Get a rigorous, prioritized code review that flags real bugs before your teammates do.

You are a staff engineer performing a thorough code review. Review the following [LANGUAGE] code that implements [WHAT THE CODE DOES]. Focus, in priority order, on: (1) correctness and edge cases, (2) security vulnerabilities, (3) concurrency or race conditions, (4) performance, and (5) readability and naming. For each issue, cite the exact line or block, explain the concrete failure it could cause, rate its severity as Blocker / Major / Minor, and propose a specific fix with a short code snippet. Do not comment on style already enforced by a linter. End with a 3-bullet summary of the most important changes. Here is the code:

[PASTE CODE]

Systematically diagnose a bug from symptoms, logs, and code instead of guessing.

Act as a debugging partner using the scientific method. I am seeing this bug: [DESCRIBE THE OBSERVED BEHAVIOR VS. EXPECTED BEHAVIOR]. Environment: [LANGUAGE / FRAMEWORK / VERSIONS]. Relevant code, logs, and stack trace are below. Do the following: (1) restate the problem in one sentence, (2) list the 3-5 most likely root causes ranked by probability with your reasoning, (3) for the top hypothesis, describe the smallest experiment or log line I can add to confirm or rule it out, and (4) once confirmed, give the minimal fix. Ask me for any missing information before speculating. Code and logs:

[PASTE CODE, LOGS, STACK TRACE]

Refactor messy code into a cleaner design without changing its external behavior.

You are an expert in refactoring and clean code. Refactor the following [LANGUAGE] code while preserving its exact external behavior and public API. Goals: reduce complexity, remove duplication, improve naming, and separate concerns. Constraints: [LIST ANY CONSTRAINTS, e.g. no new dependencies, must stay backward compatible]. Deliver: (1) the refactored code, (2) a bullet list of each change and why it improves the design, and (3) any behavior-preserving tests I should add to lock in the current behavior before merging. Do not introduce features or fix unrelated issues. Original code:

[PASTE CODE]

Generate a thorough unit test suite covering happy paths, edge cases, and failures.

Write a complete unit test suite for the following [LANGUAGE] function/module using [TEST FRAMEWORK, e.g. Jest / pytest]. Cover: the happy path, boundary values, empty and null inputs, invalid inputs and expected errors, and any concurrency or side-effect behavior. Use clear Arrange-Act-Assert structure and descriptive test names that read as specifications. Mock external dependencies such as [LIST DEPENDENCIES]. After the tests, list any behaviors that are ambiguous or untestable as written and suggest how the code could be made more testable. Code under test:

[PASTE CODE]

Pressure-test an architecture for scalability, failure modes, and cost before you build it.

You are a principal architect reviewing a proposed system design. The system must [DESCRIBE WHAT IT DOES] at a scale of [EXPECTED LOAD / DATA VOLUME / USERS], with these non-functional requirements: [LATENCY, AVAILABILITY, CONSISTENCY, BUDGET]. My proposed design is: [DESCRIBE COMPONENTS, DATA STORES, APIS, DATA FLOW]. Evaluate it on scalability, reliability and failure modes, data consistency, security, operational complexity, and cost. Identify the top 3 risks and single points of failure, propose concrete mitigations, and note where I am over-engineering. If a materially simpler design would meet the requirements, describe it. Use clear sections and diagrams-as-text where helpful.

Produce a correct regex or shell command plus a plain-English explanation you can trust.

I need a [regex / shell command / SQL snippet] for [TARGET ENVIRONMENT, e.g. JavaScript regex, bash on Linux]. Goal: [DESCRIBE EXACTLY WHAT IT SHOULD MATCH OR DO]. It must handle these examples correctly: matches -> [POSITIVE EXAMPLES]; must NOT match -> [NEGATIVE EXAMPLES]. Provide: (1) the final expression/command, (2) a token-by-token explanation of each part, (3) any edge cases or gotchas, and (4) a safer or more readable alternative if one exists. Prefer clarity over cleverness.

Turn a business question into a correct, efficient SQL query for your exact schema.

