A curated collection of landmark prompting frameworks extracted from pioneering arXiv research papers, deconstructed for prompt engineering mastery, and reframed for daily productivity, digital operations, marketing, and financial markets.
When elite AI labs publish papers, their prompts consistently outperform generic "chat prompts." Studying their architectures reveals several transferable skills:
Instead of asking the LLM to output a perfect response in a single generation step, paper prompts force iterative transformations (e.g., CoD, Self-Refine). The model progressively compresses, critiques, or enriches content across defined rounds.
Models suffer from sycophancy and attention distraction when irrelevant or biased context is present. Techniques like System 2 Attention teach us to prompt the LLM to regenerate an unbiased, noise-free version of the input before producing an answer.
ReAct showed that pure reasoning creates hallucinations, and pure tool calling lacks planning. Explicitly requiring Thought: -> Action: -> Observation: prevents premature conclusions and grounds operational execution.
Research prompts rigorously define what not to do (e.g., "Do not introduce new entities without fusing previous ones", "Never include meta-discourse like 'Here is your summary'").
Top prompts provide objective grading rubrics or persona perspectives within the prompt so the LLM critiques itself against explicit rubrics rather than vague standards.
Article: {{ ARTICLE }}
You will generate increasingly concise, entity-dense summaries of the above Article.
Repeat the following 2 steps 5 times:
Step 1. Identify 1-3 informative Entities (";" delimited) from the Article which are missing from the previously generated summary.
Step 2. Write a new, denser summary of identical length which covers every entity and detail from the previous summary plus the Missing Entities.
A Missing Entity is:
- Relevant: to the main story.
- Specific: prior knowledge not required to understand it, but not too generic.
- Novel: not in the previous summary.
- Faithful: present in the Article.
- Anywhere: located anywhere in the Article.
Guidelines:
- The first summary should be long (4-5 sentences, ~80 words) yet sparse in information, containing few entities. Use placeholders like "some people" to allow space for future entities.
- Make every word count: rewrite the previous summary to improve flow and make space for additional entities.
- Make space with fusion, compression, and removal of uninformative phrases like "the article discusses".
- The summaries should become highly dense and concise yet remain self-contained and grammatically clean, without lost readability.
- Missing entities can appear anywhere in the new summary.
- Never drop an entity from the previous summary. If space cannot be made, do not add that entity.
Remember, use the exact same number of words for each summary.
Answer in JSON. The JSON should be a list (length 5) of dictionaries whose keys are "Missing_Entities" and "Denser_Summary".Optimized for Financial Markets & Executive Daily Briefings:
Market Data / Report:
"""
{{ PASTE_EARNINGS_CALL_OR_FINANCIAL_REPORT }}
"""
Task: Produce an ultra-dense, executive market intelligence briefing using a 3-step Chain of Density approach.
Follow these strict steps:
Iteration 1 (Baseline Context):
- Write an initial 3-sentence summary (~60 words) highlighting high-level macro movement, company guidance, and market sentiment. Keep it readable but leave room for specific metrics.
Iteration 2 (Adding Key Financial Metrics):
- Identify 3 crucial quantitative metrics from the text (e.g., EPS surprises, revenue growth %, margin compression, CAPEX revisions) not included in Iteration 1.
- Rewrite the summary keeping the exact same length (~60 words), compressing sentences and fusing clauses to integrate these 3 metrics seamlessly.
Iteration 3 (Adding Catalyst & Risk Entities):
- Identify 2 specific forward-looking catalysts or operational risks (e.g., regulatory scrutiny, inventory write-downs, competitor launch).
- Rewrite the summary one final time at the exact same ~60-word length, retaining all previous metrics while integrating these catalysts. Eliminate all passive voice and filler words (e.g., "the company reported that").
Output Format:
Return only the final Iteration 3 summary, followed by a bulleted table of the core metrics and catalysts embedded in the text.Given the following text by a user, extract the part that is unbiased and not containing any opinion, speculation, or irrelevant context. Only return the regenerated context without the distracting details.
