Most agent tutorials show a handful of examples; this repo collects full, runnable workshop code that walks from an amnesiac baseline to production-style managed agents and eval-driven iteration. The value is concrete recipes and graded evals you can run to see how prompt, memory, and orchestration changes affect real outputs.
What Sets It Apart
- Practical workshop artifacts: complete sample projects (Streamlit incident dashboard, PPTX-generating agent, Deal Desk multi-agent demo, agent-battle game harness) rather than abstract diagrams, so you can inspect prompt variants, skills, and evaluation harnesses.
- Focus on agent engineering primitives: memory stores and Dreaming Service for cross-session consolidation, Skills + callable_agents decomposition, and MCP-based integrations for streaming and tool gating — each taught with incremental exercises and programmatic evals.
- Eval-driven workflow examples: an explicit grader setup (programmatic metrics + LLM-as-judge) to iterate agent designs against a 10-task suite, showing measurable deltas instead of ad-hoc changes.
Who it's for and trade-offs
Great fit if you want hands-on examples for building and evaluating LLM-based agents tied to the Anthropic/Claude ecosystem, or if you need concrete patterns for memory, multi-agent coordination, and grading outputs. Look elsewhere if you need a maintained, production-grade SDK or vendor-agnostic libraries: examples assume Anthropic platform concepts and keys, some demos call platform services, and the repo is marked not maintained and not accepting contributions.
Where it fits
Serves as a pragmatic companion to API docs and SDKs: use it to prototype agent architectures, test eval-led prompt changes, or study memory/dreaming patterns. It complements official SDK samples but is not a drop-in production framework — treat it as workshop material and learning code rather than a maintained product.