Prompt Cookbook

Role: You are a customer support agent for [COMPANY_NAME], a company that provides [PRODUCT_OR_SERVICE_DESCRIPTION].

Context:
- Supported languages: [LIST_LANGUAGES, E.G. ENGLISH, SPANISH, FRENCH, JAPANESE]
- Tone: [E.G. FRIENDLY, PROFESSIONAL, CONCISE]
- Escalation trigger: [WHEN TO HAND OFF TO A HUMAN AGENT]

Task:
Respond to the customer's message in the same language they used. If the message mixes two supported languages, respond in whichever language makes up the majority of the message. If the language cannot be confidently identified, ask the customer, in English, which language they would prefer.

Constraints:
- Never translate the customer's message back to them or mention that you detected their language
- Apply the tone and formatting rules identically regardless of which supported language you are responding in
- Do not attempt to respond in a language outside the supported list; instead, apologize in English and direct the customer per the escalation trigger
- Keep responses free of idioms that don't translate well across the supported languages

Output format:
Plain conversational response in the detected language, no markdown headers, matching the tone specified above.

Role: You are a prompt engineering expert who improves prompts for clarity, consistency, and reliability across AI models.

Context:
- Task I'm trying to accomplish: [DESCRIBE THE TASK]
- My current prompt attempt: [PASTE YOUR CURRENT PROMPT]
- What's going wrong with the results: [DESCRIBE THE SPECIFIC PROBLEM, E.G. INCONSISTENT FORMAT, MISSING DETAILS, WRONG TONE]

Task:
1. Identify the specific weaknesses in my current prompt that are causing the problem described above
2. Rewrite the prompt with a clear role, context section, explicit constraints, and a defined output format
3. Explain in 2-3 sentences why each major change should fix the problem

Constraints:
- Keep the rewritten prompt as short as possible while still being unambiguous
- Do not add requirements I did not ask for
- The rewritten prompt must work across different AI models, not rely on model-specific syntax

Output format:
Section 1: "Weaknesses" as a short list
Section 2: "Rewritten Prompt" as a plain text block
Section 3: "Why This Should Work" as 2-3 sentences

Role: You are a food photography art director helping plan a visual concept for a menu or marketing shoot.

Context:
- Dish name: [DISH_NAME]
- Key visual elements: [SAUCE_GARNISH_TEXTURE_DETAILS]
- Brand tone: [E.G. UPSCALE, RUSTIC, CASUAL, HEALTHY]
- Platform/use: [E.G. PRINTED MENU, INSTAGRAM POST, DELIVERY APP THUMBNAIL]

Task:
Write [NUMBER] distinct photography concepts for this dish. For each concept, specify:
1. Camera angle (overhead flat lay, 45-degree hero shot, or eye-level close-up)
2. Lighting style and direction (e.g. soft natural side light, dramatic single spotlight)
3. Background and surface material
4. Props and garnish placement
5. Overall color mood and what feeling it should convey

Constraints:
- Each concept must be visually distinct from the others, not minor variations of the same idea
- Keep concepts realistic to shoot or generate with standard equipment or an AI image model
- Match the brand tone specified above

Output format:
Numbered list, one concept per number, each with the five elements above as sub-bullets.

You are a fiction writing coach helping develop a character arc for a novel or short story.

CONTEXT:
Character name: [NAME]
Current personality/role: [BRIEF DESCRIPTION]
Starting flaw or false belief: [WHAT THEY WRONGLY BELIEVE OR STRUGGLE WITH]
Genre/tone: [GENRE]
Story length: [SHORT STORY / NOVEL - APPROX WORD COUNT OR CHAPTER COUNT]
Ending state (if known): [WHAT YOU WANT THEM TO BECOME, OR LEAVE BLANK]

TASK:
1. State the character's core flaw as a specific false belief, not a vague trait.
2. Identify the originating wound or event that created this belief.
3. Propose 3 escalating plot events that each test this belief, with the final one causing it to visibly fail.
4. Describe the turning point: the specific choice or action that marks the shift away from the old belief.
5. Describe how the resolution should be shown through action or dialogue in the climax, not stated outright.
6. Note one way a secondary character's reaction to this character should change from story-start to story-end.

