Router

View as Markdown
Intermediate2 min readLast updated August 20, 2026

Use this guide when a skill isn't being selected the way you expect, or when you want to understand how the router matches a request to a skill. It covers where the routing table lives for each tool, the match and load sequence, and how to disambiguate or name a skill directly when more than one could fit.

Where it lives

FormatRouter file
Source of truthskills/CLAUDE.md
Cursorcursor/rules/00-router.mdc (alwaysApply: true)
Codexcodex/AGENTS.md
Claude Code / Geminiloaded from the source via the plugin/extension

The system generates all forms from skills/CLAUDE.md.

Additional Resource Refer to the Architecture documentation for details.

How routing works

  1. You make a request in natural language.
  2. The agent reads the routing table (always loaded for full-bundle installs).
  3. It matches your intent to the best-fitting row.
  4. It loads that skill's SKILL.md and follows it, including reading any references/ the task needs.

You don't have to name a skill. You can specify one directly (for example, "use the Data Modeling skill…").

The routing table

Each row maps the user's intent to a specific skill. Abbreviated:

When the user asks about…Skill
Brand Kit, Voice Profiles, Knowledge Vault, on-brand AI generationBrand Kit Assistant
Migrating/porting from Contentful to ContentstackMigration Companion
Migrating Delivery SDK code from JavaScript to TypeScriptMigrate JS to TS SDK
Delivery SDK code, queries, Live Preview setup, SSR previewDelivery SDK
Designing or refactoring content modelsData Modeling Best Practices
Debugging Live Preview or Visual BuilderLive Preview & Visual Builder Support
Fetching entries, CDA queries, pagination, bulk opsEntries
Uploading, transforming, delivering assetsAssets
Classifying content, category hierarchiesTaxonomy
Workflow stages, approvals, publish rulesWorkflows & Publish Rules
Environments, publishing, delivery/preview tokens, Sync APIEnvironments & Publishing
Languages, fallback chains, localizationLocalization
Branches, aliases, CI/CD, deployment strategyBranches & Aliases
Roles, permissions, teams, token capabilitiesRoles & Permissions
Deploying multiple content changes together, campaignsReleases
Authentication, token types, API keys, rate limits, SSOTokens & Authentication
Webhooks, event channels, payloads, signaturesWebhooks
Matching a Launch env to .env.exampleSync Launch env vars
Triggering and monitoring Launch deploymentsTrigger and Monitor Deployments
Personalization, A/B testing, audience segmentation, variantsVariants & Personalization
Building a Developer Hub or Marketplace appDeveloper Hub App Architect

The full, authoritative table is in skills/CLAUDE.md and the skills reference.

When multiple skills fit

Many real tasks span skills.

For example, "deploy a campaign of 50 entries without my site rebuilding hundreds of times" touches the following skills:

  • Releases: Deploys all changes as a single atomic operation
  • Webhooks: Prevents the flood of rebuilds those webhook events trigger.

The agent routes to the primary skill and reads cross-referenced skills as needed. Skills include explicit cross-references for this purpose.

Disambiguating

Some routes prompt for clarification first. Localization, for instance, behaves differently for the editorial UI versus CDA delivery, so the skill clarifies which you mean before answering. This is by design.

Additional Resource Refer to the Skill Anatomy documentation for details.