499 saved links, 2020–2026 (peak: 2023). A tag that documents its own obsolescence, gracefully: the early layer is magic-words folklore, the middle is rigorous technique (The Prompt Report, CoT, taxonomies), and the recent layer records the discipline dissolving into something bigger — context engineering, automated optimization (DSPy), and agents that write their own prompts. “Stop prompting and start operating it,” as one 2026 save puts it.
Related: LLMs · AI Agents · Generative AI · RAG
The canon
- The Prompt Report — 200+ techniques, 1,500+ papers; “the single best thing you can read.” The survey site is the interactive version.
- the taxonomy of prompting techniques — “best overview to date,” per your save; and the 150-study survey with 21 prompt-quality properties.
- Lee Boonstra’s prompt engineering whitepaper — “the best free book you can find. Concise & practical.”
- Anthropic’s interactive prompt-engineering tutorial + the Claude docs overview and the prompt library.
- promptingguide.ai · DAIR.AI’s free resource collection · the prompt-engineering techniques repo.
- Model-specific, always worth rereading on release day: “read the prompt guides front to back whenever a new model drops” · the GPT-5.5 prompting guide · Gemini 3 best practices · prompting reasoning models.
From prompting to context engineering
The tag’s big turn, mid-2025.
- “The new skill is not prompting, it’s context engineering” — the article that named it.
- the Context Engineering Handbook · the hands-on tutorial for closing the prototype-to-production gap.
- “prompts are terrible for defining the behaviors of systems” and how Claude Code assembles context — Drew Breunig’s excellent thread of posts.
- stop prompting, start operating: rules, tests, skills, reviewers — the operator’s mindset.
- agents.md and AGENTS.md in action — prompts becoming repo-level configuration.
- treating team standards as prompt material · prompts are code, .md files are state.
- compound engineering — where this all ends up.
Automated prompt optimization
Letting the model do the prompt engineering — the DSPy thread.
- Drew Breunig’s DSPy talk — “the clearest explanation I’ve seen yet” · the talk itself.
- automating prompt creation with DSPy · the Hugging Face course on DSPy GEPA.
- LLMs writing prompts for themselves — now trustworthy · “Let the Model Write the Prompt”.
- the definitive hands-on guide to automated prompt engineering.
- Vertex AI Prompt Optimizer · Salesforce’s prompt-optimization automation.
- the prompt tuning playbook · Uber’s prompt templating toolkit — versioned, evaluated, industrial-scale prompting.
System prompts in the wild
Reading other people’s homework — a genuinely useful genre.
Testing, evals & injection
- Promptfoo: with Vertex AI for evaluation and security · local prompt testing with Docker Model Runner.
- detecting prompt regressions with the Evals API · LLM-Evalkit — metrics for prompt engineering.
- why your prompts don’t belong in Git — the ops argument.
- Hamel Husain’s FAQ on synthetic data for evals.
- Injection: design patterns for securing agents against prompt injection (“a superb paper”) · prompt injection: the risk and mitigations · OpenAI’s CISO on Atlas injection mitigations · Meta’s PromptGuard.
Prompting for everyone
- Google Prompting Essentials — the course to recommend to non-engineer colleagues.
- MIT’s course on writing effective prompts.
- prompt engineering for Copilot Chat · smaller prompts, better results: custom instructions · the Copilot Chat cookbook.
- the chain-of-thought deep dive — the technique that started the whole field.