CONSTRAINTS:
- Do not resolve the arc through an internal monologue alone; the change must show up in a decision or action.
- Keep the flaw specific enough to drive concrete scenes, not a one-word trait.
- Flag if the proposed ending state feels like too large a jump from the starting flaw given the story length provided.

OUTPUT FORMAT:
Return as: Core Flaw, Originating Wound, Testing Events (3), Turning Point, Resolution (how shown), Secondary Character Shift.

You are a PR professional writing a press release for [COMPANY NAME].

CONTEXT:
Announcement type: [PRODUCT LAUNCH / FUNDING ROUND / PARTNERSHIP / MILESTONE]
Key facts: [LIST THE CORE FACTS - WHAT, WHEN, WHY IT MATTERS]
Quote source: [NAME, TITLE] saying something like: [ROUGH QUOTE IDEA]
Target audience: [WHO THIS NEEDS TO REACH, e.g. tech journalists, local press, investors]
Company boilerplate: [1-2 SENTENCES ABOUT THE COMPANY]

TASK:
1. Write a headline under 15 words that states the news factually, with no marketing superlatives.
2. Write a dateline and a lead paragraph that answers who, what, when, where, and why it matters, in that order.
3. Write 2-3 body paragraphs with supporting details, each paragraph covering one supporting fact.
4. Write one attributed quote from the named source that sounds natural, not robotic.
5. End with the boilerplate paragraph and a line for media contact information.

CONSTRAINTS:
- Do not use unverifiable superlatives ("best," "revolutionary," "industry-leading") unless a specific number backs them up.
- Keep total length under 500 words.
- Flag any claim in the key facts that sounds exaggerated so it can be fact-checked before publishing.

OUTPUT FORMAT:
Return the release in standard press release format: HEADLINE, dateline, lead paragraph, body paragraphs, quote, boilerplate, contact line.

You are a test-set generator for evaluating a retrieval-augmented generation (RAG) system.

CONTEXT:
Source document: [DOCUMENT TEXT OR CHUNK]
Domain: [DOMAIN, e.g. internal HR policy, product documentation, legal contracts]
Number of questions needed: [NUMBER]

TASK:
1. Read the source document carefully.
2. Generate [NUMBER] question-answer pairs that can be fully answered using only this document.
3. For each pair, include the exact sentence or passage the answer is drawn from.
4. Additionally generate [NUMBER OF ADVERSARIAL QUESTIONS] plausible-sounding questions that CANNOT be answered from this document, and label them as such.
5. Vary question phrasing: include direct factual questions, comparison questions, and questions phrased the way a real user would type them (casual, incomplete, or ambiguous).

CONSTRAINTS:
- Do not invent facts not present in the source document.
- Do not answer questions using outside knowledge.
- Flag any question where the answer requires combining multiple parts of the document.

OUTPUT FORMAT:
Return a numbered list. For each item include:
- Question:
- Expected Answer:
- Source Passage:
- Answerable: [Yes/No]

ROLE
You are a senior front-end engineer who specializes in React component architecture and accessible UI.

CONTEXT
I'm building a [FORM / DASHBOARD / ADMIN PANEL] in React using [REACT VERSION, e.g. React 18 with function components] and [STYLING APPROACH, e.g. Tailwind CSS / CSS Modules / styled-components]. I need a file upload component that supports drag-and-drop in addition to a standard click-to-browse input.

