oracle-ai-developer-hub

Workshop TODO Checklist

9 hands-on coding TODOs across 11 parts. Every “true setup” task (Oracle DDL, seed data, ONNX models, DDS policies, scheduler job) was run by the Codespace before you opened the notebook — you don’t see it in the workshop notebook unless you open notebook_complete_with_setup_code.ipynb (the full-source version for advanced learners).

Each TODO has a hard-stop assert checkpoint right below it. If you skip a TODO or get it wrong, the next cell raises an AssertionError so you can’t accidentally barrel forward with broken state.

Part 1 — Setup (Guide)

No TODO. Just run the imports + agent_conn + chat client cells.

Part 2 — Long-Term Memory with OAMP (Guide)

  1. Implement _scan_tables — mine ALL_TABLES + ALL_TAB_COMMENTS into Facts. TODO 1

Part 3 — Retrieval (Guide)

  1. Implement retrieve_knowledge — cosine search + rerank. TODO 2
  2. Implement hybrid_rrf_search_memories — vector + keyword fused via Reciprocal Rank Fusion in one SQL. TODO 3

(The three-way RRF probe is a demo cell — run it and observe the r_vec / r_txt ranks.)

Part 4 — DBFS Scratchpad (Guide)

No TODO. Read the DBFS Python wrapper, smoke-test with scratch.write / scratch.read.

Part 5 — Oracle MLE Compute (Guide)

No TODO. Read the exec_js helper, run the percentile smoke test.

Part 6 — Tools & Skills (Guide)

  1. Register tool_run_sql with the @register decorator. TODO 4

Part 7 — The Agent Loop (Guide)

  1. Implement agent_turn — the dispatch loop. TODO 5

(The three-turn end-to-end demo runs after the assert passes.)

Part 8 — Identity-Aware Authorization with DDS (Guide)

  1. Implement set_identity — bridge end_user / clearance into the EDA_CTX namespace via the trusted AGENT.set_eda_ctx procedure. TODO 6

(The DDS-aware agent_turn wrapper and the two-identity comparison run after the assert.)

Part 9 — JSON Relational Duality Views (Guide)

  1. Register tool_get_document — read a full document by primary key. TODO 7

Part 10 — Continuous Scans via DBMS_SCHEDULER (Guide)

No TODO. The scheduler job + AGENT_REQUEST_SCAN procedure are pre-installed. The notebook defines drain_queued_scans() and shows how to consume the queue.

Part 11 — Tool-Output Offload (Guide)

  1. Implement log_tool — persist the full tool output as an OAMP memory keyed by tool_call_id. TODO 8
  2. Register tool_fetch_tool_output — recover full bytes by tool_call_id. TODO 9