Module DD-11 — OpenAI Agents SDK: 2-Layer Harness/Compute Split

OpenAI Agents SDK: 2-Layer Harness/Compute Split

The most architecturally thoughtful SDK. 2-layer harness/compute split. 7 sandbox providers. Handoffs + agents-as-tools. The SDK reference. Module 5 (Sandboxing): 5/5 — the highest-scoring sandbox in the roster.

60
minutes
8
artifacts
3
sub-sections
The OpenAI Agents SDK is not a finished harness — it is the primitives for building one. Its value is the architectural decisions it bakes in as defaults so the harness you build on top starts from a strong foundation rather than a blank file. Three contributions define it: the 2-layer harness/compute split (Module 5), the 7-provider sandbox abstraction (Module 5, 5/5 — the highest in the roster), and handoffs + agents-as-tools as first-class primitives (Module 1.3). Credential isolation is architectural, not optional — the same principle as NemoClaw (DD-09), applied to a different trust boundary.
Key Claims
Load-Bearing Claims

The SDK scores 5/5 on Module 5 (Sandboxing) — the highest in the roster — because of the 7-provider abstraction. One tool, written once, targets 7 isolation regimes (local bash, Docker, E2B, Modal, Daytona, Cloudflare, Vercel) through a uniform interface. Swap by configuration, not code change. The harness code is provider-agnostic. Compare to OpenCode (DD-03): Docker-or-nothing. This is the most architecturally thoughtful sandboxing in the roster.

The 2-layer harness/compute split makes credential isolation architectural, not optional. The harness layer (loop, model calls, credentials) runs on your backend; the compute layer (tool execution) runs in a sandbox. There is no code path in the SDK for credentials to enter the sandbox — you cannot accidentally put creds in the sandbox because the architecture prevents it. A discipline you maintain converted into a property you inherit.

The SDK is the only framework that formalizes BOTH Module 1.3 subagent patterns as first-class primitives. Handoffs (delegation by transfer — terminal, subagent owns task) and agents-as-tools (delegation by query — parent retains control, gets structured result). Most harnesses implement one or neither. This is the 4/5 on Module 10 (Subagents) and the SDK-level realization of Module 1.3.

The NemoClaw connection is load-bearing: same principle, different layer. NemoClaw (DD-09) places the governance gate OUTSIDE the agent's reach; the SDK places credentials OUTSIDE the sandbox's reach. Both enforce a property by placing the enforcement outside the principal's address space — structurally unreachable. This cross-cutting pattern is what the course wants you to internalize.

After This Module
01
State the SDK's defining contribution — the 2-layer harness/compute split — and explain why credential isolation is architectural rather than optional.
02
Name the 7 sandbox providers and explain why the provider abstraction makes the harness code provider-agnostic (swap by config, not code change).
03
Distinguish handoffs from agents-as-tools (Module 1.3) and explain why the SDK is the only framework that formalizes both as first-class primitives.
04
Score the SDK 38/60 and defend the shape: 5/5 on Module 5 (reference), 4/5 on Module 10, low on observability and verification (bring-your-own layers).
05
Articulate the NemoClaw connection — same principle (enforcement outside the principal's reach), different trust boundary (credentials outside the sandbox's reach vs governance outside the agent's reach).
Artifacts
01
Teaching Document
Teaching document — the SDK thesis (primitives not product), the 2-layer harness/compute split, the 7-provider sandbox abstraction (Module 5 reference at 5/5), handoffs + agents-as-tools (Module 1.3), declarative-by-composition orchestration, the 38/60 score profile, the NemoClaw connection, anti-patterns; with learning objectives, key terms, references
READ
02
Diagrams
6 Mermaid/n8n diagrams — the 2-layer harness/compute split, the 7-provider sandbox abstraction, handoffs vs agents-as-tools, declarative-by-composition orchestration, the NemoClaw same-principle-different-layer connection, the n8n 2-layer dispatch workflow
READ
03
Slide Deck
10 slides — reveal.js, dark theme, design-system teal; covers the primitives-not-product thesis, the 2-layer split, the 7 providers, handoffs + agents-as-tools, the NemoClaw connection, declarative-by-composition, the score profile, anti-patterns, the lab
READ
04
Teaching Script
Verbatim teaching transcript with [SLIDE N] cues, ~2,100 words spoken across 10 slide cues
READ
05
Flashcards
22 flashcards (TSV) — mix of recall, application, and analysis; covers the SDK category, the 2-layer split, the 7 providers, credential isolation as architecture, handoffs + agents-as-tools, the NemoClaw connection, the score profile, anti-patterns
TEST
06
Exam
15 questions, 20/40/40 Bloom distribution (3 recall / 6 application / 6 analysis), 70% pass; validated JSON with rationale per question; covers the 2-layer split, 7-provider abstraction, handoffs vs agents-as-tools, the NemoClaw parallel, the orchestration spectrum, the OpenAI-camp contrast with Codex CLI
TEST
07
Lab Spec
Simulate the 2-Layer Harness/Compute Split — runnable simulation (Python 3.10+, type hints, no GPU, no external deps): confirm credentials cannot cross the boundary, swap all 7 providers by configuration, demonstrate both subagent patterns (handoffs + agents-as-tools), verify the NemoClaw parallel (~45-60 min)
DO
08
Module Web Page
Single-file HTML hub
HERE