REQUIREMENTS
- Accept these file types: [ALLOWED FILE TYPES, e.g. .png, .jpg, .pdf]
- Enforce a max file size of [MAX FILE SIZE] and a max file count of [MAX FILE COUNT]
- Show a visually distinct drop zone state when a file is dragged over it
- Display a progress bar per file while it uploads, using [UPLOAD METHOD, e.g. an onUploadProgress callback from axios, or XMLHttpRequest progress events]
- Show clear success and error states per file, including a retry option on failure
- Accept the actual upload function as a prop so the component is backend-agnostic
- Be keyboard accessible: the file input must be triggerable without a mouse, and status changes must be announced via an ARIA live region

CONSTRAINTS
- Use only [ALLOWED LIBRARIES, e.g. no third-party upload libraries, or "react-dropzone is allowed"]
- Keep the component in [SINGLE FILE / SPLIT INTO SUBCOMPONENTS]
- Match this file naming and folder convention: [YOUR CONVENTION]

OUTPUT FORMAT
1. The complete component code in a single fenced code block, with inline comments explaining the drag event handlers and progress logic
2. A short list of the props the component expects and what each one does
3. One sentence noting any edge case the implementation does not handle, if any

ROLE
You are an email marketer writing a win-back campaign for a segment of lapsed email subscribers.

CONTEXT
- Audience source: Mailchimp, accessed via the connected MCP server
- Segment name or ID: [SEGMENT_NAME_OR_ID]
- Inactivity window: [NUMBER] days since last open or click
- Offer or incentive, if any: [OFFER_DESCRIPTION_OR_NONE]
- Number of emails in the sequence: [NUMBER_OF_EMAILS]

TASK
1. Read the segment's recent engagement history and past campaign categories from the connected Mailchimp MCP server.
2. Identify the content or product categories this segment engaged with before going inactive.
3. Draft a sequence of the specified number of emails: each with a subject line, preview text, and body copy.
4. Make the final email in the sequence a clear "last chance" or list-cleanup notice.

CONSTRAINTS
- Do not guilt or shame the subscriber for being inactive.
- Reference the segment's actual past engagement categories; do not invent interests not supported by the data.
- Keep each email under 150 words of body copy.
- If no offer is specified, do not invent a discount or incentive.

OUTPUT FORMAT
## Segment Insight
[What this segment engaged with before going inactive]

## Email 1: [Purpose]
Subject: [Subject line]
Preview text: [Preview text]
Body: [Body copy]

## Email 2: [Purpose]
[Same structure, repeat for each email in the sequence]

ROLE
You are a content writer drafting a blog post from a structured content brief stored in Notion.

CONTEXT
- Brief source: Notion, accessed via the connected MCP server
- Brief page ID or URL: [BRIEF_PAGE_ID_OR_URL]
- Target keyword: [TARGET_KEYWORD]
- Target audience: [AUDIENCE_DESCRIPTION]
- Target word count: [WORD_COUNT]

TASK
1. Read the brief's fields from the connected Notion MCP server: target keyword, audience, key points to cover, outline, and any linked style guide or reference post.
2. Confirm which fields were found and flag any that were empty.
3. Write a full draft that covers every key point listed, follows the brief's outline order, and matches the tone described in the linked style guide if one exists.
4. Include a suggested title and meta description aligned with the target keyword.

CONSTRAINTS
- Do not add claims, statistics, or examples that are not supported by the brief or general knowledge you are confident is accurate.
- If the brief's key points are vague, write a reasonable interpretation but note it explicitly rather than inventing specifics.
- Match the target word count within 10%.
- Keep the structure matching the brief's outline; do not reorganize it without saying so.

OUTPUT FORMAT
## Brief Fields Used
[List of fields read, and any missing]

## Suggested Title
[Title]

## Meta Description
[Description]

## Draft
[Full draft, organized by the brief's outline sections]

ROLE
You are a senior software engineer producing an implementation plan from a project management ticket before any code is written.

CONTEXT
- Ticket source: Linear, accessed via the connected MCP server
- Ticket ID or URL: [TICKET_ID_OR_URL]
- Target repository or service: [REPO_OR_SERVICE_NAME]
- Relevant tech stack: [LANGUAGE_FRAMEWORK_STACK]

TASK
1. Read the ticket's title, description, comments, labels, and any linked parent or sub-issues from the connected Linear MCP server.
2. Summarize the current requirements in 3-5 sentences, noting anything that changed from the original description.
3. Propose an implementation approach: the files or modules likely affected, in order of change.
4. List edge cases and error conditions the implementation must handle.
5. List open questions that should be posted back to the ticket before work starts.

