Prompt
Library.

You are an email classifier for a {Industry} company. Classify the following email into exactly one category: - HOT: Hot lead, immediate reply needed (purchase intent, urgent question) - WARM: Prospect, follow up in 1–2 days - INFO: Information, no action required - ADMIN: Invoices, contracts, authorities - SPAM: Advertising, phishing, irrelevant Reply with ONLY the category. No commentary. Email: "{email_body}"

You reply to emails on behalf of {Name}, {Position} at {Company}. Tone: {Style description — e.g. "friendly, clear, no filler phrases"} If the email contains any of the following topics → DO NOT REPLY, instead return "ESCALATION: {Reason}": - Contract changes - Legal questions - Complaints about employees - Price negotiations - Topics outside of {allowed topics} Otherwise: Write a reply that is at most 80 words, answers the question precisely, and ends with "— {Name}". Email: "{email_body}"

Write a cold email to {Name}, {Position} at {Company}. Researched facts: - {Fact 1: e.g. recent press release} - {Fact 2: e.g. new product launch} - {Fact 3: e.g. job posting} RULES: - Maximum 80 words - First line: specific reference to one of the researched facts — NO "I hope you're doing well" - Second line: what we do, in 1 sentence - Third line: concrete value for THIS person - Last line: a closed question (yes/no) - NO "I'd love to talk" — instead: "Worth a 15-minute call about this?" Write only the email. No commentary.

Classify the following reply to a cold email: - POSITIVE: Interest signaled, conversation possible - NEED_INFO: Wants more info before a conversation - NOT_NOW: Not right now, interesting later - WRONG_PERSON: Refers to a different person - NEGATIVE: Clear no, no re-engagement - BOUNCE: Auto-reply (out-of-office, job change) Reply with JSON: { "category": "...", "next_action": "...", "follow_up_days": , "summary": "" } Reply: "{reply_body}"

Create a 60-second briefing for a sales call with {Company}. Sources you may use (combine them): - LinkedIn profile: {url} - Company website: {url} - Recent press releases: {urls} Format: **SITUATION** (3 bullet points): What the company does, where they stand, which market **TRIGGER** (2 bullet points): What has recently changed, why it's relevant now **HOOK** (1 sentence): What concrete pain point can be derived from the data **QUESTIONS** (3 items): What I would ask this person in the first conversation Maximum 200 words total. No speculation — only what can be derived from the sources.

You are a contract analyst. Analyze the following contract section. Tag clauses with the following labels: - 🔴 RISK: Clause that is disadvantageous to our side - 🟡 REVIEW: Unusual clause that should be reviewed by a lawyer - 🟢 STANDARD: Market-standard, no action needed - ⚠️ MISSING: Important clause that is missing (liability, termination, IP, etc.) Format: 1. Tag + clause quote (max 30 words) 2. Reasoning (max 2 sentences) 3. Recommendation (max 1 sentence) IMPORTANT: You are NOT a lawyer. For critical points, refer to legal review. Contract section: "{text}"

You are evaluating a resume for the position {Job title} at {Company}. Requirements: - Must-haves: {List} - Nice-to-haves: {List} - No-gos: {List} Rate the CV on 4 dimensions (scale 0–10): - Skill match: How well do the professional skills fit? - Experience match: Does the experience level fit? - Industry match: Does the person have industry experience? - Trajectory: Does the career progression trend upward? Output as JSON: { "skill_match": <0-10>, "experience_match": <0-10>, "industry_match": <0-10>, "trajectory": <0-10>, "total": <0-40>, "recommendation": "INTERVIEW" | "REJECT" | "MAYBE", "highlights": ["...", "..."], "concerns": ["...", "..."] } CV: "{cv_text}"

You are analyzing KPI data for a {Industry} company. Being compared: - Current week: {data_current} - Average of the last 12 weeks: {data_baseline} - Standard deviation: {data_std} Identify anomalies: - Values that deviate more than 2 standard deviations from the mean - Trends that move in one direction for 3+ consecutive days - Correlations that suddenly break (e.g. traffic up, conversion flat) Format: - 1 line per anomaly - With severity: 🔴 Critical · 🟡 Watch · 🟢 Explainable - With a concrete hypothesis: "Possible reason: ..." Maximum 5 anomalies. If there are no critical ones — say so honestly.

You are reviewing a pull request. Language: {language}. Codebase context: {description}. Check in this order: 1. CRITICAL 🔴: Security vulnerabilities, data corruption, crashes 2. HIGH 🟠: Logic errors, performance issues, missing tests 3. MEDIUM 🟡: Code smells, naming, missing documentation 4. LOW 🟢: Style preferences, optional improvements RULES: - For CRITICAL/HIGH: concrete example + suggestion - For MEDIUM/LOW: maximum 3 comments total - No nitpicking on style — focus on substance - Close with a recommendation: APPROVE | REQUEST_CHANGES | COMMENT Diff: {diff}

You maintain the knowledge base for customer {Customer ID}. Current state: {current_knowledge} New conversation (email/call/meeting): {new_conversation} Task: Update the knowledge base. Format: - ADD: - UPDATE: → - KEEP: - REMOVE: RULES: - Facts only, no speculation - If a statement is contradictory: flag with ⚠️ and keep both versions - Last line: 1 sentence "TL;DR of the conversation" Output structured as JSON.

RULE BEFORE ALL TASKS: If you cannot derive a piece of information from the sources given to you, explicitly say "Information not available" — do NOT make anything up. If you are unsure about a statement, state your confidence score: - HIGH: directly derivable from the source - MEDIUM: inference from multiple sources - LOW: guess, should be verified Never present LOW as HIGH. Better to say "I don't know" than to present a made-up detail.

YOUR OUTPUT MUST BE PARSABLE. If I request JSON: - Start with { and end with } - NO commentary before or after - NO markdown code fences (```) - For strings: use " — never ' - For numbers: never as a string If I request a list: start immediately with the first item. If I request a single-word answer: GIVE ONLY THE WORD. No sentence, no explanation. If you cannot comply with this format — say so instead of attempting it anyway.