22 Kimi Prompts for Agent Mode & OK Computer (2026)

Goal: build a high-converting landing page for [PRODUCT], a [ONE-LINE DESCRIPTION] aimed at [TARGET AUDIENCE].
Deliverable: a single responsive HTML page (mobile-first) I can preview and download, with hero, benefits, social proof, FAQ, and one clear call-to-action to [CTA / LINK].
Context: brand tone is [TONE], primary color [HEX], key differentiators are [LIST].
Constraints: write real, specific copy (no lorem ipsum), keep it accessible, and finish with a short summary of the layout choices you made. Ask me nothing you can reasonably decide yourself.

Goal: design and build a small marketing website for [COMPANY / PROJECT].
Deliverable: a linked multi-page site — Home, [PAGE 2], [PAGE 3], Contact — as responsive HTML/CSS I can download, with consistent navigation and footer.
Context: audience is [AUDIENCE]; value proposition is [VALUE PROP]; include these sections/content: [LIST].
Constraints: coherent visual system (typography, spacing, color [HEX]), real copy on every page, and working internal links. Plan the sitemap first, show it to me, then build. End with a list of what a developer would need to change to go live.

Goal: recreate a landing page in the style of [REFERENCE URL or uploaded screenshot], adapted for [MY PRODUCT].
Deliverable: one responsive HTML page that matches the reference's structure and feel but uses my content and brand.
Context: my product is [DESCRIPTION]; keep [WHAT TO KEEP]; change [WHAT TO CHANGE].
Constraints: do not copy their exact text or logos — reproduce the layout pattern and rewrite everything for me. Note any section where you deviated from the reference and why.

Goal: build an investor pitch deck for [STARTUP], raising [ROUND] to [USE OF FUNDS].
Deliverable: a 10–12 slide deck (Slides) I can present, covering problem, solution, market size, product, traction, business model, competition, team, and the ask.
Context: what we do is [DESCRIPTION]; traction so far is [METRICS]; market is [MARKET].
Constraints: one clear idea per slide, concrete numbers over adjectives, speaker-note under each slide. Where you lack a number, insert a [PLACEHOLDER] and flag it rather than inventing data. Summarize the story arc at the end.

Goal: turn the uploaded [DOCUMENT/REPORT] into a clear explainer deck for [AUDIENCE].
Deliverable: a Slides deck of [N] slides that teaches the key ideas in order, with a simple diagram or chart where it helps.
Context: the audience knows [LEVEL] about this topic; the goal of the talk is [OBJECTIVE].
Constraints: pull only from the document, cite the section on any data slide, and keep each slide to one takeaway plus supporting detail. End with a 3-bullet "so what" summary slide.

Goal: build a 3-year financial model for [BUSINESS], a [BUSINESS TYPE].
Deliverable: a Sheet with tabs for Assumptions, Revenue, Costs, and a monthly P&L that rolls up to annual, plus a summary tab with the headline numbers and a simple chart.
Context: revenue drivers are [DRIVERS]; pricing is [PRICING]; main costs are [COSTS]; starting assumptions: [KEY ASSUMPTIONS].
Constraints: put every assumption in one place so I can change inputs and watch the model update; use real formulas (not hard-coded totals); label units and currency. List the top 3 assumptions the outcome is most sensitive to.

Goal: build a KPI dashboard for [TEAM / PRODUCT] from the data I've uploaded ([FILE]).
Deliverable: a Sheet (or interactive HTML dashboard) that shows [METRICS] with trend charts, a top summary row, and filters for [DIMENSION].
Context: the data columns are [COLUMNS]; the period is [TIMEFRAME]; the audience is [AUDIENCE].
Constraints: clean and validate the data first (flag anything missing or malformed), compute each metric transparently, and lead with the 3 numbers that matter most. End with one paragraph on what the data is telling us.

Goal: run Deep Research on [TOPIC / QUESTION] and produce a decision-ready report for [AUDIENCE].
Deliverable: a structured report — executive summary, key findings, the evidence for each, open questions, and a source list.
Context: I care most about [ANGLE]; timeframe is [TIMEFRAME]; prior beliefs to test: [ASSUMPTIONS].
Constraints: search broadly, prefer primary and recent sources, cite every claim with a link and publication date, and clearly separate what's confirmed from what's uncertain. If sources disagree, show the disagreement rather than averaging it away.

