AI 에이전트에 프롬프트를 작성하는 것은 챗봇에 프롬프트를 작성하는 것과 근본적으로 다릅니다. 챗봇에서는 대화를 나누는 반면, 에이전트에서는 직무 기술서를 작성하는 것입니다. 에이전트는 목표, 제약 조건, 기대하는 결과물, 그리고 문제가 발생했을 때 어떻게 대처할지 이해해야 합니다. 사용자가 매 단계마다 안내할 수 없기 때문입니다.
이 15개의 템플릿은 에이전트 사용에 최적화되어 있습니다. 명확한 목표, 구체적인 제약 조건, 정의된 출력 형식, 그리고 오류 처리 지침을 포함하고 있습니다. Hermes Agent, Claude Code, OpenAI Codex, ChatGPT의 다단계 모드에서 사용할 수 있습니다.
핵심 요약
에이전트 프롬프트에는 챗봇 프롬프트와 다른 세 가지 요소가 필요합니다. 명확한 성공 기준("작업이 완료되었는지 어떻게 알 수 있는가?"), 구체적인 제약 조건("/src 외부의 파일은 수정하지 마세요"), 그리고 오류 처리("API가 다운되면 3번 재시도한 후 오류를 보고하세요")입니다. 이러한 요소가 없으면 에이전트는 토큰을 낭비하고 예측할 수 없는 결과를 생성합니다.
리서치 템플릿
1
경쟁사 리서치
Research the top [5-10] competitors to [your product/company] in [market].
For each competitor, find:
- Company name, URL, founding year
- Pricing (all tiers)
- Key features (top 5)
- Target audience
- Recent news or product updates (last 3 months)
Output: A comparison table with all competitors side by side.
Format: Markdown table.
Success: Table has complete data for all competitors with sources cited.
If a competitor's pricing isn't public, note "pricing not public" rather than guessing.
2
시장 동향 분석
Research [topic/industry] trends for [time period].
Find:
- 5 major trends with supporting data points
- 3 expert quotes or predictions (with attribution)
- 2 contrarian viewpoints
- Key statistics with sources
Output: A 1,000-word briefing document.
Structure: Executive summary (3 bullets) → Trends → Expert perspectives → Data → Contrarian views.
Success: Every claim has a cited source. No unsourced statistics.
3
개인/회사 리서치 브리프
Research [person name] at [company].
Find:
- Current role, tenure, previous roles
- Recent public statements, interviews, or posts (last 6 months)
- Their company's recent news and strategic direction
- Mutual connections or shared interests with me: [your background]
- 3 conversation starters specific to their work
Output: A 1-page brief I can review in 5 minutes before a meeting.
Success: No fabricated claims. If information isn't publicly available, say so.
📬 이 콘텐츠가 도움이 되셨나요? 에이전트 관련 가이드를 매주 발행합니다. 이메일로 받아보세요 →
---자동화 템플릿
4
일일 요약 리포트
Generate a daily summary covering:
1. Top 3 news items in [industry/topic] from the last 24 hours
2. Any mentions of [your company/competitors] in news or social media
3. Key metrics: [list your metrics sources]
Format: Bullet points, under 300 words total.
Tone: Factual, no commentary. Just the information.
Schedule: Run at 8:00 AM daily.
Delivery: Post to [Slack channel / email / Discord].
5
이메일 분류
Review my unread emails and categorize them:
Categories:
- URGENT: Needs response within 2 hours (client issues, boss requests, deadlines)
- TODAY: Should respond today but not immediately
- THIS WEEK: Can wait a few days
- FYI: No response needed, just information
- SKIP: Newsletters, promotions, automated notifications
For each URGENT and TODAY email:
- One-sentence summary of what they need
- Suggested response (2-3 sentences)
Output: Categorized list with summaries.
Constraint: Do NOT send any responses — only draft suggestions for my review.
6
콘텐츠 재활용
Take this article: [paste URL or content]
Create:
1. A LinkedIn post (150-200 words, hook + value + question)
2. A Twitter/X thread (5-7 tweets, first tweet is the hook)
3. A Reddit post for r/[subreddit] (educational, no self-promotion in body)
4. 3 short-form video script hooks (under 15 seconds each)
Constraints:
- No links in any version (kills reach)
- Each version should feel native to its platform
- LinkedIn: professional, uses line breaks
- Twitter: punchy, uses numbers
- Reddit: humble, leads with value
- Video hooks: pattern interrupt opening
코딩 템플릿
7
기능 구현
Implement [feature description].
Requirements:
- [Requirement 1]
- [Requirement 2]
- [Requirement 3]
Constraints:
- Only modify files in /src and /tests
- Follow existing code style and patterns
- Add tests for every new function
- Do not modify any database schemas
Steps:
1. Read the relevant existing code first
2. Plan the changes before making them
3. Implement
4. Run tests
5. Fix any failures
6. Report what was changed and why
Success: All tests pass, feature works as described, no unrelated files modified.
8
버그 조사
Bug report: [describe the bug]
Expected behavior: [what should happen]
Actual behavior: [what actually happens]
Steps:
1. Reproduce the bug by reading the relevant code
2. Identify the root cause (not just the symptom)
3. Propose a fix with explanation
4. Implement the fix
5. Add a test that would have caught this bug
6. Run the full test suite to verify no regressions
Constraint: Do not refactor unrelated code. Fix only the bug.
Report: What caused it, what you changed, and what test you added.
