Awesome Prompt Engineering — Papers, Tools & Courses

Why curation beats search

Google "prompt engineering" and you get 800 million results. 700 million are repackaged Twitter threads. The remaining 100 million contain the 60 papers that actually matter and a long tail of noise. Awesome Prompt Engineering is the filtered signal — a maintained repo of the work that practitioners actually cite and tools they actually use.

What's inside

The 5 papers you must read first

1. Chain-of-Thought Prompting (Wei et al., 2022) — the foundational technique.
2. ReAct: Reasoning and Acting (Yao et al., 2022) — the pattern underlying most agents.
3. Tree of Thoughts (Yao et al., 2023) — deliberate search over reasoning paths.
4. Constitutional AI (Bai et al., 2022) — how Anthropic shapes model behavior.
5. DSPy: Compiling Declarative LM Calls (Khattab et al., 2023) — the modern alternative to string prompts.

Tools by use case

Structured output

• Instructor (Python) — Pydantic-first structured output for any LLM
• Zod Schema (TS) — runtime type validation for OpenAI function calls

Prompt compilation

• DSPy — declarative programs compiled into optimized prompts
• Guidance — templating with control-flow primitives

Evaluation

• Promptfoo — A/B test prompts across models
• DeepEval — pytest-style LLM evals

Red teaming

• Garak — LLM vulnerability scanner
• Rebuff — prompt injection detection library

Context engineering — the next wave

"Context engineering" emerged in 2025 as the umbrella term for everything beyond prompt wording: retrieval relevance, token budget allocation, memory strategies, tool orchestration. The repo added a dedicated section tracking the ~20 papers on context compression, session management, and long-context scaling laws.

How to use the list in practice

1. Starting fresh? Read the 5 seed papers above plus one survey per month.
2. Building a production app? Pick one tool per axis: structured output, eval, red-team.
3. Running an AI team? Use the courses section as onboarding material — each has quizzes.
4. Doing research? Follow the papers' citation graph; the repo links arxiv + GitHub repros.

Update cadence and license

The list is updated roughly monthly, with 5-15 new entries per release. Entries are removed when linked projects die or when newer work supersedes them. Apache 2.0 license — fork freely.