Tutorial 201 — Building with LCDL

Developers who finished 101 and need to design governed tasks, wire JSON schemas, compose operators, and follow RAG / verification patterns—still offline-first.

Updated

Audience

Developers who finished 101 and need to design governed tasks, wire JSON schemas, compose operators, and follow RAG / verification patterns—still offline-first.

Tutorial 201 building blocks

How authored contracts become registered, offline-runnable tasks verified with operators and pytest.

  1. contract.md + contract.jsonDefines task intent and machine-readable input/output schemas.
  2. TASK_REGISTRY_V1 registrationRegisters the task_id so TaskRunner can resolve the contract.
  3. TaskRunner + fake_chatExecutes tasks offline with a deterministic fake chat adapter.
  4. operators → evidence tasks → pytestComposes retries and evidence paths, then proves behavior with pytest.

Prerequisites

Track Requirement
Lessons 201-01–03 Tutorial 101 complete
201-04–05 Comfortable reading contract.json and running **pytest

Final artifact

You can:

  • Sketch a new task_id folder layout and pass audit_contract_docs.py --strict
  • Run fallback_chain / until_ok recipes without LLM calls
  • Execute answer_from_evidence with fake chat + stub evidence
  • Map a failing Err to verification vs repair actions

Lesson table

Step Page Focus
1 Author a task contract Layout + registry mental model
2 Add a ContractSpec sidecar contract.json + audits
3 Use operators fallback_chain, until_ok, try_catch
4 RAG and evidence Fake evidence + answer_from_evidence
5 Verification and repair Taxonomy + hooks
6 SharePoint library reads Canonical sharepoint_* tasks + strategy guide

Supporting reference

ContractSpec · Schemas · Operators · RAG · Verification · Repair loops

Next track

Tutorial 301 — architecture, gateways, MCP/Cursor, CI, production readiness.