You are a senior data engineer writing [SQL DIALECT, e.g. PostgreSQL / BigQuery] SQL. My schema is:

[PASTE CREATE TABLE STATEMENTS OR COLUMN LISTS]

I need to answer this business question: [DESCRIBE THE QUESTION]. Write a single, correct query. Then: (1) explain the logic step by step, (2) flag any assumptions you made about the data (e.g. how NULLs, duplicates, or time zones are handled), (3) suggest indexes or query rewrites that would improve performance at scale, and (4) show a small sample of what the output rows would look like. Do not use SELECT *; return only the columns needed.

Get a structured EDA plan and Python code before diving into a new dataset.

Act as a data scientist starting exploratory data analysis. My dataset is described as: [DESCRIBE COLUMNS, TYPES, ROW COUNT, AND SOURCE]. My analytical goal is: [WHAT DECISION OR QUESTION THIS SUPPORTS]. Produce: (1) a prioritized EDA checklist covering data quality, distributions, missingness, outliers, and key relationships to investigate; (2) runnable Python (pandas + matplotlib/seaborn) for the highest-value checks; (3) the specific hypotheses each check would confirm or reject; and (4) common traps to watch for with this kind of data. Keep code idiomatic and commented.

Design an end-to-end retrieval-augmented generation pipeline tailored to your corpus.

You are an applied AI engineer designing a production RAG system. Use case: [DESCRIBE THE APP AND USERS]. Corpus: [DESCRIBE DOCUMENT TYPES, VOLUME, UPDATE FREQUENCY, SENSITIVITY]. Constraints: [LATENCY, BUDGET, ON-PREM VS CLOUD, PREFERRED MODELS]. Design the full pipeline and justify each choice: chunking strategy and size, embedding model, vector store, retrieval method (dense/hybrid/reranking), metadata and filtering, prompt-assembly and context-window budget, citation handling, and guardrails against hallucination. Then list the top 3 failure modes for this use case and how to mitigate each. Present as clearly labeled sections.

Define metrics and a test set to objectively measure whether your RAG system works.

Help me build an evaluation harness for a RAG system that [DESCRIBE WHAT IT ANSWERS]. Define: (1) the core metrics for retrieval (e.g. context recall, precision) and generation (faithfulness, answer relevance, correctness) with plain-English definitions; (2) how to construct a gold test set of question/answer/expected-source triples for my domain, including how many examples and how to source them; (3) an LLM-as-judge rubric with a 1-5 scale and explicit anchor descriptions for each score; and (4) how to detect regressions when I change chunking, the embedding model, or the prompt. Output a concrete, actionable plan.

Systematically audit a dataset for quality issues before it feeds a model or report.

You are a data quality analyst. I am about to use a dataset for [PURPOSE, e.g. training a model / an executive dashboard]. Columns and types: [LIST COLUMNS]. Known context: [SOURCE, HOW IT IS COLLECTED, ANY KNOWN ISSUES]. Produce a data quality audit plan covering completeness, uniqueness, validity, consistency, timeliness, and accuracy. For each dimension, give the specific check to run (with SQL or pandas), a reasonable threshold for pass/fail, and the downstream risk if it fails. Finish with a prioritized list of what to fix first and why.

Design a normalized, future-proof database schema from plain requirements.

Act as a database architect. Design a schema for [DESCRIBE THE DOMAIN / APPLICATION]. Key entities and relationships: [DESCRIBE THEM]. Expected scale and access patterns: [READS/WRITES, QUERY TYPES, GROWTH]. Target database: [e.g. PostgreSQL]. Deliver: (1) CREATE TABLE statements with appropriate types, keys, and constraints; (2) an explanation of normalization decisions and any deliberate denormalization for performance; (3) recommended indexes tied to the stated query patterns; and (4) how the schema should evolve to handle [LIKELY FUTURE REQUIREMENT]. Note any trade-offs explicitly.

Craft a robust system prompt that gives an AI agent a clear role, boundaries, and behavior.

You are an expert in prompt engineering for autonomous AI agents. Help me write a production system prompt for an agent whose job is: [DESCRIBE THE AGENT'S PURPOSE]. It has access to these tools: [LIST TOOLS AND WHAT EACH DOES]. Users are: [WHO INTERACTS WITH IT]. The system prompt must define: role and objective, scope and explicit out-of-scope behavior, how and when to use each tool, how to handle uncertainty and missing information, tone, safety and refusal rules, and the required output format. Write the full system prompt, then explain the reasoning behind each section so I can tune it.

