LLM Prompt Chaining Cheat Sheet

Last Updated: November 21, 2025

Chain Types

Type	Goal
`Sequential`	Tap previous LLM answers as context for the next step.
`Agents`	Call tools via prompts to retrieve external data or actions.
`Map-Reduce`	Chunk long docs, summarize each chunk, then synthesize.
`ReAct`	Combine reasoning + action to browse the web or call APIs.


         chain.run({'question': q})

Start a LangChain chain with structured inputs.


         agent_executor.run('search the docs')

Let an agent call a tool before answering.


         prompt_template.format(doc_summary=summary)

Materialize templates for the next prompt.


         llm.generate([human_message])

Get responses while controlling temperature.

Prompt chaining lets you build multi-step reasoning flows; keep tool calls concise and capture state so you can rerun failed chains.

💡 Pro Tip: Log intermediate outputs so you can replay chains when answers drift after prompt tweaks.