skeem

Relational-aware CLI for headless backends (Directus, PocketBase, NocoDB, Strapi) and public reference datasets (WikiData, MusicBrainz, OpenStreetMap, Open Library, CrossRef, TMDB), built for AI agents. Agents should look facts up, not make them up.

Try it in ten seconds — no install, no backend, no API key:

npx skeem get items Q42 --adapter wikidata --url https://www.wikidata.org --json

Every command returns a stable JSON envelope (ok / error.code), writes are schema-aware with a compact relation grammar (@id, ?field=value, ??field=value), coordination primitives (claims, idempotency, provenance, soft delete) make multi-agent writes safe, and --enrich folds a reference record into a create/upsert so your agent never types a release date from memory:

skeem create movies --enrich tmdb:movies:603 --map externalIds.imdb=imdb_id --json

Full docs, the agent playbook (SKILL.md), and the design notes live in the GitHub repo.

This is the core package: the CLI entrypoint and backend-agnostic runtime.

What Lives Here

• CLI argument parsing
• output formatting
• config loading
• schema cache
• relation-aware data operations
• compound exec
• schema discover, diff, and define
• system-feature behaviors for aliases, provenance, versions, trash, claims, annotations, and idempotency

Useful Paths

• bin/skeem.ts
• src/cli/index.ts
• src/core/runtime.ts
• src/config/load-config.ts

Common Commands

npm run build --workspace skeem
npm run test --workspace skeem
npm run typecheck --workspace skeem

The root-level docs live at:

• README.md
• docs/README.md