Write clean tool definitions and descriptions an LLM will actually call correctly.

Help me define a tool for an LLM agent to call. The tool should [DESCRIBE WHAT THE TOOL DOES] by wrapping [API / FUNCTION / SYSTEM]. Produce a JSON-schema-style function definition with: a clear name, a description written so the model knows exactly when to use it and when not to, and a parameters object where every field has a description, type, and whether it is required. Then write 3 example user requests: two where the model should call the tool (with the exact arguments) and one where it should not. Flag any parameters that are ambiguous or easily hallucinated and suggest how to constrain them.

Decompose a complex task into a coordinated multi-agent system with clear responsibilities.

You are an architect of multi-agent AI systems. I want to automate this workflow: [DESCRIBE THE END-TO-END TASK AND SUCCESS CRITERIA]. Design a multi-agent solution: (1) list each agent with its single clear responsibility, inputs, outputs, and tools; (2) show how they hand off work and who orchestrates them (supervisor, sequential, or graph); (3) specify the shared state or memory passed between them; (4) define where a human should be in the loop; and (5) identify the top failure modes (loops, context loss, conflicting outputs) and how the design prevents them. Justify why this is better than a single agent for this task.

Turn an agent workflow into an explicit graph of nodes, edges, and state transitions.

Act as a LangGraph/state-machine expert. I want to build an agent that [DESCRIBE THE GOAL]. Model it as a graph. Define: (1) the state object schema (what fields it carries between steps); (2) each node as a function with its purpose, what it reads from state, and what it writes back; (3) the edges, including conditional edges and the routing logic for each decision point; (4) the entry point and the termination conditions; and (5) where retries, human review, or tool calls fit. Present it as a labeled node/edge list plus a short pseudocode skeleton. Keep the graph as simple as the task allows.

Diagnose why an agent is misbehaving and design evals to prevent regressions.

My AI agent is misbehaving. Its job: [DESCRIBE]. Observed problem: [DESCRIBE THE BAD BEHAVIOR, e.g. wrong tool calls, loops, hallucinated facts, ignores instructions]. Below are the system prompt and a failing trace. Diagnose the likely cause, distinguishing between prompt issues, tool-description issues, missing context, and model limitations. Give the specific edit that would most likely fix it. Then design 3-5 evaluation cases (input + expected behavior) that would catch this class of failure automatically in the future. System prompt and trace:

[PASTE SYSTEM PROMPT AND TRACE]

Generate a search-optimized, well-structured article outline that matches user intent.

You are an SEO content strategist. Create a detailed outline for a blog post targeting the primary keyword "[KEYWORD]" for an audience of [TARGET READER]. Search intent: [informational / commercial / transactional]. Deliver: (1) 3 title options with the keyword placed naturally, each under 60 characters; (2) a meta description under 155 characters; (3) an H1 plus a full H2/H3 structure where each heading maps to a sub-question the reader is asking; (4) secondary keywords and related entities to include for topical depth; and (5) suggested internal-link anchors and a clear call to action. Do not write the full post yet; make the outline comprehensive enough that any writer could execute it.

Produce multiple ad variations built on distinct angles for A/B testing.

You are a direct-response copywriter. Write ad copy for [PRODUCT/SERVICE] targeting [AUDIENCE] on [PLATFORM, e.g. Google Search / LinkedIn / Meta]. The core value proposition is [VALUE PROP] and the single desired action is [ACTION]. Produce 5 distinct variations, each using a different psychological angle (e.g. pain-point, social proof, curiosity, urgency, aspiration). For each, respect the platform's character limits and include a headline, body, and CTA. Label the angle used and note which audience segment it should resonate with most. Avoid hype words and unverifiable claims.

Design a multi-touch email sequence that moves leads from signup to conversion.

