensions, with clear guidance on when each approach is the right investment.

Topics Covered in This Guide

Read the Full Guide + Download Free Sample

36 pages · Instant PDF download · Available in the SimuPro Knowledge Store

View Guide Summary & Sample on SimuPro → 📋 Browse Complete Guide Index →

Frequently Asked Questions

What are Claude Skills and how are they different from Claude's built-in capabilities?
Claude Skills are modular Markdown files containing structured expert guidance — best practices, API references, code patterns, and explicit pitfall lists — that Claude reads before generating output for a matching task. They are not hardcoded into the model; they are runtime knowledge files loaded on demand. The key difference from Claude's baseline capabilities is consistency and quality: without a skill, Claude applies general training-data heuristics; with a skill, it follows verified, domain-specific rules that prevent known failure modes and enforce professional standards on every matching request. Skills can also be updated instantly without any model retraining, making them far more maintainable than trained knowledge for fast-moving technical domains.
How does Claude know which skill to apply to my request?
Each skill has a trigger description in its SKILL.md front matter. When you submit a request, Claude performs semantic matching between your intent and each available skill's trigger description — you do not need to name a skill explicitly. For example, a request mentioning "Word document," "report," or ".docx" automatically triggers the DOCX skill, while "slides," "deck," or "presentation" triggers the PPTX skill. Multiple skills can fire simultaneously for a single request when the task spans several domains, and the skill selection process has five explicit steps: intent analysis, skill matching, SKILL.md reading, guided execution, and QA assurance.
Which skill should I use for creating a professionally branded multi-page PDF?
The PDF skill, which uses Python's ReportLab Platypus layout engine. It covers the complete creation workflow: BaseDocTemplate with explicit Frame geometry (never SimpleDocTemplate, which adds 6pt padding causing misalignment), onFirstPage and onLaterPages callbacks for running headers and footers, the complete Flowable building-block inventory, and critical rules such as never using Unicode subscript characters in ReportLab (use XML sub and super tags inside Paragraph objects instead — Unicode renders as solid black boxes). The guide's Example 01 walks the complete branded multi-page report workflow from story-list planning to final output validation, including the KeepTogether pattern that prevents awkward mid-table page breaks.
Can multiple Claude Skills fire at the same time for a single request?
Yes — multi-skill composition is one of the most powerful features of the system. Complex tasks naturally invoke several skills simultaneously: reading a financial PDF and building an Excel model from its tables triggers both the PDF Reading skill and the XLSX skill; converting Word documents into a summary presentation invokes the DOCX skill for content extraction and the PPTX skill for slide design and QA. There is no penalty for invoking multiple skills, and each contributes its specialised expert guidance independently. The guide covers five common multi-skill composition patterns and the "ingest once, output many" architectural principle that guarantees consistent data across all outputs in a multi-format pipeline.
How do I build a custom skill for my company's specific document conventions?
Use the skill-creator meta-skill at /mnt/skills/examples/skill-creator/SKILL.md. The process has five steps: decide what institutional knowledge the skill should encode, write a precise trigger description (the single most important step — it determines whether your skill fires reliably), structure the SKILL.md with an Overview, Quick Reference table, Detailed Workflow with code examples, Critical Rules section, and Dependencies list, save it to /mnt/skills/user/your-skill-name/SKILL.md (user skills take highest priority over public skills), then benchmark against representative requests and iterate. The guide's Example 09 walks the complete build of a SimuPro Branded PDF skill that encodes company colour constants, the rainbow stripe pattern, and page template geometry as a single version-controlled source of truth.
What is the 4-output pipeline and when should I use it?
The 4-output pipeline (Example 10 in Part III) reads a single source PDF and simultaneously produces four professional deliverables: an executive Word summary, a live-formula Excel data workbook, a management PowerPoint presentation, and a branded PDF briefing — all from the same extracted content map. The core architectural principle is "ingest once, output many": all data is extracted from the source document into a shared Python dictionary once, then passed to each builder function. This guarantees identical numbers across all four outputs with no risk of re-extraction inconsistencies. It is the recommended pattern whenever a report or analysis needs to be distributed in multiple formats to different stakeholder groups.
How does Claude Skills compare to OpenAI Custom GPTs?
They solve different problems. A Custom GPT is a packaged mini-assistant with a fixed persona, system prompt, and tools — ideal for distributing a specialised assistant to non-technical users via the GPT Store, with a shareable link and no setup required. A Claude Skill is a plain Markdown file encoding operational expertise for a specific technical domain — it changes what Claude knows to do, not who Claude is. Skills are composable (multiple fire per request), version-controllable as plain text in Git, private to your own environment (never transmitted to third-party servers), and updateable without retraining. For consistent expert-grade document production governed by precise institutional rules, Claude Skills are the right investment; for consumer-facing distribution with a branded persona, Custom GPTs win. The guide's Appendix covers the full side-by-side comparison across eight dimensions.

