Skills for the
AI Agent Era
Skills are portable instruction sets that tell your AI agent how to operate within a specific context. They extend what Claude, Gemini, and other agents know how to do — turning a general-purpose AI into a specialized practitioner.
This page is TVN's guide to navigating the skills ecosystem safely and strategically — not just finding skills, but understanding them, adapting them, and building your own.
What is a Skill?
A skill is a SKILL.md file — a structured set of instructions, context, and workflow patterns that you load into an AI agent. When an agent has a skill installed, it operates with that skill's knowledge and constraints in scope, without you having to re-explain your context every session.
Skills are how you turn Claude or Gemini from a general assistant into your Strategy Engineer, your ServiceNow Architect, your Org Design Practitioner, or any specialized role you need it to play — consistently and at scale.
As of early 2026, the Agent Skills standard — announced by Anthropic in December 2025 — has been adopted by 16+ major AI tools including Claude Code, Gemini CLI, Cursor, Codex, and others. Skills you build are portable across agents.
When NOT to use a Skill
Skills are powerful, but they are not the right tool for every job. A skill shapes how an AI agent thinks — it does not replace dedicated infrastructure, persistent processes, or production-grade automation. Using a skill where a discrete agent is warranted is one of the most common mistakes in early AI adoption.
If any of the conditions on the right describe your use case, you are past the boundary of what a SKILL.md can safely deliver. At that point, you need a purpose-built AI agent — deployed as its own system, with its own tools, memory, and execution environment.
You need guaranteed, repeatable execution
Skills are interpreted by an LLM. They guide behavior — they don't enforce it. For workflows that must execute identically every time, a deterministic agent with hard-coded logic is the right answer.
The workflow runs at volume or without human oversight
A skill-guided agent is designed for collaborative, human-in-the-loop work. High-volume, automated pipelines (hundreds of runs per day, unattended) need dedicated infrastructure — not a conversational agent with a prompt overlay.
Compliance or audit trails are required
Skills leave no native audit log. If your organization requires a traceable record of every action taken — for SOC 2, HIPAA, or internal policy — a discrete agent with proper logging and access controls is mandatory.
Credentials or sensitive data are in scope
A skill is a text file that rides inside a conversation. It should never contain secrets, and it cannot enforce data isolation. Workflows touching credentials, PHI, or client confidential data need security-isolated agent infrastructure.
The task requires persistent memory across many sessions
Skills are stateless. They do not remember what happened last week. If your workflow depends on accumulating context across dozens of sessions over time, you need an agent with a proper memory layer — a vector store, a knowledge graph, or a structured database.
The Alternative
A discrete AI agent is a purpose-built system: its own execution environment, its own tool integrations, its own memory, and its own access controls. Think of it as the difference between a reference manual (skill) and a specialist you've hired full-time (agent). Partners like Echelon AI and OperatorZero on the TVN network build exactly this — agents deployed inside your ServiceNow instance that operate at production scale, with full traceability.
The TVN Approach to Skills
There are 700,000+ skills available across marketplaces and repos. The question isn't which ones to install — it's how to think about adopting any skill into your professional AI practice.
Your skill is a reflection of how you work
The most powerful skill you can build isn't one you found in a library — it's one you wrote yourself, that encodes your specific methodology, your domain knowledge, and your professional standards. Community skills are starting points. Your own skills are professional assets.
Start with the official repo
Before pulling from any community library, read through Anthropic's official skills first. They show you the standard — what a well-structured skill looks like, what it should and shouldn't do, and what the agent is actually being asked to execute.
Treat every third-party skill as a prompt you haven't read yet
A skill is a set of instructions your AI agent will follow. Before installing, open the SKILL.md file and read it. Ask: does this instruct the agent to call external URLs? Does it request permissions it doesn't need? Would I be comfortable if a new employee followed these instructions?
Use community skills as examples, not installs
The most valuable use of a skills library isn't installing the skill — it's using it as a reference for Claude or Gemini to adapt. Paste a skill into your session and say: "Understand the pattern here, then create a version scoped to [your specific context], without inheriting anything you can't explain." This gives you the benefit of community thinking without the risk of opaque instructions.
Scope every skill to its purpose
A skill that does too much is a skill you can't trust. If a community skill bundles five different behaviors, break it apart. The narrower the skill's scope, the easier it is to verify what it's actually doing — and the more reliably your agent will use it correctly.
Document your skill decisions
Your PKG (Personal Knowledge Graph) on TVN will eventually capture the skills you've built and deployed. Start treating your skill library as a professional asset — not a collection of downloads, but a set of deliberate decisions about how your AI agents operate on your behalf.
Skill Libraries & Repos
These are the most referenced sources in the skills ecosystem. Treat them as a research library — read before you install, adapt before you deploy.
Anthropic Official Skills
github.com/anthropics/skills
The canonical reference. Skills built and maintained by Anthropic — the authoritative baseline for understanding how skills are designed, structured, and intended to be used. 107k+ stars.
Google Gemini CLI Skills
geminicli.com
Official Google documentation for skills in Gemini CLI. Full getting-started guide, skill authoring reference, and tutorials — the authoritative source for Gemini-native skills.
SkillsMP Marketplace
skillsmp.com
700,000+ agent skills with filtering by occupation, author, and popularity. The largest searchable skills marketplace — useful for discovering niche use cases across every domain.
SkillHub
skillhub.club
7,000+ AI-evaluated skills for Claude, Codex, Gemini, and OpenCode. Every skill is rated for quality and security before listing — one of the most carefully moderated skill directories available.
Awesome Claude Skills
ComposioHQ/awesome-claude-skills
Curated Claude skills organized by category and workflow, maintained by Composio. Quality-focused with nearly 50k stars — the highest-starred community collection for Claude specifically.
Awesome Agent Skills
VoltAgent/awesome-agent-skills
1,000+ skills compatible with Claude, Gemini CLI, Cursor, Codex, and more. The broadest cross-platform library available — useful for discovering what's possible across the ecosystem. 13.5k stars.
Official Platform Documentation
Claude Code Skills Documentation
Official Anthropic documentation on extending Claude with skills — how they work, how to install them, and how to build your own.
Gemini Agent Skills Guide
Google's official guide to setting up Gemini with MCP and Skills — the authoritative reference for Gemini-native skill development.
Build skills that reflect how you work
Value Producers on TVN are building AI agents that encode their domain expertise and deliver it at scale. Your skills are the difference between an AI that behaves generically and one that operates as your professional extension.