Act as a lifecycle marketing expert. Design a [NUMBER]-email nurture sequence for [PRODUCT/SERVICE]. The subscriber just [TRIGGER EVENT, e.g. downloaded a guide / started a trial]. Goal of the sequence: [DESIRED OUTCOME]. For each email give: the send timing (e.g. Day 0, Day 2), the single objective of that email, a subject line plus preview text, a concise body with one clear CTA, and the reader emotion or objection it addresses. Ensure the sequence tells a coherent story and escalates toward the goal without being pushy. End with the key metric to watch for each email.

Rewrite any copy to match a defined brand voice while keeping the message intact.

You are a brand copy editor. Our brand voice is: [DESCRIBE VOICE, e.g. confident, warm, plain-spoken, no jargon] and we avoid [WORDS/TONES TO AVOID]. Rewrite the text below so it fully matches this voice while preserving the meaning, facts, and any legal or technical accuracy. Keep it roughly the same length unless tightening improves clarity. Provide the rewrite, then a short bullet list of the specific voice changes you made so I can apply the same rules myself next time. Text to rewrite:

[PASTE TEXT]

Turn one long-form asset into a full week of platform-native content.

Act as a content repurposing strategist. Below is a [BLOG POST / WEBINAR TRANSCRIPT / REPORT]. Repurpose it into: (1) a LinkedIn post with a strong hook and a clear takeaway; (2) a 5-tweet thread; (3) 3 short-form video hooks/scripts (under 30 seconds each); (4) a plain-text email teaser with a CTA to the original; and (5) 3 quote-card snippets suitable for graphics. Keep the core insight intact but adapt tone and format to each platform's norms. Preserve any data or claims exactly as stated in the source. Source content:

[PASTE CONTENT]

Write a researched, non-spammy cold email that earns a reply.

You are a top B2B SDR who writes concise, researched cold emails. Prospect: [NAME, ROLE, COMPANY]. What I know about them or a relevant trigger: [RECENT NEWS, POST, INITIATIVE, OR PAIN]. What we sell: [PRODUCT] which helps [ICP] achieve [OUTCOME]. Write a cold email under 120 words that: opens with a specific, personal observation (not flattery), connects it to a concrete problem we solve, states one quantified proof point, and ends with a low-friction ask (not "got 15 minutes?"). No buzzwords, no "I hope this finds you well." Then give 2 alternative subject lines under 6 words each.

Prepare sharp discovery questions that surface pain, impact, and buying process.

Act as a sales coach trained in consultative selling (SPIN / MEDDIC). I have a discovery call with [PROSPECT ROLE] at [COMPANY TYPE] who may need [PRODUCT/CATEGORY]. Prepare a question bank organized into: (1) current-state and situation, (2) problems and their business impact, (3) cost of inaction / urgency, (4) decision process, budget, and stakeholders, and (5) success criteria. Make questions open-ended and non-leading. For each section, note what a strong answer sounds like versus a red flag, and suggest a natural follow-up probe. Keep it to the 12 highest-signal questions.

Turn common objections into confident, empathetic responses that keep deals alive.

You are an enterprise sales expert. For my product [PRODUCT] sold to [ICP], build an objection-handling playbook for these objections: [LIST OBJECTIONS, e.g. "too expensive", "we already use X", "no time right now", "need to check with my boss"]. For each objection: (1) name the real concern likely behind it, (2) give a response using the acknowledge-reframe-evidence-question structure, (3) provide the exact words to say, and (4) note what to do if it is a genuine dealbreaker versus a stall. Keep responses honest and non-manipulative; the goal is to advance or disqualify, not to pressure.

Draft a tailored proposal that frames value and ROI in the buyer's language.

Act as a solutions consultant writing a proposal for [PROSPECT COMPANY]. Context from discovery: their goals are [GOALS], current pains are [PAINS], and success looks like [DESIRED OUTCOMES]. We propose [SOLUTION] priced at [PRICING]. Write a proposal with: an executive summary in their language, a problem statement that mirrors what they told us, the proposed solution mapped point-by-point to their goals, a quantified ROI / business case, implementation timeline and responsibilities, and clear next steps. Keep it outcome-focused, not feature-focused, and flag any assumptions I should validate before sending.

Analyze call notes to surface deal risks and recommend the best next move.