분석 템플릿
9
데이터 분석 리포트
Analyze this dataset: [attach file or describe location]
Steps:
1. Describe the dataset: columns, row count, date range, missing values
2. Calculate [specific metrics: MoM growth, averages, distributions]
3. Identify the top 3 trends or anomalies
4. Create visualizations for key findings
5. Write a 3-paragraph executive summary
Output: Report with summary at top, analysis in middle, raw data notes at bottom.
Audience: [Non-technical executive / Technical team / Client]
Success: Every number is traceable to the source data. No made-up statistics.
10
문서 검토
Review this document: [attach file]
Check for:
1. Factual accuracy — flag any claims that seem incorrect or unsourced
2. Consistency — are there contradictions between sections?
3. Completeness — what important topics are missing?
4. Clarity — which paragraphs are confusing or ambiguous?
5. Actionability — does each section lead to a clear next step?
Output: Annotated feedback organized by section.
Format: Section heading → Issue → Suggestion → Priority (High/Medium/Low)
Constraint: Be specific. "This section is unclear" is not useful. "Paragraph 3 says 'significant growth' without a number — add the specific percentage" is useful.
전략 템플릿
11
의사결정 매트릭스
Help me decide between these options: [list 3-5 options]
Context: [your situation, constraints, priorities]
Steps:
1. Identify the 5 most important criteria for this decision
2. Weight each criteria by importance (total = 100%)
3. Score each option on each criteria (1-10)
4. Calculate weighted scores
5. Provide a recommendation with reasoning
Output: Decision matrix table + 1-paragraph recommendation.
Constraint: Be honest about trade-offs. Don't make one option look perfect — every choice has downsides.
12
미팅 준비 브리프
I have a meeting about [topic] with [who] in [timeframe].
Prepare:
1. 3 key points I should make (with supporting data)
2. 5 questions they might ask me (with prepared answers)
3. 2 questions I should ask them
4. 1 potential objection and how to address it
5. A 30-second opening statement
My goal for this meeting: [specific outcome]
Constraint: Keep the total brief under 1 page. I need to review it in 5 minutes.
크리에이티브 템플릿
13
콘텐츠 캘린더
Create a 30-day content calendar for [platform] about [topic/niche].
Requirements:
- 1 post per day
- Mix: 40% educational, 30% personal story, 20% promotional, 10% engagement
- Each post: headline, hook (first line), key point, CTA
- Include trending topic tie-ins where natural
- No two consecutive days with the same content type
My audience: [describe]
My product/service: [describe]
Constraint: Every promotional post must deliver standalone value even without clicking the CTA.
14
랜딩 페이지 카피
Write landing page copy for [product/service].
Target customer: [describe]
Main problem solved: [describe]
Key differentiator: [what makes this different]
Sections needed:
1. Hero: headline (under 10 words) + subheadline + CTA button text
2. Problem section: 3 pain points the customer recognizes
3. Solution section: how the product solves each pain point
4. Social proof: placeholder for testimonials (write 3 example testimonials)
5. Features: top 5 with one-sentence benefits
6. FAQ: 5 questions (answer objections, not features)
7. Final CTA: urgency-driven close
Tone: [professional/casual/bold]
Constraint: No "revolutionary," "game-changing," "leverage," or "unlock."
15
고객 피드백 분석
Analyze this customer feedback: [paste reviews, surveys, or support tickets]
Steps:
1. Categorize each piece of feedback (feature request, bug, praise, complaint)
2. Identify the top 5 themes by frequency
3. Sentiment analysis: what percentage is positive/neutral/negative?
4. Extract specific, actionable improvement suggestions
5. Identify any urgent issues that need immediate attention
Output: Summary report with theme frequency table + prioritized action items.
Success: Every action item maps to specific customer feedback (quote the relevant text).
이 템플릿들을 특정 사용 사례에 맞게 개선하려면 무료 프롬프트 옵티마이저에 붙여넣으세요. 맥락, 제약 조건, 세부 사항을 추가하여 에이전트 출력의 신뢰성을 높여줍니다.
---📬 이런 콘텐츠를 더 원하시나요? 에이전트 전용 템플릿과 가이드, 매주 제공합니다. 무료 구독하기 →
---자주 묻는 질문
이 템플릿들은 모든 AI 에이전트에서 작동하나요?
네. (목표 + 제약 조건 + 출력 형식 + 성공 기준) 구조는 에이전트에 구애받지 않습니다. 에이전트의 구체적인 기능은 다를 수 있습니다. Hermes는 메시징 통합을, Claude Code는 코드베이스 접근을, ChatGPT는 웹 리서치를 처리합니다. 그러나 프롬프트 구조는 그대로 적용됩니다.
에이전트 프롬프트에 오류 처리 지침이 필요한 이유는 무엇인가요?
에이전트는 자율적으로 행동하기 때문입니다. 챗봇은 응답을 보여주고 사용자가 다음 단계를 결정합니다. 하지만 에이전트는 응답에 따라 행동을 취합니다. 행동이 실패하고 오류 처리 지침이 없다면, 에이전트는 멈추거나(시간 낭비) 잘못된 방식으로 재시도합니다(비용 낭비).
템플릿을 조합할 수 있나요?
네. 복잡한 워크플로는 종종 템플릿을 연결합니다. 리서치 → 분석 → 콘텐츠 생성. 일부 에이전트(Hermes, CrewAI)는 연결된 작업을 기본적으로 처리합니다. ChatGPT의 경우, 복합 작업을 순차적인 프롬프트로 분리하세요.
고지 사항: 본문의 일부 링크는 제휴 링크입니다. 우리는 직접 테스트하고 정기적으로 사용하는 도구만 추천합니다. 자세한 내용은 고지 정책을 참조하세요.