Adrian2016/awesome-prompts

Awesome Prompts

Prompts from ai-boost/awesome-prompts, in two subsets:

• engineering: 372 system prompts for agents, coding, research and safety work. 73 of them cite arXiv papers by id (104 papers in all), and 166 link the open-source projects they draw on.
• library: 2777 everyday prompts for writing, marketing, study and design, each in English and Chinese.

The engineering prompts can also be browsed and searched online at agentcraft.dev.

from datasets import load_dataset

eng = load_dataset("Adrian2016/awesome-prompts", "engineering", split="train")
lib = load_dataset("Adrian2016/awesome-prompts", "library", split="train")

engineering

Each prompt is a full system prompt written to put one paper, product or practice to work, such as a Lean 4 theorem prover built from Goedel-Architect, or a coding agent following OpenAI's Codex prompting guide.

Field	Meaning
id	File name in the GitHub repo, without extension
name	Display name
category	Section of the GitHub README
description	One-line summary, usually naming the source and date
prompt	Full prompt text
lang	en or zh
arxiv_ids	arXiv ids cited in the description or prompt
github_repos	owner/repo of open-source projects it is based on
url	Source file on GitHub

Category	Prompts
Coding & Development	103
Meta & Prompt Engineering	72
Safety & Compliance	27
AI & ML	24
Product & Strategy	22
Research & Analysis	20
Writing & Academic	17
Image, Video & Audio Generation	13
Knowledge & Documentation	9
Productivity & Tasks	9
Legal & Compliance	8
DevOps & SRE	7
Healthcare & Clinical	6
Industrial & Automotive	6
Learning & Education	6
Game Development	6
Data Engineering	4
Project Management	4
Creative & Role-play	4
Translation	4
Context Engineering	1

library

Collected from prompts.chat and Chinese prompt projects, then categorized and given titles in both languages. 2052 rows have both versions written by people. In the other 725, the language named in translated was machine translated.

Columns: id, category, lang (language of the original: en, zh or both), title_en, title_zh, prompt_en, prompt_zh, translated, for_devs, source, license.

Category	Prompts
image	743
coding	613
business	174
education	151
writing	138
marketing	117
lifestyle	116
research	109
prompting	108
productivity	90
creative	76
finance	72
games	69
data	56
health	56
language	54
roleplay	35

Papers cited

Each line links a paper to the engineering prompts built on it.

