tomevault-io LangChain Agent Architecture Review
When To Use
- The main decision is about LangChain chain/agent structure, tool definitions, memory configuration, or retrieval pipeline.
- Reviewing prompt engineering, callback boundaries, or evaluation setup.
Workflow
- Identify the chain/agent topology — linear chains, LCEL pipelines, or agent loops.
- Review tool definitions for side-effect boundaries, idempotency, and credential isolation.
- Check memory configuration — is it scoped per session, per user, or global?
- Review retrieval setup — vector store, chunking strategy, retriever configuration.
- Check callback and tracing setup for observability.
- Review prompt templates for injection risks and output structure.
- Recommend the smallest structural change that improves safety or clarity.
Output Format
LangChain architecture review:
- Chain/agent topology:
- Tool definitions & side effects:
- Memory configuration:
- Retrieval pipeline:
- Callbacks & tracing:
- Prompt safety:
- Recommended change:
- Verification:
Source: WangYao-GoGoGo/architecture-agent-skills — distributed by TomeVault.