prompt-atlas

A curated, versioned, searchable library of production-grade prompts for LLM trainers, AI product managers, and evaluation teams. 100 cards across 8 directions — RAG, Agent, RLHF, SFT, Multimodal, Chain-of-Thought, Evaluation, Code. Triggers when the user asks for a prompt for retrieval scoring, multi-hop QA, query rewriting, HyDE, citation auditing, hallucination detection, chunk summarization, context compression, agent planning / tool-call schema, agent reflection, tool-call repair, plan-and-execute, trajectory memory compression, multi-agent sub-task delegation, pairwise preference labeling, pointwise reward scoring, constitutional critique-and-revise, best-of-N selection, red-team prompt generation, instruction augmentation, self-instruct, SFT data filtering, SFT response generation, multi-turn conversation generation, few-shot example selection, structured image captioning, visual question answering, VLM caption verification, OCR structured extraction, chart and table extraction, structured reasoning, least-to-most decomposition, self-consistency aggregation, verify-then-finalize, tree-of-thoughts, LLM-as-judge rubrics, reference-based judging, per-claim factuality, pointwise quality scoring, safety output classification, position-bias-aware pairwise judging, multi-turn dialogue judging, code review, test case generation, code explanation, code evaluation, or refactor suggestion. Use to locate and adapt a Prompt Card rather than writing prompts from scratch.