awesome-deepseek-prompts
A curated guide of DeepSeek prompting techniques in Chinese, covering the V4 model's thinking-mode settings, official prompt libraries, and ready-to-use templates for analysis, math, coding, and deep-research tasks.
This repository collects prompt-writing guidance and template prompts for DeepSeek models, focused mainly on DeepSeek-V4's thinking mode. It explains how to choose between non-thinking, high, and max reasoning effort depending on the task, documents API-level rules such as which sampling parameters thinking mode ignores and how reasoning content must be replayed on later turns when tool calls are involved, and gives ten writing rules, for example describing the task and acceptance criteria directly instead of asking the model to think step by step, since reasoning is already built in.
It provides ready-to-use templates for complex analysis, competition-style math problems, coding tasks aimed at agents such as Claude Code or Codex, and structured deep-research reports, plus links to DeepSeek's own official guides for thinking mode and function calling.
The material is written entirely in Chinese and links out to a mix of official DeepSeek prompt libraries and community role-play and other prompt collections.