Skills Reference

You don't need to configure, invoke, or manage skills. They are 100% automatic. When an agent works on a task, the relevant skills load silently in the background. There is nothing to install, nothing to enable, nothing to learn. Just use Forge normally — skills do the rest.

---

How Skills Work

Skills are contextual knowledge documents, not executable agents. When an agent starts working, Claude Code matches the task context against skill descriptions and loads the most relevant ones automatically. You never invoke skills directly — they activate when needed.

Example: You ask Builder to implement a REST API. Claude Code loads the architecture skill (system design patterns), the coding-standards skill (naming conventions, error handling), and the core-testing skill (test strategy). Builder's output reflects all three — proper structure, clean code, testable interfaces.

Skills are the difference between "AI-generated code" and "code that looks like your senior engineer wrote it."

---

Why Skills Matter

Without skills, agents start from zero on every request. They might:

• Use inconsistent naming conventions across files
• Write tests that pass but prove nothing (hollow assertions)
• Miss your project's specific patterns and conventions
• Generate code that works but doesn't fit your architecture

With skills loaded, agents know:

• Your project's architecture patterns and boundaries
• Testing standards — coverage targets, assertion quality, integration patterns
• Security best practices — OWASP patterns, secrets handling, auth flows
• How Claude Code, Codex, and Gemini discover context (framework skills)
• Performance optimization strategies for your stack

---

Skills by Category

The plugin ships 27 skills (forge-plugin v3.10.7). Some pages below describe skills that are no longer in the plugin; the list is being reconciled against the release (build B2).

Core

Foundation knowledge loaded across many contexts. These define the baseline standards.

Skill	What It Provides
Core Architecture	Module boundaries, dependency rules, layer separation, component relationships
Core Coding Standards	Code style, naming conventions, error handling patterns, type safety rules
Core NXTG-Forge	Forge-specific conventions — MCP tool usage, governance patterns, plugin architecture
Core Testing	Test strategy, coverage targets, assertion quality standards, test infrastructure

Domain Knowledge

Deep expertise in specific engineering domains.

Skill	What It Provides
Architecture	System design patterns, data flow, component relationships, trade-off analysis
Coding Standards	Language-specific conventions, formatting rules, documentation standards
Documentation	Doc structure, API documentation patterns, README conventions, JSDoc standards
Security	Vulnerability categories, remediation approaches, auth patterns, secrets management
Testing Strategy	Test pyramid, integration patterns, mock guidelines, coverage strategies
Testing	Test framework usage, assertion patterns, coverage tools, fixture management

Framework Knowledge

How AI coding tools discover context — critical for multi-tool workflows.

Skill	What It Provides
Claude Code Framework	Claude Code architecture, tool usage, agent patterns, extension points
Claude Code Best Practices	Effective Claude Code usage, prompt patterns, workflow optimization
Codex Framework	Codex CLI conventions — AGENTS.md, task mode, skill discovery (reference only)
Gemini Framework	Gemini CLI conventions — GEMINI.md, context loading, modular imports (reference only)

Workflow

Development process patterns — from environment setup to deployment.

Skill	What It Provides
Dev Environment Patterns	Environment setup, WSL2 patterns, toolchain configuration, CI/CD integration
Git Workflow	Branching strategy, commit conventions, PR review practices, release patterns
Runtime Validation	Input validation, error boundaries, data integrity, type safety at system boundaries
Parallel Execution	Plan Mode + Agent Teams — Claude Code's highest-leverage superpowers

Performance & Debugging

Optimization and troubleshooting knowledge.

Skill	What It Provides
Optimization	Performance profiling, bundle optimization, caching strategies, hot path analysis
Browser Debugging	Playwright MCP for console errors, screenshots, network monitoring, UI debugging

Knowledge & Learning

Meta-skills about the system itself.

Skill	What It Provides
Skill Development	How to create new skills — SKILL.md format, progressive disclosure, triggers
Domain Knowledge	Project-specific domain concepts, business rules, terminology
Verify Governance	Governance verification patterns, health score interpretation, compliance checking
CEO Loop	ORBIT protocol for autonomous governance — iteration model, decision tracking
Crucible Audit	Forensic test quality methodology — the 8 fraud patterns, detection techniques

Agent Roles

Expertise modules that inform specific agent personalities.

Skill	What It Provides
Agent Lead Architect	System design leadership, architecture decisions, trade-off analysis frameworks
Agent Backend Master	Backend patterns — API design, data layer, server architecture, caching
Agent CLI Artisan	CLI tool design — argument parsing, output formatting, terminal UX
Agent Platform Builder	Platform engineering — infrastructure, deployment, scaling, observability
Agent QA Sentinel	QA leadership — test coverage analysis, bug detection, quality gates
Agent Development	Agent creation guide — frontmatter, system prompts, tool selection, examples
Agent Integration Specialist	Service integration — API connections, webhooks, auth flows, data sync

---

How Skills Load

You type a prompt
  │
  Claude Code analyzes context
  │
  ├── Matches task against skill descriptions
  │     └── "Implement a REST API" → loads: architecture, coding-standards, core-testing
  │
  ├── Matches project type
  │     └── TypeScript project → loads: claude-code-framework, dev-environment-patterns
  │
  └── Matches agent invocation
        └── Builder agent → loads: parallel-execution (knows about Plan Mode)

Skills don't conflict — they layer. When multiple skills activate, their knowledge combines to give agents a comprehensive understanding of what "good" looks like for your specific task.

---

Creating Custom Skills

Skills are SKILL.md files in directories under your plugin's skills/ folder. See Skill Development for the complete guide.

---
name: My Custom Skill
description: When this skill is relevant — triggers automatic loading
---

# Skill Content

Your knowledge, patterns, best practices...

The description field is the trigger — Claude Code matches it against task context to decide when to load the skill. Write descriptions that capture the situations where this knowledge helps, not just the topic.