Input Text: {{ INPUT_TEXT }}
Regenerated Context:Followed by the downstream task prompt:
Given the regenerated context:
{{ REGENERATED_CONTEXT }}
Answer the following question: {{ QUESTION }}Optimized for Market Research, Competitor Intelligence & Digital Operations:
Raw Source Material (News, Social Media Sentiment, Analyst Commentary):
"""
{{ PASTE_RAW_MARKET_OR_COMPETITOR_NEWS }}
"""
Task: Conduct an objective impact assessment by running a System 2 Attention protocol to eliminate hype, bias, and leading opinions.
Phase 1: Noise & Bias Scrubbing (Context Regeneration)
1. Read the raw text and strip away:
- Emotional/sensationalist vocabulary (e.g., "skyrocketing", "bloodbath", "game-changer", "revolutionary").
- Speculative commentary without attributable hard evidence.
- Irrelevant background anecdotes or leading user opinions.
2. Output a sanitized, neutral factual baseline that retains only verifiable events, validated metrics, dates, and direct actions taken.
Phase 2: Operational Impact Analysis
Using ONLY the sanitized factual baseline generated in Phase 1:
- What is the tangible operational or financial implication for our business/portfolio?
- What are the concrete, verifiable downside risks versus verified upside opportunities?
- What metric must we monitor next week to validate this trend?
Output:
Provide the "Sanitized Factual Baseline" followed by the "Operational Impact Analysis".Solve a question answering task with interleaving Thought, Action, Observation steps.
Thought can reason about the current situation, and Action can be three types:
(1) Search[entity], which searches the exact entity on Wikipedia and returns the first paragraph if it exists. If not, it will return some similar entities to search.
(2) Lookup[keyword], which returns the next sentence containing keyword in the last page successfully searched.
(3) Finish[answer], which returns the answer and finishes the task.
Here are some examples:
Question: What is the elevation range for the area that the eastern sector of the Colorado orogeny extends into?
Thought 1: I need to search Colorado orogeny, find the area that the eastern sector of the Colorado orogeny extends into, then find the elevation range of the area.
Action 1: Search[Colorado orogeny]
Observation 1: The Colorado orogeny was an episode of mountain building in Colorado and surrounding areas...
Thought 2: The eastern sector extends into High Plains. So I need to search High Plains and find its elevation range.
Action 2: Search[High Plains]
Observation 2: High Plains refers to one of two distinct land regions...
Thought 3: High Plains elevation range is from 1,800 to 7,000 ft. So the answer is 1,800 to 7,000 ft.
Action 3: Finish[1,800 to 7,000 ft]Thought step.Optimized for Digital Operations, Customer Support Triage & Issue Remediation:
Incident / Operational Ticket:
"""
{{ PASTE_CUSTOMER_TICKET_OR_SYSTEM_ALERT }}
"""
You are an Autonomous Operations Triage Agent. Resolve or diagnose this operational issue using an explicit ReAct (Thought-Action-Observation) trajectory.
Available Simulated Tools:
- QueryDatabase[query_string]: Simulates looking up user billing, system logs, or account flags.
- CheckSystemStatus[service_name]: Checks uptime, API latency, or service degradation.
- DraftCommunication[target_audience, core_message]: Drafts user or stakeholder notices.
- ResolveTicket[root_cause, action_taken, follow_up]: Completes the investigation.
Instructions:
1. Proceed strictly in sequential cycles of:
- Thought N: Analyze the current situation, identify what knowledge is missing, and determine the exact next step.
- Action N: Call one tool with exact parameters (e.g., QueryDatabase["user_id=1234 status"]).
- Observation N: State what data would be returned or verified based on realistic system behavior.
2. Continue until the root cause is isolated.
3. Finish with Action: ResolveTicket[...].
Do not jump straight to the conclusion. Show at least two Thought-Action-Observation loops before resolving.Task: {{ TASK_DESCRIPTION }}
Input: {{ INPUT }}
Initial Output:
Generate an initial solution for the task.
