Skip to content

Deploy Agent Skills

Give AI agents access to knowledge, procedures, and operational playbooks — no API endpoint required. Agent Skills are structured instructions that AI models discover progressively, loading only what's relevant to the current task.

This is the deployment mode for SOPs, compliance checklists, troubleshooting guides, onboarding flows, and any domain knowledge your AI needs to reason correctly.

No API
PURE KNOWLEDGE
SKILL.md
OPEN STANDARD
Progressive
DISCLOSURE MODEL

How agent skills differ from API tools

API-based MCP servers give AI the ability to do things — query databases, create records, trigger webhooks. Agent Skills give AI the ability to know things — how your refund policy works, which escalation steps to follow, what your compliance requirements are.

The AI discovers skills through progressive disclosure: it sees the skill name and description first, then loads the full body only when the skill is relevant to the conversation. This keeps the context window clean and prevents information overload.

The SKILL.md open standard

SKILL.md is an open standard for packaging AI procedural knowledge, published at agentskills.io. Initially developed by Anthropic and publicly released in December 2025, it defines a structured Markdown format that gives AI agents persistent domain expertise — turning them from generic assistants into specialists that follow your exact procedures.

The standard is supported natively by every major AI coding platform:

Claude Code
Anthropic
OpenAI Codex
OpenAI
Gemini CLI
Google
GitHub Copilot
Microsoft
Cursor
Anysphere
VS Code
Microsoft
Windsurf
Codeium
Antigravity
Google DeepMind

What Vinkius Cloud changes

On every platform above, skills exist as local files in a repository. They live in .cursor/skills/, .claude/skills/, or similar directories. They work — but only for that specific developer, in that specific project, in that specific IDE session.

Vinkius Cloud takes the same SKILL.md standard and deploys it to the cloud as a governed MCP server:

LOCAL SKILL.MD
Lives in a git repo
One developer, one project, one IDE. No access control, no audit trail, no DLP. Skills are discoverable only in the local workspace.
VINKIUS CLOUD
Deployed to the global edge
Any MCP client, any team member, any IDE. HMAC-authenticated tokens, DLP redaction, 40ms revocation, full audit log. Skills become organizational infrastructure.

The same refund policy, the same escalation playbook, the same compliance checklist — accessible from Claude Desktop, from Cursor, from a custom agent, from any MCP-compatible client. Managed centrally, governed uniformly, updated once for everyone.


THE WIZARD
Teach AI
how to think.
Skills are structured instructions that AI agents discover progressively — only loading what's needed for the current task. Less noise, better reasoning.

Step 1 — Define your skills

01 — SKILLS
Teach AI
how to think.
Skills are structured instructions that AI agents discover progressively — only loading what's needed for the current task. Less noise, better reasoning.
Following the SKILL.md open standard from agentskills.io.
SKILLS (3)
Refund Policy
ACTIVE
Step-by-step procedure for processing customer refunds within 30 days.
Escalation Playbook
ACTIVE
Rules for when and how to escalate a conversation to a human agent.
Product FAQ
ACTIVE
Canonical answers for the most common customer questions.
+ Add Skill

Each skill has three fields:

FieldPurpose
NameShort identifier (up to 64 characters). This is what the AI reads first when scanning available skills.
DescriptionA concise explanation of what the skill covers. This helps the AI decide whether to load the full body.
BodyThe complete content — Markdown with headings, lists, code blocks, tables. This is what the AI reads when it selects this skill.

Writing effective skills

The body field supports full Markdown. Structure your content the same way you would write documentation for a colleague:

markdown
## Refund Policy

### Eligibility
- Order must be within 30 days of delivery
- Product must be unused and in original packaging
- Digital products are non-refundable after activation

### Process
1. Verify the order date in the Payments system
2. Confirm the item condition with the customer
3. Issue the refund through Stripe → Refunds → Create
4. Send confirmation email using the "Refund Processed" template

### Exceptions
- VIP customers (tier Gold+): extend window to 90 days
- Defective products: no time limit, escalate to Quality team

SKILL.md compatible

Skills follow the same format used by Cursor, Windsurf, and other AI-native editors. If you already have SKILL.md files in your repositories, paste them directly into the body field.

