# How Domain Expertise Is Encoded in Skills within Knowledge Work Plugins

> Learn how domain expertise is encoded in skills within anthropics knowledge work plugins. Discover self-contained markdown documents combining YAML metadata, taxonomies, and references.

- Repository: [Anthropic/knowledge-work-plugins](https://github.com/anthropics/knowledge-work-plugins)
- Tags: deep-dive
- Published: 2026-06-01

---

**Domain expertise is encoded in skills as self-contained markdown documents that combine YAML front-matter metadata, embedded domain taxonomies, and linked reference materials to create a declarative, version-controlled knowledge base.**

The `anthropics/knowledge-work-plugins` repository implements a unique approach where domain expertise lives directly in the file system rather than hidden in application code. Each **skill** is a markdown document located at `*/skills/*/SKILL.md` that captures specialized knowledge through a three-layer architecture, making complex business logic searchable, maintainable, and portable across different operational domains.

## The Three-Layer Architecture of Domain Expertise

### Layer 1: YAML Front-Matter and Compatibility Declarations

In [`small-business/skills/ticket-deflector/SKILL.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/small-business/skills/ticket-deflector/SKILL.md), the encoding of domain expertise begins with YAML front-matter. The `compatibility` field explicitly lists domain-specific integrations, such as "Requires PayPal, HubSpot, Mail. Optional: Intercom, Square." This metadata tells the orchestrator exactly which external services and APIs belong to the skill's operational domain.

Additional front-matter fields—including `name`, `description`, and `argument-hint`—define the skill's identity and purpose, while the compatibility declaration functions as a machine-readable domain dependency manifest.

### Layer 2: Embedded Taxonomies and Decision Logic

The markdown body contains executable domain logic rendered as human-readable structures. In [`customer-support/skills/ticket-triage/SKILL.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/customer-support/skills/ticket-triage/SKILL.md), the file embeds a complete **Category Taxonomy** and **Priority Framework** as declarative tables. These structures capture nuanced classification logic—such as routing rules and priority matrices—directly in markdown rather than programming code.

This approach transforms support-domain expertise into explicit decision tables that the runtime can interpret and execute during skill invocation.

### Layer 3: Linked Reference Materials and Examples

Domain expertise extends beyond the main skill file through relative links to supplemental content. Skills reference markdown files in `reference/` and `examples/` directories that house best-practice guides, "gotchas," and external API documentation. For instance, the ticket-deflector skill links to [`reference/gotchas.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/reference/gotchas.md), which contains PayPal lookup patterns and tone guidelines essential for small-business customer interactions.

These links create a federated web of domain-specific knowledge that the orchestrator surfaces contextually during execution.

## How Connector Hints Bind Skills to External APIs

A critical encoding mechanism appears as a short note within the skill document: `> If you see unfamiliar placeholders … see CONNECTORS.md`. This points to [`CONNECTORS.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/CONNECTORS.md) at the repository root, which catalogs concrete APIs including Zoom, PayPal, and HubSpot. By referencing this shared connector catalog, skills make their domain dependencies explicit and reusable across the entire plugin ecosystem, ensuring that required external systems are discoverable and version-controlled.

## Practical Skill Invocation Examples

When invoking a skill, the system leverages all three layers of encoded expertise to produce domain-accurate output:

```bash

# Ticket-deflector pulls order info from PayPal and drafts replies

/skill ticket-deflector <<EOF
Forwarded email from john@example.com asking about a refund for order #12345
EOF

```

The skill reads its front-matter `compatibility` list, contacts PayPal and HubSpot, applies refund-approval gates, and returns a draft reply with inline notes about any missing data.

```bash

# Ticket-triage categorizes using embedded taxonomies

/ticket-triage "User cannot connect their SSO – receiving a 403 on the callback URL"

```

This invocation triggers the **Category Taxonomy** and **Priority Framework** defined in [`customer-support/skills/ticket-triage/SKILL.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/customer-support/skills/ticket-triage/SKILL.md), returning structured JSON with `category`, `priority`, and routing recommendations.

```bash

# Data-visualization validates against declared compatibility

/create-viz <<EOF
{
  "data": [{"date":"2024-01-01","sales":120},{"date":"2024-01-02","sales":150}],
  "type": "line",
  "title": "Daily Sales"
}
EOF

```

The skill checks its front-matter for Python plotting library requirements, validates the input schema, and returns a rendered chart URL.

