Skip to content

skills.md

Latest commit

 

History

History
122 lines (83 loc) · 4.79 KB

skills.md

File metadata and controls

122 lines (83 loc) · 4.79 KB

Skills System

Skills are pluggable capability modules for AI workers in LinkWork. Each Skill is an independent, declaratively defined capability unit that can be composed and installed across different roles as needed.

Design Philosophy

Traditional Agent capabilities have blurry boundaries — what plugins are installed and what capabilities exist are only known at runtime, and the AI can install and modify them on its own.

LinkWork takes a different approach:

  • Skills are declaratively defined — each Skill has a clear name, description, version, and dependencies
  • Skills are pinned at build time — Skills are injected into the image during role building, read-only at runtime
  • Skills are admin-governed — which roles can use which Skills is decided by administrators, not by the AI

Install capabilities for AI workers like installing apps on a phone. Admins are the app store reviewers; AI workers are the users.

Three-tier Capability Architecture

Role
  └── Skills (Capability Modules)
        └── MCP Tools (External Tool Access)

Roles define "who the AI worker is", Skills define "what the AI worker can do", MCP Tools define "what tools the AI worker can use". Three decoupled layers, freely composable, access-controlled.

Skill Composition

A Skill consists of the following core elements:

Element Description
Name & Description Skill identifier and functionality description
Version Number Semantic versioning (SemVer), supports rollback
Knowledge Content Markdown knowledge files injected into the AI's context
Dependency Declaration Other Skills or tools this Skill depends on
Scope Tags and categories for marketplace discovery

Skill Knowledge Injection

The core value of Skills lies in knowledge injection. Each Skill is essentially a set of structured knowledge files (Markdown) that are automatically loaded into the AI's context during task execution:

skills/
├── code-review/
│   ├── SKILL.md           # Skill definition & metadata
│   ├── guidelines.md      # Code review guidelines
│   └── checklist.md       # Review checklist
└── data-analysis/
    ├── SKILL.md
    ├── methodology.md     # Analysis methodology
    └── templates.md       # Report templates

AI workers automatically receive the knowledge from all Skills configured for their role during task execution — no additional setup required.

Skills Marketplace

LinkWork provides a built-in Skills marketplace supporting team sharing and reuse:

Marketplace Features

Feature Description
Publish Publish personally validated Skills to the team marketplace
Discover Search available Skills by category, tags, and popularity
Version Management Independent version management per Skill, supports rollback
Usage Analytics View call count and success rate for each Skill
Team Sharing Set Skill visibility scope (personal / team / global)

Lifecycle

Write Skill → Personal validation → Publish to marketplace → Admin review → Team available → Continuous iteration

Skills you've refined on your personal tools can go straight into LinkWork, becoming standardized capabilities your entire team can use.

Build-time Pinning

Skills are injected into the container image during the role build phase:

  1. Pull the corresponding version of Skills per role configuration
  2. Pin the exact version number, written into the image
  3. Read-only at runtime, cannot be modified

This means:

  • All instances of the same role have identical Skills configurations
  • Skills changes require rebuilding the image — no "some instances updated, some didn't" scenarios
  • Build records are fully traceable, Skills configuration can be reproduced at any point in time

See Harness Engineering for details.

Common Skills Examples

Skill Description Applicable Roles
Code Review Code quality, security vulnerability, performance issue review Development Engineer
Data Analysis Data cleaning, statistical analysis, visualization reports Data Analyst
Document Writing Technical docs, API docs, user manuals Technical Writer
Literature Tracking Paper fetching, summary extraction, trend analysis Research Assistant
Ops Inspection Service health checks, log analysis, alert handling Operations Engineer

Further Reading