You are a sales manager reviewing a deal. Below are my notes from recent interactions with [PROSPECT/COMPANY]. Deal value: [AMOUNT], target close: [DATE]. Analyze the deal and return: (1) a qualification assessment against MEDDIC (metrics, economic buyer, decision criteria, decision process, pain, champion) noting what is confirmed vs. unknown; (2) the top 3 risks that could kill or stall it; (3) the single most important next action and exactly how to execute it; and (4) 2 questions I still must get answered. Be candid if this deal looks unlikely to close. Notes:

[PASTE NOTES]

Turn a messy decision into a crisp one-page memo with options and a recommendation.

You are a chief of staff who writes sharp decision memos. Help me structure a decision about [DECISION]. Context and constraints: [BACKGROUND, DEADLINE, BUDGET, STAKEHOLDERS]. Write a one-page memo with: (1) the decision to be made stated as a clear question, (2) relevant context and what has changed, (3) 2-3 viable options each with pros, cons, cost, and risk, (4) a clear recommendation with the reasoning, (5) what could make us wrong and how we'd know, and (6) what we need from the reader (approve / input / FYI). Be concise and neutral in the options, decisive in the recommendation.

Draft measurable, ambitious OKRs aligned to a higher-level goal.

Act as an OKR coach. Help my [TEAM/FUNCTION] set OKRs for [TIME PERIOD, e.g. Q3]. Our company-level priority is: [COMPANY GOAL]. Our team's mission is: [MISSION]. Propose 2-3 Objectives that are qualitative, inspiring, and clearly ladder up to the company goal. For each, write 3-4 Key Results that are measurable, outcome-based (not task lists), and ambitious yet realistic. For every Key Result, note the metric, a suggested baseline-to-target, and how it would be measured. Then flag any proposed KR that is actually an activity in disguise and rewrite it as an outcome.

Build a fair, signal-rich interview plan and scorecard for a role.

You are an expert in structured hiring. I am hiring a [ROLE] whose top responsibilities are [KEY RESPONSIBILITIES] and who must succeed at [WHAT SUCCESS LOOKS LIKE IN 6-12 MONTHS]. Design an interview kit: (1) the 4-6 core competencies to assess and why each matters for this role; (2) for each competency, 2 behavioral or work-sample questions plus what a strong vs. weak answer looks like; (3) a scoring rubric (1-4) with anchored descriptions; and (4) a suggested panel and how to divide competencies to avoid overlap and bias. Keep questions job-relevant and legally sound; avoid trivia and brain-teasers.

Stress-test a plan by imagining its failure before you commit resources.

Facilitate a pre-mortem on the following plan. Plan: [DESCRIBE THE INITIATIVE, GOAL, TIMELINE, AND RESOURCES]. Assume it is [TIME PERIOD] later and the initiative has clearly failed. Generate: (1) the 8-10 most plausible reasons it failed, spanning execution, market, team, technical, and financial causes; (2) which of these are within our control; (3) for the top 4 risks, an early-warning signal we could monitor and a concrete preventive action to take now; and (4) one assumption in the plan that, if wrong, would be catastrophic. Be blunt and specific, not generic.

Compress a busy period into a signal-dense update leaders will actually read.

Act as a communications advisor to executives. Turn my raw notes into a concise [board / leadership] update for [PERIOD]. Audience cares most about: [PRIORITIES/METRICS]. Structure it as: (1) a 3-sentence TL;DR with the overall status (on track / at risk / off track); (2) key metrics vs. targets with brief commentary; (3) top wins; (4) top risks or asks, each with what I need from the reader; and (5) what's next. Lead with what changed and why it matters, cut activity that has no outcome, and keep the whole thing scannable in under two minutes. Raw notes:

[PASTE NOTES]

Convert a raw transcript into clean notes, decisions, and owned action items.

You are an executive assistant summarizing a meeting. From the transcript/notes below, produce: (1) a 3-4 sentence summary of what the meeting was about and its outcome; (2) key decisions made, each stated in one line; (3) action items as a table with owner, task, and due date (write "unassigned" or "no date" where unstated rather than guessing); (4) open questions or unresolved items; and (5) anything requiring follow-up from someone not in the room. Be faithful to the source and do not invent commitments. Transcript:

[PASTE TRANSCRIPT]

Get a layered summary of a long document tuned to why you're reading it.

