Diagrams

Raw Excalidraw sources for the interactive diagrams embedded in wiki notes — nothing else lives here. Each file’s rendered PNG lives in _attachments/, embedded directly in the concept page it illustrates; this folder holds only the editable .excalidraw source, linked from that page through a [!diagram] callout so the image opens the pan-and-zoom version.

Not a concept catalog like the root index.md — a diagram isn’t a concept in its own right, it argues one visually. Each entry below points back to the note it belongs to.

Sources

• llm-wiki-pattern.excalidraw — the retrieval-only-vs-wiki-compilation argument from The LLM Wiki Pattern.
• choosing-a-pattern.excalidraw — all twelve agent design patterns mapped along the who-decides-what-runs-next axis, each drawn as the shape of its own architecture, from Choosing an Agentic Design Pattern. It backs only its own page now; the per-pattern glyph crops it used to supply were replaced by the twelve replications below.
• model-context-protocol.excalidraw — the hub diagram from MCP: AI applications on the left, data sources and tools on the right, bidirectional flow through one standardized protocol in the middle. Unlike the other sources here this one is a replication of the official docs’ own figure rather than an original argument, redrawn in stock Excalidraw colors so it adapts to dark mode.
• a2a-protocol.excalidraw — the three arguments of the A2A protocol on one canvas: an agent narrowed by a tool interface beside two agents talking as peers, MCP and A2A drawn as perpendicular axes across an organizational boundary, and the four-step request lifecycle as a timeline.

The three AI-native SDLC figures

One file per figure, each a replication of the corresponding figure in Anthropic’s “The AI-Native SDLC Playbook”, redrawn in stock Excalidraw colors so they adapt to the site’s dark mode. Like the figures below they reproduce a source figure rather than making an original argument. All three sit on The AI-Native SDLC.

• sdlc-bottleneck-moved.excalidraw — the six stages before and after agents, with Build collapsed to a sliver and the freed width marked “cycle time reclaimed.”
• sdlc-line-vs-loop.excalidraw — the traditional lifecycle as a straight line beside the AI-native one as a continuous loop around Claude.
• sdlc-play-dependency-graph.excalidraw — the twelve plays in five rows, solid arrows for a real prerequisite and dotted for a play that helps but is not required.

The twelve pattern figures

One file per pattern, each a replication of the corresponding figure in Google Cloud’s “Choose a design pattern for your agentic AI system”, redrawn in stock Excalidraw colors so they adapt to the site’s dark mode. Like model-context-protocol.excalidraw these reproduce a source figure rather than making an original argument, so the labels and flows follow the source rather than this wiki’s own framing.

• single-agent.excalidraw — Single-Agent Pattern.
• sequential.excalidraw — Sequential Pattern.
• parallel.excalidraw — Parallel Pattern.
• loop.excalidraw — Loop Pattern.
• review-and-critique.excalidraw — Review and Critique Pattern.
• iterative-refinement.excalidraw — Iterative Refinement Pattern.
• coordinator.excalidraw — Coordinator Pattern.
• hierarchical-task-decomposition.excalidraw — Hierarchical Task Decomposition Pattern.
• swarm.excalidraw — Swarm Pattern.
• react.excalidraw — ReAct Pattern.
• human-in-the-loop.excalidraw — Human-in-the-Loop Pattern.
• custom-logic.excalidraw — Custom Logic Pattern.