Brief Summary

Claude Skills are modular knowledge packages that transform Claude from a capable generalist into a domain expert — reading curated SKILL.md files before every matching task to apply verified best practices, tested code patterns, and explicit pitfall lists. This guide is the definitive reference for every skill in the public library: PDF creation and manipulation, Word document engineering, PowerPoint design, Excel financial modelling, frontend React development, universal file reading, PDF content extraction, and Anthropic product knowledge.

Each skill is documented at three levels: trigger conditions and capability summary for quick reference, full technical deep-dives covering library APIs, command-line toolchains, QA pipelines, and critical rules, and annotated real-world examples that walk every workflow from first read to final validated output. Multi-skill composition patterns show how complex tasks invoke several skills simultaneously — and how the outputs of each integrate into a single coherent deliverable.

You will finish this guide knowing exactly which skill fires for any given request, how to extend or override the public library with your own institutional rules, how to benchmark trigger accuracy, and how to build a complete 4-output report pipeline that produces a branded PDF, an Excel model, a PowerPoint deck, and a Word summary from a single source document — all at expert grade.

Extended Summary

If you use Claude for any kind of professional document production — PDFs, spreadsheets, presentations, Word reports, web components, or data extraction — this guide gives you the complete operational blueprint: how the skill system works, what each of the 8 public skills can do, and exactly how to extend it with your own custom rules.

Part I opens with a precise architecture walkthrough: how skills are stored in layered filesystem directories, how Claude matches skill trigger descriptions against your request semantics, and how multiple skills compose simultaneously for multi-format tasks. The comprehensive skill catalog then documents all eight public skills — PDF, DOCX, PPTX, XLSX, Frontend Design, File Reading, PDF Reading, and Product Self-Knowledge — with capability tables, key tools, and the critical rules that prevent the most common failure modes.

Part II delivers full technical deep-dives for every skill: the complete ReportLab Platypus building-block reference, the DOCX dual-width table pattern and XML unpack/edit/pack workflow, the pptxgenjs object API and 10 named colour palettes with mandatory visual QA pipeline, openpyxl formula-first principles and zero-error recalc policy, and the Frontend Design aesthetic system with its explicit anti-pattern list. File Reading adaptive protocols and PDF Reading strategy selection round out the reference.

Part III presents 10 expert real-world examples spanning every skill individually and in combination: a branded multi-page ReportLab PDF, a DCF financial model with sensitivity table, a full-design-system pitch deck, contract editing with tracked changes, an OCR pipeline from scanned invoices to Excel, an interactive React analytics dashboard, a merge-watermark-encrypt PDF knowledge base, messy CSV cleaning to formatted XLSX, a custom domain skill built with the skill-creator, and the ultimate 4-output pipeline that produces four professional deliverables from a single PDF source.

The Appendix provides an honest side-by-side comparison of Claude Skills against OpenAI Custom GPTs — architectural differences, when each approach wins, and a step-by-step worked example building the SimuPro Branded PDF custom skill that encodes company-specific colour constants, rainbow stripe pattern, and page template geometry as a living, version-controlled institutional knowledge file.

SimuPro Data Solutions
SimuPro Data Solutions
Cloud Data Engineering & AI Consultancy  ·  AWS  ·  Azure  ·  GCP  ·  Databricks  ·  Ysselsteyn, Netherlands  ·  simupro.nl
SimuPro is your end-to-end cloud data solutions partner — from in-depth consultancy (research, architecture design, platform selection, optimization, management, team support) to tailor-made development (proof-of-concept, build, test, deploy to production, scale, automate, extend). We engineer robust data platforms on AWS, Azure, Databricks & GCP — covering data migration, big data engineering, BI & analytics, and ML models, AI agents & intelligent automation — secure, scalable, and tailored to your exact business goals.
Data-Driven AI-Powered Validated Results Confident Decisions Smart Outcomes

Related Guides in the SimuPro Knowledge Store

SimuPro Data Solutions — Cloud Data Engineering & AI Consultancy

Expert PDF guides · End-to-end consultancy · AWS · Azure · Databricks · GCP

Visit simupro.nl →
📋 Browse All Guides — Complete Index →