• arXiv:2211.01910: self_distillation_code_strategist
• arXiv:2312.08935: verifier_engineering_strategist
• arXiv:2406.07524: diffusion_lm_prompt_engineer
• arXiv:2412.06559: verifier_engineering_strategist
• arXiv:2501.12948: verifier_engineering_strategist
• arXiv:2502.09992: diffusion_lm_prompt_engineer
• arXiv:2502.18600: chain_of_draft, reasoning_theater_diagnostician
• arXiv:2503.06692: reasoning_drift_auditor, test_time_compute_scaling_strategist
• arXiv:2507.19457: self_distillation_code_strategist
• arXiv:2510.04618: agentic_context_engineering_architect
• arXiv:2601.01885: agent_memory_architect
• arXiv:2601.06112: agent_reliability_engineer
• arXiv:2601.07190: prompt_compression_strategist, agent_memory_architect
• arXiv:2601.09625: agent_red_team_architect
• arXiv:2601.13671: vendor_diverse_multi_agent_designer
• arXiv:2601.16507: reprompt_requirements_engineering_prompt_architect
• arXiv:2601.18554: structured_schema_instruction_designer
• arXiv:2601.21557: meta_context_engineering_architect
• arXiv:2601.22311: lookahead_planning_specialist, web_agent_failure_diagnostician, test_time_compute_scaling_strategist
• arXiv:2602.03442: procedural_knowledge_architect
• arXiv:2602.13517: reasoning_drift_auditor, reasoning_theater_diagnostician, test_time_compute_scaling_strategist
• arXiv:2602.16666: verifier_engineering_strategist, agent_reliability_engineer
• arXiv:2602.16873: vendor_diverse_multi_agent_designer
• arXiv:2602.20021: agent_red_team_architect
• arXiv:2603.00729: qwen3_coder_next_agentic_architect
• arXiv:2603.04421: vendor_diverse_multi_agent_designer
• arXiv:2603.05488: verifier_engineering_strategist, reasoning_drift_auditor, reasoning_theater_diagnostician, test_time_compute_scaling_strategist, meta_cognitive_tool_use_specialist
• arXiv:2603.07379: procedural_knowledge_architect
• arXiv:2603.09619: context_engineering_maturity_architect
• arXiv:2603.11088: agent_red_team_architect
• arXiv:2603.12372: reasoning_drift_auditor, test_time_compute_scaling_strategist
• arXiv:2603.14248: lookahead_planning_specialist, web_agent_failure_diagnostician
• arXiv:2603.15309: meta_cognitive_tool_use_specialist
• arXiv:2603.15401: agent_skill_effectiveness_auditor
• arXiv:2603.15727: agent_skill_supply_chain_auditor, agent_red_team_architect
• arXiv:2603.19461: hyperagents_designer
• arXiv:2603.22862: meta_cognitive_tool_use_specialist
• arXiv:2603.23509: internal_safety_collapse_auditor
• arXiv:2603.25379: 5w3h_intent_architect
• arXiv:2603.27277: codebase_memory_mcp_architect
• arXiv:2604.00025: caveman_mode
• arXiv:2604.00356: agent_trajectory_triage_specialist
• arXiv:2604.00487: vendor_diverse_multi_agent_designer
• arXiv:2604.01161: reasoning_drift_auditor, reasoning_theater_diagnostician
• arXiv:2604.01193: self_distillation_code_strategist, verifier_engineering_strategist
• arXiv:2604.01346: agent_world_model_architect
• arXiv:2604.01348: self_distillation_code_strategist, procedural_knowledge_architect
• arXiv:2604.01438: agent_skill_supply_chain_auditor, agent_red_team_architect
• arXiv:2604.01707: prompt_compression_strategist, agent_memory_architect, procedural_knowledge_architect
• arXiv:2604.02795: structured_schema_instruction_designer
• arXiv:2604.02910: lookahead_planning_specialist
• arXiv:2604.02985: prompt_compression_strategist
• arXiv:2604.03081: agent_skill_supply_chain_auditor
• arXiv:2604.04247: self_distillation_code_strategist, parallel_prompt_learning_strategist
• arXiv:2604.06260: test_time_compute_scaling_strategist, diffusion_lm_prompt_engineer
• arXiv:2604.08222: agent_world_model_architect
• arXiv:2604.08224: cognitive_externalization_architect
• arXiv:2604.08545: meta_cognitive_tool_use_specialist
• arXiv:2604.09059: agent_world_model_architect
• arXiv:2604.10866: agent_world_model_architect
• arXiv:2604.12231: agent_memory_architect
• arXiv:2604.12285: agent_memory_architect
• arXiv:2604.12890: long_horizon_multimodal_search_agent, web_agent_failure_diagnostician
• arXiv:2604.12986: plan_execute_safety_architect
• arXiv:2604.13006: structured_schema_instruction_designer
• arXiv:2604.13531: web_agent_failure_diagnostician
• arXiv:2604.14862: structured_schema_instruction_designer
• arXiv:2604.22709: abstract_chain_of_thought_architect
• arXiv:2605.02765: constraint_typology_architect
• arXiv:2605.03314: disclosure_policy_designer, reasoning_theater_diagnostician
• arXiv:2605.05191: elastic_context_orchestrator
• arXiv:2605.06623: maspo_joint_prompt_optimizer
• arXiv:2605.06651: ai_co_mathematician
• arXiv:2605.06717: proactive_coding_agent_architect
• arXiv:2605.07937: clarification_timing_strategist
• arXiv:2605.10805: verifier_engineering_strategist, llm_judge_routing_strategist
• arXiv:2605.10913: shepherd_meta_agent_runtime_architect
• arXiv:2605.17634: contextual_integrity_agent_architect
• arXiv:2605.20530: agentatlas_trajectory_eval_architect, agent_atlas_trajectory_auditor
• arXiv:2605.22166: runtime_harness_adaptation_architect
• arXiv:2605.23904: agent_skill_optimizer_architect
• arXiv:2605.26275: sandboxed_prompt_engineer
• arXiv:2606.00448: agent_skill_compositional_risk_auditor
• arXiv:2606.02994: reasoning_primitive_induction_architect
• arXiv:2606.04329: memory_poisoning_attack_auditor
• arXiv:2606.04465: sepo_self_evolving_prompt_agent
• arXiv:2606.05009: agentic_deontic_reasoning_architect
• arXiv:2606.06468: formal_theorem_proving_architect
• arXiv:2606.13662: agent_environment_engineering_architect
• arXiv:2606.20515: s_agent_spatial_tool_use_architect
• arXiv:2606.24775: agent_native_memory_system_architect
• arXiv:2606.26057: unfireable_safety_kernel_architect
• arXiv:2606.26356: instruction_bleed_auditor
• arXiv:2606.30005: proprioceptive_context_dashboard_architect
• arXiv:2607.00038: loop_engineering_architect
• arXiv:2607.01793: agent_safety_testing_at_scale_architect
• arXiv:2607.02116: contextnest_verifiable_context_governance_architect
• arXiv:2607.02134: scientific_paper_replication_harness_architect
• arXiv:2607.02509: recontext_recursive_evidence_replay_architect
• arXiv:2607.05120: agent_data_injection_attack_auditor
• arXiv:2607.08028: auditable_enterprise_llm_agent_harness_architect
• arXiv:2607.08662: webswarm_deep_wide_research_orchestrator
• arXiv:2607.08716: proactive_memory_agent_long_horizon
• arXiv:2607.11399: opensquilla_token_efficient_agent_architect

How it was made

The engineering prompts are written and maintained in the GitHub repo, mostly in 2025 and 2026. arxiv_ids and github_repos are pulled from the text with regular expressions, so a prompt that names a paper without an id has an empty list. Two kinds of prompts are left out: 2023-era prompts by other authors, and a copied vendor system prompt.

The library comes from the sources listed in the GitHub repo's library/README.md. Prompts that were sexual, jailbreaks, spam or empty were removed before categorizing, and DeepSeek's official examples are left out because they are quoted, not licensed.

Uses and limits

Good for: starting points for system prompts, examples for prompt-optimization research, and a bilingual set for testing how models follow instructions in English and Chinese.

Limits: no prompt here has been benchmarked. The citations show which paper a prompt draws on, but do not show that the prompt reproduces the paper's results. Machine-translated library rows can read stiffly.

License

• engineering: CC-BY-4.0. Credit "ai-boost/awesome-prompts".
• library: per row, in the license column: CC0-1.0 (2592), MIT (158), Apache-2.0 (27). MIT and Apache-2.0 license texts are in licenses/.