Runtime boundary (`forge-lcdl` vs `forge-lcdl-runtime`)

Forge LCDL (this repo) owns:

Updated

Runtime boundary (forge-lcdl vs forge-lcdl-runtime)

Forge LCDL (this repo) owns:

  • Governed tasks, contracts, JSON Schema sidecars, and TASK_REGISTRY_V1
  • TaskRunner / run_task, result envelopes (Ok / Err), operators, verification primitives
  • OpenAI-compatible helpers, MCP client utilities documented here
  • Offline-first examples and deterministic game/page mechanics surfaces

forge-lcdl-runtime (optional sibling) is the place for disk-backed chat sessions, richer RAG-lite helpers, decision tables/trees, matrix utilities, and Playwright/MCP bridge stubs that pair with host applications.

Consumers own production credentials, fleet/Docker topology, data retention, SLAs, browser policies, and UX.

For SharePoint library reads, the consumer registers the live Playwright Page (or session ref); LCDL performs read-only inventory and REST download atoms. Orchestration (sync jobs, SQLite, Copilot) stays in the host — see SharePoint consumer boundary.

This handbook does not require cloning the runtime repo—keep examples here PYTHONPATH=src friendly without cross-repo imports.

Core runtime consumer lanes

How forge-lcdl core, optional forge-lcdl-runtime, and consumer repos divide governed tasks from operations.

  1. forge-lcdl core ─ contracts runners operators genericsVersioned contracts, task dispatch, operators, and generic helpers in this repo.
  2. forge-lcdl-runtime ─ disk sessions RAG-lite matrices MCP stubs optionalOptional sibling package for disk-backed sessions and richer runtime helpers.
  3. Consumer repos ─ secrets browsers fleets UX SLAsApplication layer owning credentials, deployment, browsers, and user experience.

See also