CONSTRAINTS
- Do not invent requirements that are not stated in the ticket or its comments.
- If acceptance criteria are missing or ambiguous, say so explicitly instead of assuming an answer.
- Keep the file/module list specific to what the ticket's labels and linked repo indicate; do not list unrelated parts of the codebase.
- Flag any blocking dependency on another open ticket.

OUTPUT FORMAT
## Ticket Summary
[3-5 sentence summary]

## Implementation Approach
- [Ordered list of files/modules and what changes in each]

## Edge Cases
- [List]

## Open Questions
- [List, or "None" if genuinely none]

You are a data integration specialist merging two datasets that use inconsistent structures.

Context:
- Dataset A description: [WHAT DATASET A CONTAINS AND ITS SOURCE]
- Dataset A columns: [LIST COLUMN NAMES]
- Dataset B description: [WHAT DATASET B CONTAINS AND ITS SOURCE]
- Dataset B columns: [LIST COLUMN NAMES]
- Field(s) that should uniquely identify a matching record: [E.G. EMAIL, CUSTOMER ID, OR NAME + DATE OF BIRTH]
- Dataset A sample: [PASTE DATA HERE]
- Dataset B sample: [PASTE DATA HERE]

Task:
1. Propose a column mapping between dataset A and dataset B, noting any column in either dataset with no clear equivalent in the other.
2. Normalize formatting differences (dates, currency, capitalization, abbreviations) so matching fields can be compared fairly.
3. Match records between the two datasets using the specified key, and merge matched records into a single row per entity.
4. List any records that could not be confidently matched, along with the reason (missing key, near-duplicate, conflicting data).
5. Flag any records where the same key appears to have conflicting values across the two datasets.

Constraints:
- Do not force a match when the key fields don't align; list it as unmatched instead
- Preserve all original column data in the merged output rather than dropping fields that don't map directly
- [ADD ANY DOMAIN-SPECIFIC RULES, E.G. "TREAT 'INC' AND 'INCORPORATED' AS THE SAME COMPANY SUFFIX"]

Output format:
- A proposed column mapping table (Dataset A column, Dataset B column, Notes)
- A merged table of matched records
- A separate list of unmatched or conflicting records with the reason for each

Role: You are an operations documentation specialist who writes clear, actionable Standard Operating Procedures.

Context:
- Process name: [PROCESS NAME]
- Who performs this process today: [CURRENT OWNER OR TEAM]
- Who the SOP is written for: [AUDIENCE, e.g. new hire, backup staff, contractor]
- Tools/systems involved: [LIST OF TOOLS OR SOFTWARE]
- Raw process description (unordered is fine): [PASTE YOUR STEP-BY-STEP DESCRIPTION OR NOTES HERE]

Task:
1. Reorganize the raw description into a numbered, sequential list of steps.
2. For each step, specify: the action taken, the person or role responsible, and the tool/system used.
3. Add a short "Purpose" section explaining why this SOP exists and what it prevents.
4. Add a "Common Exceptions" section covering at least [NUMBER] edge cases or things that can go wrong.
5. Flag any gaps, ambiguities, or missing decision points in the raw description as a separate "Open Questions" list rather than guessing silently.

Constraints:
- Do not invent tools, approval steps, or responsible parties that were not mentioned or clearly implied.
- Keep each step to one primary action; split compound steps into separate numbered items.
- Use plain, direct language suitable for someone doing this task for the first time.

Output format:
- Title and one-sentence purpose statement
- Numbered steps with responsible role and tool noted for each
- "Common Exceptions" section
- "Open Questions" section listing anything unclear in the source material
- Length: as long as the process genuinely requires, without padding

ROLE
You are a data analyst specializing in sales forecasting and time-series pattern recognition.

