Skip to main content

Skills & Plugins

Skills and plugins extend coding agent capabilities with custom commands and reusable functionality. This guide covers what they are, when to create them, and lessons from building a skills ecosystem.

What Are Skills?​

Skills are reusable prompts or workflows triggered by commands. Instead of typing the same complex instructions repeatedly, you define them once and invoke them with a short command.

/deploy-staging    → Runs your deployment workflow
/review-pr 123     → Reviews a specific pull request
/techdebt          → Finds code quality issues

The One-Day Rule​

If you do something more than once a day, make it a skill.

Common candidates:

  • Deployment workflows
  • Code review patterns
  • Version management
  • Log monitoring
  • Documentation generation
  • Test execution patterns

When to Create Skills​

Repetitive Workflows​

Any multi-step process you run frequently:

  • Deploying to different environments
  • Running specific test suites
  • Generating boilerplate code
  • Checking system health

Team Standardization​

Ensure everyone follows the same process:

  • PR review checklists
  • Release procedures
  • Code quality checks
  • Security audits

Complex Operations​

Tasks that require specific steps or context:

  • Database migrations with validation
  • Multi-service deployments
  • Performance profiling workflows

What Are Plugins?​

Plugins are a packaging mechanism that bundles multiple agent extensions together. They enable sharing and distribution of skills, hooks, MCP servers, and other configurations.

What Plugins Can Include​

ComponentDescription
SkillsCustom commands and prompts
HooksLifecycle scripts (pre/post actions)
MCP ServersExternal tool connections
ConfigurationDefault settings and preferences

Plugin Namespacing​

To avoid conflicts between plugins, components are typically namespaced:

/commit                   → Built-in skill
/team-tools:deploy        → Skill from "team-tools" plugin
/security:audit           → Skill from "security" plugin

Our Experience: Building a Skills Marketplace​

The PwC AI-CoE team built an internal skills marketplace for our coding agent workflows. Here's what we learned.

What We Built​

Skills for common operations:

  • Azure DevOps Deployments - Build and release through Azure pipelines
  • GCP Log Monitoring - Check Cloud Run and GKE logs across environments
  • Version Management - Semantic versioning and release branch creation
  • Code Quality - Linting, type checking, and test execution

Key Learnings​

Keep skills focused on single workflows

A skill should do one thing well. Instead of a "deploy-everything" skill, create:

  • /deploy-dev - Deploy to development
  • /deploy-staging - Deploy to staging
  • /deploy-prod - Deploy to production (with extra confirmations)

Include clear instructions

Skills should be self-documenting. Include:

  • What the skill does
  • Required prerequisites
  • Expected inputs/outputs
  • Error handling guidance

Document prerequisites

Before a skill can run, the user may need:

  • Environment variables set
  • CLI tools installed
  • Authentication configured
  • Permissions granted

Make these requirements explicit in the skill documentation.

Version your skills

As workflows evolve, skills need updates. Treat skills like code:

  • Keep them in version control
  • Review changes before deploying
  • Test after updates

Team Adoption​

What made skills successful for our team:

  1. Low friction - Skills are faster than remembering complex commands
  2. Consistency - Everyone follows the same process
  3. Discoverability - A central place to find available workflows
  4. Iteration - Easy to improve skills based on usage

Skill Best Practices​

Clear, Action-Oriented Names​

  • /deploy-staging not /staging
  • /run-e2e-tests not /tests
  • /check-security not /security

Self-Contained Prompts​

The skill prompt should include everything needed:

  • What to do
  • How to do it
  • What to check for
  • How to report results

Handle Failure Gracefully​

Skills should:

  • Report clear error messages
  • Suggest next steps on failure
  • Not leave the system in a broken state

Keep Prompts Focused​

One skill, one purpose. Complex workflows should chain multiple skills rather than creating one massive skill.

Getting Started​

Check your coding agent's documentation for skill syntax and configuration:

  • Claude Code: Skills Documentation
  • Other agents: Check the agent's plugin or extension documentation

Most coding agents support some form of custom commands or prompt templates. The concepts above apply regardless of the specific implementation.