## Key Implementation Files

- **[`small-business/skills/ticket-deflector/SKILL.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/small-business/skills/ticket-deflector/SKILL.md)**: Demonstrates how front-matter encodes required services and how domain-specific workflow steps are written in declarative markdown.

- **[`customer-support/skills/ticket-triage/SKILL.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/customer-support/skills/ticket-triage/SKILL.md)**: Shows embedded taxonomy, priority logic, and routing tables that capture support-domain expertise.

- **[`CONNECTORS.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/CONNECTORS.md)**: Serves as the central catalog of external APIs and services referenced by the `compatibility` fields across all skills.

- **[`design/skills/ux-copy/SKILL.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/design/skills/ux-copy/SKILL.md)**: Illustrates non-technical domain expertise encoded in style guides and tone examples.

- **[`partner-built/zoom-plugin/skills/zoom-mcp/SKILL.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/partner-built/zoom-plugin/skills/zoom-mcp/SKILL.md)**: Captures partner-product expertise via required "domain allowlist" and SDK configuration metadata.

## Summary

- Domain expertise in `anthropics/knowledge-work-plugins` lives in self-contained markdown files under `*/skills/*/SKILL.md`, not buried in application code.
- **YAML front-matter** encodes domain identity and external service dependencies through the `compatibility` field, declaring requirements like PayPal or HubSpot integrations.
- **Embedded taxonomies and decision tables** capture nuanced business logic—such as priority frameworks and routing matrices—as declarative markdown structures.
- **Reference links and connector hints** extend expertise to external documentation and the central [`CONNECTORS.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/CONNECTORS.md) API catalog.
- This architecture creates a structured, searchable, and version-controlled representation of specialized knowledge that orchestrators can validate and execute at runtime.

## Frequently Asked Questions

### How does the compatibility field in YAML front-matter encode domain requirements?

The `compatibility` field explicitly lists required and optional external services—such as "Requires PayPal, HubSpot, Mail. Optional: Intercom, Square"—allowing the orchestrator to verify domain-specific integrations before executing the skill. This declaration acts as a domain dependency manifest within the skill's metadata block, ensuring the runtime only invokes skills when their required external systems are available.

### Where are complex business rules like priority matrices stored in a skill?

Complex business rules live directly in the markdown body of the skill file rather than in separate code files. For example, [`customer-support/skills/ticket-triage/SKILL.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/customer-support/skills/ticket-triage/SKILL.md) contains full **Category Taxonomy** and **Priority Framework** tables that encode support-domain classification logic as declarative, human-readable structures that both developers and domain experts can maintain.

### What is the role of CONNECTORS.md in encoding domain expertise?

[`CONNECTORS.md`](https://github.com/anthropics/knowledge-work-plugins/blob/main/CONNECTORS.md) serves as a central catalog of all external APIs and services that skills can reference. Skills include hints pointing to this file, making their domain dependencies explicit and ensuring that required integrations—such as Zoom, PayPal, or HubSpot—are discoverable and reusable across the entire plugin ecosystem without duplicating configuration data.

### How does this architecture differ from traditional code-based domain logic?

Unlike traditional approaches where expertise is buried in source code, this pattern stores domain knowledge in version-controlled markdown with structured front-matter, making it readable by both humans and AI systems. The `*/skills/*/SKILL.md` pattern allows non-technical domain experts to modify taxonomies, compatibility lists, and reference materials without modifying application code or redeploying services.