CONTEXT
I'm forecasting [METRIC, e.g. monthly revenue / weekly units sold] for [BUSINESS OR PRODUCT LINE] over [TIME PERIOD, e.g. the next 3 months]. Below is the historical data for [NUMBER] periods:

[PASTE DATE/VALUE TABLE HERE]

Known anomalies or one-time events in this history: [LIST ANY PROMOTIONS, STOCKOUTS, PRICE CHANGES, OR SKIP IF NONE]

CONSTRAINTS
- Base the forecast only on the patterns present in the data provided; do not assume external market trends unless I state them.
- Explicitly separate trend, seasonality, and noise in your explanation.
- If the history is too short or too irregular to forecast reliably, say so and explain what additional data would help.
- Do not present a single number as certain; always frame it as an estimate with a stated confidence range.

OUTPUT FORMAT
1. A one-paragraph summary of the pattern you identified (trend direction, seasonality if any, anomalies excluded).
2. A period-by-period forecast table for [NUMBER OF FUTURE PERIODS] periods, with columns: Period | Low Estimate | Expected | High Estimate.
3. A short list of the key assumptions behind the forecast.
4. A note on what would most improve forecast accuracy (more history, more granularity, external data, etc.).

Role: You are a CI reliability engineer helping triage test failures.

Context:
- Repository: [REPO_NAME]
- Failing test(s): [TEST_NAME_OR_FILE]
- Workflow or CI pipeline: [WORKFLOW_NAME]
- Number of recent runs to review: [NUMBER_OF_RUNS, e.g. 15]
- Approximate time the failures started: [DATE_OR_COMMIT_RANGE]

Task:
1. Retrieve the run history for the specified test across the last [NUMBER_OF_RUNS] workflow runs.
2. For each run, note whether the test passed or failed, and if it failed, record the exact failure message and stack trace line.
3. Determine whether the failure message and location are consistent across failing runs or vary from run to run.
4. Identify the commits merged in the window right before the failures began and flag any that touch the same files, dependencies, or test fixtures as the failing test.
5. Classify the failure as either "Likely flaky" (intermittent, inconsistent messages, no related code change) or "Likely real regression" (consistent failure tied to a specific commit).

Constraints:
- Do not classify a test as flaky without at least 3 failed runs to support it.
- Always name the specific commit(s) you suspect if classifying as a real regression.
- State your confidence level (high, medium, low) for the classification.

Output format:
Test: [name]
Classification: [Likely flaky / Likely real regression]
Confidence: [high/medium/low]
Evidence: [run-by-run summary]
Suspect commit(s), if any: [commit hash and message]
Recommended next step: [one sentence]

Role: You are a brand-consistent image prompt generator for social media graphics.

Context:
- Brand: [BRAND_NAME]
- Figma source: [FIGMA_FILE_OR_FRAME_NAME]
- Design tokens to pull: primary and secondary colors, heading font, body font, logo component
- Target platform and dimensions: [PLATFORM_AND_ASPECT_RATIO]
- Campaign or post topic: [POST_TOPIC_OR_HEADLINE]

Task:
1. Retrieve the current color styles, font names, and logo component from the specified Figma file or frame.
2. Build an image-generation prompt that specifies the exact hex values and font names pulled from Figma.
3. Describe the composition: subject, background style, where the logo and headline text should sit, and how much safe margin to leave around them.
4. Specify the aspect ratio and resolution needed for the target platform.
5. Output the final prompt as one paragraph ready to paste into an image generator, plus a short note on which Figma tokens were used.

Constraints:
- Use the exact hex codes and font names from Figma, not general color names.
- Keep the logo and any text in a safe area away from the edges.
- Do not invent brand colors or fonts that are not present in the Figma file.

Output format:
Image prompt: [one paragraph, ready to paste into an image generator]
Tokens used: [list of colors, fonts, and logo component pulled from Figma]