Search the Practicum

Type two letters · ↑↓ select · Enter open · Esc close

Type at least two letters.

Libraries of ready-made skills and plugins

A living Practicum page. Checked: 2026-08-11. Skill catalogs change every month, so check the date and follow the source link. Repository contents, versions, and prices go stale quickly. The underlying principle does not.

A skill in plain language: a skill is an instruction file, usually named SKILL.md, sometimes paired with small scripts. An AI agent loads it when a task falls within its area. Instead of explaining "how we do this" every time, you describe the process once and reuse it.

This is a harness, the operating system around a model, made portable. You can keep it in a repository, give it to a team, and update it over time.

A plugin is a larger package built around a role. It can include several skills, service connections, and slash commands.

A. Where to find ready-made skills

Source What it contains Link Caveat
anthropics/skills The open Agent Skills standard from December 2025, plus ready-made skills for documents, design, development, and company communication. The repository includes spec/ and template/. github.com/anthropics/skills Official source and reference format
openai/skills A skill catalog for Codex that uses the same format github.com/openai/skills Built for the OpenAI and Codex ecosystem
NVIDIA/skills Signed skills for NVIDIA products. The catalog syncs from product repositories. Install with npx skills add nvidia/skills. github.com/NVIDIA/skills Contents change over time. It has a strong engineering focus. See section C for signing and verification.
VoltAgent/awesome-agent-skills More than 1,000 community skills across Microsoft tools, business, marketing, cloud services, data, and web development github.com/VoltAgent/awesome-agent-skills Community catalog. Check the author and read the files.
skillmatic-ai/awesome-agent-skills A curated list with an explanation of the architecture github.com/skillmatic-ai/awesome-agent-skills A useful starting point for understanding the format
phuryn/pm-skills More than 100 product management skills for hypotheses, interviews, prioritization, launch, and growth github.com/phuryn/pm-skills Wired to Western data sources. Check that they cover your market before relying on the output.
coreyhaines31/marketingskills Marketing skills for conversion rate optimization, copywriting, search engine optimization, analytics, and growth. Installs through SkillKit. github.com/coreyhaines31/marketingskills The same source caveat applies.

Check the data sources before you trust a kit. Mature product and marketing kits are wired to specific services, usually Crunchbase, Product Hunt, Google Trends, Reddit, and X. That wiring is invisible until it returns nothing useful for your market, your language, or your industry. Open the skill files and read which sources it actually queries before you rely on its output. If the sources do not cover your market, the kit is still worth keeping: swap the sources for ones that do and adjust the process. A ready-made kit plus a deliberate source swap beats writing one from scratch.

B. Role plugins as a ready-made function team

Anthropic created role-based Knowledge Work Plugins at github.com/anthropics/knowledge-work-plugins. The collection includes about 15 official plugins for Claude Cowork and Claude Code. Each one turns the agent into a specialist through a configuration file called a manifest, Model Context Protocol (MCP) connections, slash commands, and skills. Roles include sales, engineering, data, design, HR, legal, and finance.

This follows the same logic as the profession library. You do not need one universal AI. You need a set of roles, each guided and reviewed by a person.

An important limit: even a role plugin sends nothing without your confirmation. It can prepare work, but you approve the email, publication, or payment. The person stays in control. This is the human-in-the-loop principle from Book 1, Chapter 6.

C. Trusting a skill: signatures and scans

As skill catalogs grow, you need to ask whether an instruction file is safe. A skill may look like harmless Markdown, yet its wording can redirect an agent through prompt injection or hidden instructions.

What to watch Why it matters Status
NVIDIA Verified Agent Skills. The process uses SkillSpector, then a Skill Card and a cryptographic signature based on OpenSSF Model Signing. It records where a skill came from before the agent receives it. The catalog states that it checks instruction safety and supply-chain integrity. A valid signature shows that the published package has not changed. It does not prove that the skill is useful for your task.
OWASP Top 10 for large language models and OWASP Agentic AI Risks A shared vocabulary for risks in agent skills Public standards
MITRE ATLAS A map of attacks against AI systems Public resource

Rule for the reader: treat someone else's skill like someone else's code. Read what it does before you let an agent access email, files, money, or customer data. A signature or verification record is a useful trust signal, but it does not replace your judgment.

How to use these libraries

  1. When you find a repeated task, check whether one of the catalogs above already has a suitable skill.
  2. If it does not, build your own in 20 minutes with Turn a process into a skill.
  3. Use the profession library to decide what to automate for your role and which tools to use.

Update rule: review every quarter, add new entries at the top, and mark outdated entries instead of deleting them. Related pages: Personal system · Turn a process into a skill. Version: 2026-08-11.

Sources checked on 2026-08-11

Primary sources for skills and plugins:

Before you allow an outside skill into your environment, read what it does. Use the trust checks in section C.

Libraries of ready-made skills and plugins