Summarize the document below for a reader who is a [YOUR ROLE] and needs it in order to [YOUR GOAL / DECISION]. Provide three layers: (1) a one-sentence bottom line; (2) a 5-7 bullet executive summary of the most decision-relevant points; and (3) a section-by-section breakdown of the key claims, data, and any caveats. Then list: the 3 most important takeaways for my goal, anything that seems weak, missing, or unsupported, and 2 questions I should ask after reading. Quote figures and specifics exactly; do not round or embellish. Document:

[PASTE DOCUMENT]

Prioritize a pile of emails and draft appropriate replies in your voice.

Act as my executive assistant. Below are emails I need to handle. My role and priorities are: [ROLE / WHAT MATTERS MOST]. For each email: (1) classify it as Urgent / Important / Low / Ignore with a one-line reason; (2) recommend an action (reply, delegate, schedule, archive); and (3) for anything needing a reply, draft a concise response in a [TONE, e.g. warm but efficient] voice, leaving [BRACKETS] where I must add specifics. Then give me a suggested order to work through them. Emails:

[PASTE EMAILS]

Get a structured briefing on an unfamiliar topic with sources and open questions.

Act as a research analyst briefing me on [TOPIC] because I need to [REASON, e.g. make a decision / prep for a meeting / write a report]. My current knowledge level is [BEGINNER / INTERMEDIATE / EXPERT]. Produce a structured brief: (1) a plain-language overview of what it is and why it matters; (2) the key concepts, players, and terminology I must know; (3) the main debates, trade-offs, or open questions; (4) the most important facts and figures with a note on how confident we can be in each; and (5) what I should read or ask next. Clearly separate well-established facts from areas of uncertainty, and flag anything you are unsure about rather than guessing.

Turn a chaotic task list into a focused, realistic weekly plan.

Act as a productivity coach using outcome-based planning. Here is everything on my plate this week: [DUMP ALL TASKS, MEETINGS, AND GOALS]. My most important objective this week is [TOP GOAL], and I realistically have about [HOURS] of focus time. Help me: (1) identify the 3 outcomes that would make this a successful week; (2) sort tasks into do-now, schedule, delegate, and drop, with a one-line reason each; (3) propose a day-by-day plan that protects deep-work time and is realistic given my meetings; and (4) call out where I am over-committed and what to renegotiate or cut. Be honest about trade-offs.

Draft clear, kind, and effective feedback or a sensitive message.

You are a communication coach skilled in giving feedback. I need to communicate the following to [PERSON / RELATIONSHIP, e.g. a direct report]: [WHAT NEEDS TO BE SAID AND THE DESIRED OUTCOME]. Context and any history: [BACKGROUND]. Help me draft it using a clear structure (situation, specific behavior, impact, and request), keeping it direct but respectful. Provide: the message itself, an explanation of the choices you made, and how to open and close the conversation. Also anticipate how they might react defensively and how I should respond. Avoid vague praise sandwiches and softening that hides the message.

Understand any hard concept through a layered, analogy-driven explanation.

Explain [CONCEPT] to me using the Feynman technique. My background: [WHAT I ALREADY KNOW]. I want to understand it well enough to [GOAL, e.g. explain it to my team / use it in a decision]. Give me: (1) an explanation as if I were 12, using a concrete everyday analogy; (2) then the accurate, precise version with the real terminology; (3) where the analogy breaks down; (4) the 2-3 most common misconceptions; and (5) one or two questions to test whether I actually understood it. Keep it engaging and build up in layers rather than dumping jargon.

Rewrite a weak prompt into a clear, structured one that gets far better results.

You are a prompt engineering expert. Improve the prompt below so it reliably produces high-quality output. First, identify what is unclear, missing, or likely to cause a vague or wrong answer (e.g. no role, no context, no output format, ambiguous goal). Then rewrite it into a strong prompt that includes: a clear role, the task and context, explicit constraints, the desired output format, and where the user should insert their own details as [PLACEHOLDERS]. Finally, explain the key changes you made and why, so I can improve my own prompts next time. My original prompt:

[PASTE YOUR PROMPT]