Skip to content

How Agent Lexicon and SkeinRank work together

Agent Lexicon and SkeinRank are separate products with separate deployment boundaries. They can be used independently. Together they describe one terminology governance lifecycle.

1Discover

Agent Lexicon scans repository code and documentation for candidate project language.

2Review

Humans accept canonical terms, aliases, deprecations, scopes, and tool rules.

3Publish

A git-tracked lexicon snapshot becomes the repository vocabulary used by agents and CI.

4Promote

Approved organization-level terminology is mapped into a SkeinRank profile through an explicit review or import workflow.

5Bind

SkeinRank binds a reviewed snapshot to a search index, application context, or runtime workflow.

6Observe

Evidence and drift findings feed the next vocabulary review instead of silently mutating production language.

LayerAgent LexiconSkeinRank
Primary scopeRepository, branch, diff, tool callOrganization, profile, index, runtime binding
Main userDeveloper or coding agentSearch/RAG/platform team
StateGit-tracked lexicon + local review workspaceProfiles, bindings, evidence, snapshots
Decision pathResolve, guard, lint, check mergeCanonicalize, enrich, bind, publish, roll back
Typical integrationCLI, CI, local MCPSDK, REST API, governance console, Elasticsearch

This page describes the operating model, not an automatic synchronization service. Today:

  • each product can run without the other;
  • vocabulary handoff should remain explicit and reviewable;
  • teams decide which repository terms belong in an organization-level runtime profile;
  • the source lexicon and the runtime snapshot retain their own version and provenance.

That boundary is intentional. It prevents a newly discovered identifier in one repository from silently becoming production vocabulary everywhere.

  1. Start with Agent Lexicon in one repository.
  2. Review whether naming drift is a measurable cost in pull requests or agent sessions.
  3. Promote stable, cross-team concepts into a SkeinRank profile.
  4. Bind that profile to one search or RAG context.
  5. Expand only after the review and rollback workflow is understood.