Step 2 — Governance

02 — GOVERNANCE
Invisible
armor.
PII is masked before it ever touches the model. Large skill bodies shrink automatically — fewer tokens, same intelligence.
No Security step needed — there is no external API to authenticate.
GOVERNANCE CONTROLS
Zero-Trust PII Redaction
Masks any PII that appears in skill content before it reaches the AI.
Smart Array Truncation
Prevents oversized skill bodies from consuming too many tokens.

Both controls are enabled by default. The Security step is skipped entirely — Agent Skills don't connect to any external API, so there's nothing to authenticate.


DEPLOY
You're live.
AI already knows.

What happens when you deploy

Click Deploy. The platform indexes your skills, builds a progressive disclosure tree, and distributes the server to the global edge.

DEPLOYING...
Indexing skill definitions…done
Building progressive disclosure tree…done
Sealing in V8 Isolate sandbox…done
Live on the edge…done

After deployment, you receive the same outputs as every server type: MCP URL, Config JSON, and a Connection Token (HMAC-SHA256, shown once).

Progressive disclosure in action

When an AI client connects to your skills server, it doesn't download every skill body immediately. Instead:

PROGRESSIVE DISCLOSURE
01AI discovers skill names and descriptions — lightweight metadata only
02User asks a question that matches a skill description
03AI calls the relevant skill tool to load the full body
04AI reasons using the skill content and responds accordingly

This means a server with 20 skills doesn't consume 20× the tokens upfront. The AI loads only what it needs — keeping the context window efficient and responses accurate.


SKILL FILES
Attach more
than just text.

Skill files

On paid plans, you can attach files to individual skills. These files are organized into three directories:

scripts/
Helper scripts
Shell scripts, Python utilities, or any executable the skill references in its instructions.
references/
Reference documents
PDFs, spreadsheets, or data files the AI can cite when answering questions.
assets/
Templates and assets
Email templates, configuration files, sample data, or any resource the skill needs.

Paid plans only

Skill files are available on paid plans. On the free plan, skills are text-only (body Markdown). Save the skill first, then attach files from the detail page.


Real-world skill examples

CUSTOMER SUPPORT
Support agent playbooks
Refund policies, escalation rules, SLA definitions. The AI follows the exact same procedures your human agents follow.
ENGINEERING
Deployment runbooks
Incident response procedures, release checklists, infrastructure patterns. The AI references your specific runbooks — not generic advice.
COMPLIANCE
Regulatory procedures
GDPR data handling rules, SOC 2 controls, HIPAA patient data protocols. The AI enforces compliance without you having to repeat the rules.
ONBOARDING
New hire knowledge base
Architecture overviews, coding standards, team conventions. New team members get answers from the AI trained on your actual documentation.

Next steps


Frequently Asked Questions

What are Agent Skills in Vinkius Cloud?

Agent Skills are structured instructions — procedures, playbooks, FAQs, and decision trees — that AI models discover progressively through MCP. Unlike API-based tools that let AI do things, skills let AI know things — your refund policy, escalation rules, compliance requirements, or any domain knowledge.

Do I need an API or coding knowledge to deploy Agent Skills?

No. Agent Skills require no API backend, no code, and no technical setup. You write the skill content in Markdown (following the SKILL.md open standard), paste it into the editor, and click Deploy. The platform handles hosting, authentication, and governance.

What is the SKILL.md open standard?

SKILL.md is an open standard for packaging AI procedural knowledge, published at agentskills.io. It defines a structured Markdown format supported natively by Claude Code, OpenAI Codex, Gemini CLI, GitHub Copilot, Cursor, VS Code, and Windsurf. Vinkius Cloud deploys these skills as governed, cloud-hosted MCP servers.

How does progressive disclosure work?

The AI first sees only skill names and descriptions — lightweight metadata. When a user question matches a skill description, the AI loads the full body on demand. This keeps the context window clean: a server with 20 skills doesn't consume 20× the tokens upfront.

Can I attach files to skills?

Yes, on paid plans. Each skill supports three file directories: scripts/ (helper scripts and utilities), references/ (PDFs, spreadsheets, data files), and assets/ (email templates, configuration files, sample data).

What's the difference between local SKILL.md files and Vinkius Cloud skills?

Local skills live in a git repository and work for one developer, in one project, in one IDE. Vinkius Cloud deploys the same SKILL.md content to the global edge — accessible from any MCP client, for any team member, with HMAC-authenticated tokens, DLP redaction, and audit logging.