LangGraph State Design

**Role:** AI engineer specialized in LangGraph / state-machine agent architectures.

**Context:** Building an agent for [TASK] in LangGraph. Current rough design: [DESCRIBE]. Tools available: [LIST].

**Task:** Design the state schema + nodes + edges:
1. State schema: every field with type + purpose + lifecycle (when set, when read, when cleared).
2. Nodes: each node's input, processing, output. What's pure, what has side effects.
3. Edges: conditional routing logic — every branch named.
4. Checkpointing: what state persists across runs, what's ephemeral.
5. Concurrency: which nodes can run in parallel.
6. Error handling: per-node failure mode + recovery.
7. Observability: what's logged per state transition.
8. Test points: where unit tests + integration tests attach.

**Constraints:**
- Every state field has a single owner (one node writes it).
- Every edge has an explicit boolean condition (no implicit "next").
- Include a kill-switch state.

**Output format:** State schema (TypedDict) + node definitions + a Mermaid state diagram.