Using Skills

Skills are reusable knowledge packages that provide domain expertise, procedures, and reference material to agents. Unlike commands (which you invoke), skills are loaded automatically based on context.

How Skills Load

User types "CDK test" in chat
       |
MANIFEST.json matches trigger pattern "CDK"
       |
Loads: cdk-patterns skill + testing-orchestration-reference skill
       |
Agent receives domain knowledge for CDK testing

Skills load via three mechanisms:

Mechanism	Example	When
Trigger patterns	User says "Terraform" → loads terraform-patterns	Automatic
Agent reference	qa-engineer references testing-skill-framework	Agent invocation
Explicit @import	@.claude/skills/testing/autonomous-testing.md	Manual

Core Skills (20)

Always available across all projects.

Testing (6 skills)

Skill	Purpose
testing-skill-framework	Master testing methodology
testing-orchestration-reference	3-tier test orchestration
autonomous-testing	Self-directed test execution
visual-verification	Screenshot-based validation
docusaurus-testing	Documentation site testing
cross-validation-docs	Docs vs source accuracy

Development (6 skills)

Skill	Purpose
cdk-patterns	CDK construct patterns
terraform-patterns	Terraform module patterns
git-commit-helper	Commit message standards
skill-creator	Create new skills
mcp-builder	Build MCP integrations
artifacts-builder	Build deployable artifacts

Dashboards (5 skills)

Skill	Purpose
webapp-testing	Web app validation
dashboard-testing	Dashboard data validation
cross-validation-mcp-api	MCP vs API accuracy
marketplace-testing	Component catalog testing
design-token-validation	Design token consistency

Other (3 skills)

Skill	Purpose
invest-quality-gates	User story quality scoring
reflective-learning	Kolb Cycle + session reflection
operational-excellence	PDCA enforcement

Marketplace Skills (75)

Opt-in per-project. Browse at Component Marketplace.

Key categories:

• Product Management (65 skills) — PRDs, OKRs, personas, roadmaps, GTM
• Security (4 skills) — threat modeling, compliance
• Document Processing (3 skills) — PDF extraction, analysis
• Governance (1 skill) — regulatory compliance

Enable Marketplace Skills

// settings.local.json
{
  "skills": {
    "marketplace": ["product-management/*", "security/*"]
  }
}

Skill Anatomy

Every skill follows a 13-field frontmatter standard:

---
name: skill-name
description: "What this skill provides"
agent: [agent-that-uses-it]
context: fork          # isolated context
model: sonnet          # or opus
allowed-tools: [Read, Bash, Grep, Glob]
license: Apache-2.0
metadata:
  version: 1.0.0
  adlc_version: 3.7.1
visibility: public
track: public          # public | startup | enterprise
industries: []         # FSI, Energy, Telecom, Aviation
disclosure-level: full # full | partial | minimal
compatibility: "Claude Code 1.0+, ADLC 3.2+"
---

Tips

• Skills load in a forked context — they don't bloat the main conversation
• Use /speckit.improve to audit which skills consume the most tokens
• The reflective-learning skill is mandatory for retrospectives
• PM skills require explicit opt-in (65 skills = significant token budget)