Mistral Prompt Templates (2026): Fill-in-the-Blank

Draft a [EMAIL / MESSAGE] from me ([MY ROLE]) to [RECIPIENT] about [PURPOSE]. Key points to land: [POINTS]. Desired outcome: [OUTCOME]. Tone: [TONE]. Keep it under [LENGTH] and end with one clear ask. Output: subject line plus body, ready to send, and one slightly [warmer / firmer] alternative.

Turn this [TRANSCRIPT / NOTES] into structured meeting notes: [PASTE]. Output three sections: Decisions made, Action items (a table with Owner, Task, Due date), and Open questions. Do not invent details, flag anything unresolved, and add a two-line recap email I can send to attendees.

You are a [ROLE, e.g. strategy advisor] for a [COMPANY TYPE]. We are deciding whether to [DECISION]. Context: [KEY FACTS AND CONSTRAINTS]. Use Work mode to reason it through. Lay out the options, the main trade-offs, the biggest risk, and what you would do and why. Assume a [TIME/BUDGET] constraint. Output: a one-page brief — recommendation, rationale, risks, next step.

You are a [TYPE] writer for [AUDIENCE]. Write a [LENGTH]-word [FORMAT, e.g. how-to article] on [TOPIC]. Angle: [SPECIFIC TAKE]. Tone: [TONE]. Lead each section with the takeaway, use short paragraphs and one example per section, and avoid [BANNED WORDS/CLICHES]. Draft it in Canvas. Output: the finished piece with H2 subheadings, no preamble.

Write [NUMBER] [PLATFORM] posts about [TOPIC] for [AUDIENCE]. Goal: [GOAL, e.g. drive signups / spark replies]. Voice: [VOICE]. Each post must stand alone, lead with a hook, and stay under [CHARACTER LIMIT]. No hashtag spam. Output: numbered posts, each ready to paste, with the strongest one marked.

Rewrite the text below to be [GOAL, e.g. clearer / shorter / more persuasive] for [AUDIENCE]. Keep every fact, cut filler, and match a [TONE] tone. Target [LENGTH]. Do not add new claims.

[PASTE TEXT]

Output: the rewrite, then one line on what you changed and why.

You are a senior [LANGUAGE] engineer. Write a [FUNCTION / COMPONENT] that [WHAT IT DOES]. Inputs: [INPUTS]. Output: [EXPECTED OUTPUT]. Constraints: [e.g. no external deps, handle empty input, O(n) time]. Include comments only where non-obvious and [NUMBER] unit tests covering edge cases. Output: one code block, then a two-line usage example.

Think step by step. Here is [LANGUAGE] code that should [INTENDED BEHAVIOR] but instead [ACTUAL BEHAVIOR / ERROR]:

[PASTE CODE]

Find the root cause, explain why it happens in plain terms, and give the corrected code. List any other bugs you spot. Output: root cause, fixed code block, then a bullet list of remaining risks.

Refactor this [LANGUAGE] code to be [GOAL, e.g. more readable / faster / idiomatic] while keeping behavior identical. Follow [STYLE OR PATTERN, e.g. PEP 8, our naming convention]. Do not change the public interface or add dependencies. Explain each change in one line.

[PASTE CODE]

Output: the refactored code block, then a short changelog.

You are a [DOMAIN] research analyst. Search the web and answer: [QUESTION]. Pull from at least [NUMBER] independent, recent sources and cite every one with a link and date. Distinguish established facts from contested claims. Output: a direct answer first, then supporting evidence, then a "confidence and gaps" note.

Search the web and compare [OPTION A] vs [OPTION B] vs [OPTION C] for [USE CASE / BUYER]. Evaluate on [CRITERIA, e.g. price, features, support, lock-in]. Cite current pricing and specs with links and dates. Recommend one and explain the trade-off. Output: a comparison table, then a one-paragraph recommendation for [BUDGET / CONSTRAINT].

Using the code interpreter, analyze the attached [DATA / FILE] for [AUDIENCE]. Answer: [SPECIFIC QUESTIONS]. Pull out the [NUMBER] most important findings, quantify where possible, and flag any data-quality issue (missing values, outliers, duplicates). No speculation beyond the data. Output: headline finding, key numbers, then a short "so what" for [AUDIENCE].

Translate the text below from [SOURCE LANGUAGE] into [TARGET LANGUAGE] for a [AUDIENCE / REGION]. Localize idioms, units, currency, and tone so it reads natively — do not translate word-for-word. Keep names and product terms exact. Register: [formal / informal]. Output: the translation, then a short note on any choice you made. Text: [PASTE TEXT]

Write a [formal business email] in [LANGUAGE] from [MY ROLE] to [RECIPIENT] about [PURPOSE]. Match the professional register and courtesy conventions native to [COUNTRY], not a translated English tone. Keep it under [LENGTH], end with one clear ask, and add an English gloss below so I can check it.

Be my [LANGUAGE] tutor at a [LEVEL] level. Have a natural conversation about [TOPIC]. Keep your replies short, and after each of my messages, correct my mistakes in a separate line (what I wrote, the fix, a one-line reason). Stay in [LANGUAGE] but explain corrections in English. Start now.

You are an extraction function. From the text, return valid JSON only — no prose, no fences — using exactly this schema: {[KEY1]: [TYPE], [KEY2]: [TYPE], [KEY3]: [TYPE]}. Use null for any missing field, do not invent values, and if the text is empty return {}. Text: [PASTE TEXT]

You are a classifier. For each input, return valid JSON only matching: {"category": one of [LIST CATEGORIES], "priority": one of ["low","medium","high"], "summary": string (max 15 words)}. No prose, no fences. If an input is empty, return {"category":"other","priority":"low","summary":"empty input"}. Inputs: [PASTE ITEMS]

Help me build a reusable agent in Le Chat's Agent Builder that [WHAT IT DOES]. Write: the agent's system instructions, the inputs it expects, the tools it needs ([web search / code / connectors]), the output format, and [NUMBER] test inputs with expected output. Keep the instructions specific enough that it behaves the same every run.