Feedback:
Critique the initial output. Identify specific weaknesses, missing elements, readability issues, and areas that fail to meet criteria. Be hyper-critical.
Refinement:
Rewrite the output by addressing every criticism raised in the Feedback.Optimized for High-Stakes Marketing Campaigns, Ad Copy & Cold Outbound:
Product & Campaign Brief:
"""
Product: {{ PRODUCT_OR_SERVICE_NAME }}
Target Audience: {{ ICP_OR_BUYER_PERSONA }}
Value Proposition: {{ CORE_VALUE_PROP_AND_PROOF_POINTS }}
Goal: {{ DESIRED_ACTION_E_G_BOOK_DEMO_OR_CLICK_LINK }}
"""
Act as an elite direct-response copywriter and marketing conversion specialist. Produce a winning campaign message using the 3-Stage Self-Refine Method.
### Stage 1: Initial Draft
Write a high-energy initial draft (under 150 words) targeting the audience.
### Stage 2: Brutal Conversion Critique (The "Rubric Evaluation")
Critique Stage 1 against these 4 unforgiving benchmarks:
1. The "So What?" Test: Does the copy lead with features instead of tangible customer outcomes?
2. Friction & Cognitive Load: Are sentences clunky, jargon-heavy, or passive?
3. Credibility & Specificity: Does it use vague claims (e.g., "fast, reliable, best") instead of crisp numbers, social proof, or vivid analogies?
4. Call-to-Action (CTA): Is the CTA intimidating, vague, or low-urgency?
List 3 specific, actionable flaws to fix.
### Stage 3: Polished Final Masterpiece
Rewrite the copy from scratch, resolving every flaw identified in Stage 2.
Include:
- Hook (Pattern interrupt)
- Pain & Solution Bridge
- Concrete Proof Point
- Friction-free Call to ActionYou are an expert orchestrator and problem solver.
To solve the user's task, break down the problem into independent sub-tasks.
For each sub-task:
1. Assign it to an expert persona with domain expertise.
2. Have the persona independently reason through the sub-task.
3. Verify the intermediate findings for internal consistency.
Finally, synthesize all expert inputs into a coherent, validated final answer.Optimized for Digital Operations, Tech Stack Transitions & Business Strategy:
Decision Scenario / Strategic Dilemma:
"""
{{ DESCRIBE_THE_STRATEGIC_CHOICE_OR_OPERATIONAL_CHALLENGE }}
(e.g., Migrating from legacy CRM to AI-driven agentic stack; Expanding to a new vertical)
"""
Task: Formulate a battle-tested decision memo using Meta-Prompting Task Orchestration.
Act as the Executive Orchestrator. Convene a virtual committee of 3 distinct personas:
1. The Chief Operating Officer (Focus: Operational friction, implementation time, change management, staff burnout).
2. The Chief Financial Officer (Focus: Immediate ROI, hidden subscription/maintenance costs, downside risk, cash flow impact).
3. The Head of Product & Innovation (Focus: Competitive advantage, customer velocity, long-term scalability, tech debt).
Execution Steps:
Step 1: Independent Assessments
Each persona must independently evaluate the scenario, listing their top 2 non-negotiable priorities and top 2 major concerns.
Step 2: Cross-Examination
Identify the primary point of friction or disagreement between the COO, CFO, and Head of Innovation. How does each defend their position?
Step 3: Executive Synthesis & 30-Day Action Roadmap
As the Orchestrator, synthesize the debate into:
- The Definitive Recommendation (Go / No-Go / Conditional Pivot).
- Risk Mitigation Safeguards (directly addressing the CFO/COO concerns).
- 30-Day Execution Milestones (Week 1 to Week 4).When browsing arXiv (in categories like cs.CL - Computation and Language, cs.AI - Artificial Intelligence, or cs.SE - Software Engineering), follow this 3-step extraction filter: