Choosing and Combining Skills

The skills catalog offers many options. This guide explains how to decide which skills to enable and how to combine them effectively.

See also: For the full skill list see Skills Catalog. If a skill isn't triggering, see Debug Skill Routing.

1. Filter by phase first

Every skill belongs to upstream, midstream, or downstream. Identifying the phase of your change narrows candidates significantly.

• Design docs / ADR changes → choose upstream skills
• Application code changes → choose midstream skills
• Test / QA config changes → choose downstream skills

2. Detect overlap using tags

Skills with the same tags may cover overlapping concerns. Check tags before combining.

Tag combination	Overlap risk
typescript / type-safety	typescript-nullcheck and typescript-strict share identical applyTo globs — enabling both produces duplicate findings on the same files
community / modern-web	The four skills modern-web-semantic, modern-web-performance, modern-web-browser-compat, and modern-web-a11y-interactive share the same globs — start with one or two that match your goal
community / design-system	design-token-enforcement (hardcoded values) and design-system-component-reuse (reimplemented components) cover different axes — enabling both has low overlap

3. Prioritize by severity

When multiple skills match the same file, prefer enabling higher-severity skills first.

• critical / major — can block merges; enable these first
• minor — improvement suggestions; add based on team capacity
• info — policy confirmation only; safe to enable at all times

4. Recommended starter sets by stack

TypeScript frontend (Next.js / React)

• typescript-strict — type-safety baseline
• security-basic — XSS and secret leaks
• nextjs-app-router-boundary (if using Next.js App Router)
• modern-web-a11y-interactive — interactive UI accessibility
• design-token-enforcement (if a design system is in use)

Python API

• security-basic
• logging-observability
• coverage-gap

Design / documentation focus

• adr-decision-quality
• architecture-boundaries
• security-privacy-design

Multi-agent / AI review integration

• Run independent-review-synthesis last to consolidate multiple review results.

5. Choosing between TypeScript nullcheck and strict

typescript-nullcheck and typescript-strict target the same globs.

• strict-001 covers any types, unsafe assertions, and null handling broadly.
• nullcheck-001 specializes in null/undefined safety with deeper checks.

For projects without strictNullChecks enabled, start with nullcheck-001. Once strict mode is in place, migrate to strict-001 or enable both.