Agents and Skills in Practice

CodeForge improves Claude Code in two related ways:

• Agents handle different kinds of work with different constraints and behaviors.
• Skills inject domain knowledge when a framework, pattern, or workflow is relevant.

How It Feels as a User

You usually do not need to name an agent or manually load a skill. You describe the task, and CodeForge routes it.

Examples:

• Ask for an implementation plan and the architect agent takes the lead.
• Ask to explore the codebase and the explorer agent is used.
• Ask for tests and the test-writer agent handles framework-aware test generation.

What Agents Change

Agents are not just different prompts. They can differ in:

• tool access
• model choice
• permission mode
• worktree isolation
• hook-based verification

This is why a read-only investigation behaves differently from a refactor or test-writing task.

What Skills Change

Skills act like focused reference packs. They give Claude concrete patterns instead of generic guesswork.

Examples:

• fastapi
• testing
• security-checklist
• refactoring-patterns
• ast-grep-patterns

Practical Advice

1. Start by describing the work plainly.
2. Be specific about scope and intent.
3. Let the delegated specialist do the first pass.
4. Name a specific agent only when you need to steer the workflow.

When to Be Explicit

You can still ask for a named specialist:

Use the security-auditor to review this endpoint.
Use the architect to plan this migration.

Learn More

• Agents Reference
• Skills Reference
• Agent System Plugin
• Skill Engine Plugin

Next Steps

• Learn the broader session model in Session Basics
• Start planning work in Spec Workflow