Goal: map the competitive landscape for [PRODUCT / MARKET] as of [MONTH YEAR].
Deliverable: a report plus a comparison Sheet covering the top [N] competitors — positioning, pricing, key features, strengths, weaknesses, and target segment.
Context: my product is [DESCRIPTION]; the segment I compete in is [SEGMENT].
Constraints: use Deep Research to gather current facts, cite each competitor's pricing and claims with a dated link, and flag anything you couldn't verify. Close with where the market has an unmet gap I could target.

Goal: build an evidence review on [RESEARCH QUESTION].
Deliverable: a report that groups the strongest sources by theme, summarizes what each contributes, and traces how the thinking has evolved, ending with open questions.
Context: field is [FIELD]; timeframe [TIMEFRAME]; I need this to be [DEPTH / AUDIENCE].
Constraints: run Deep Research, cite every source with a link and date, note methodological disagreements, and mark where the evidence is thin. Keep the tone rigorous but readable, and don't overstate consensus that isn't there.

Goal: run a content pipeline that turns [SOURCE — e.g. a transcript, a report, a URL] into a finished [OUTPUT — blog post / newsletter / thread].
Deliverable: the final piece plus a short outline of how you built it.
Steps: (1) extract the key points from the source, (2) research 2–3 supporting facts with citations, (3) draft in [TONE] for [AUDIENCE], (4) self-edit for clarity and length ([WORD COUNT]), (5) produce a title and meta description.
Constraints: do all steps in one run without asking me between them; cite any added facts with links; flag anything you had to assume.

Goal: triage the messages/notes I've pasted or uploaded ([SOURCE]) into a prioritized action plan.
Deliverable: a Sheet with columns — Item | Category | Priority (P1–P3) | Suggested next action | Owner | Draft reply (if needed).
Context: my priorities right now are [PRIORITIES]; categories I use are [CATEGORIES].
Constraints: read every item, don't drop any, draft short replies only where one is clearly warranted, and put the P1s at the top with a one-line "why now". End with the 3 things I should do first today.

Goal: build a reusable weekly report for [PURPOSE — e.g. product metrics, market news, project status].
Deliverable: a report template plus this week's filled-in version, and a checklist of the exact steps to regenerate it next week.
Context: the inputs each week are [DATA SOURCES]; the audience is [AUDIENCE]; the sections I want are [SECTIONS].
Constraints: define the format once so future runs are consistent, pull current data where relevant (cite it), and keep it skimmable — headline first, detail below. Make the regeneration checklist specific enough that a future run needs no extra instructions.

Goal: collect a structured dataset about [SUBJECT] from the web.
Deliverable: a Sheet with one row per item and these columns: [COLUMNS]. Include a source link and the date collected for each row.
Context: good starting points are [URLS or "search for X"]; I need roughly [N] items matching [CRITERIA].
Constraints: only include items you can actually source, mark any field you couldn't find as "N/A" rather than guessing, dedupe near-identical rows, and note how many you found and any you excluded. Keep the columns consistent so I can sort and filter.

Goal: clean and enrich the list I've uploaded ([FILE]).
Deliverable: a Sheet with the cleaned rows plus new columns for [FIELDS TO ADD — e.g. company, industry, size, website], and a separate tab listing duplicates and rows you couldn't verify.
Context: the existing columns are [COLUMNS]; "duplicate" means [MATCH RULE].
Constraints: enrich only from sources you can cite, never invent contact details, standardize formats (casing, phone, URLs), and flag low-confidence enrichments. Report how many rows you kept, merged, and dropped.

Goal: extract structured data from the uploaded files ([FILES — PDFs / invoices / contracts / images]).
Deliverable: a Sheet where each file becomes a row (or set of rows) with these fields: [FIELDS].
Context: the files are [DESCRIPTION]; the fields matter because [WHY].
Constraints: read each file fully (use multimodal/OCR for scans), pull only what's actually present, mark missing fields as blank, and add a "confidence" note where a value was ambiguous. List any file you couldn't parse and why.

Goal: convert the uploaded design ([SCREENSHOT / FIGMA EXPORT]) into working front-end code.
Deliverable: responsive HTML/CSS (and JS if the design implies interaction) that matches the layout, spacing, and colors, plus a preview.
Context: target is [DESKTOP / MOBILE / BOTH]; brand color is [HEX]; the interactive bits should do [BEHAVIOR].
Constraints: match the design closely, use semantic HTML and accessible markup, and keep the CSS clean and commented. Note anywhere the design was ambiguous and how you interpreted it.

Goal: build a working prototype of [APP IDEA — e.g. a habit tracker, a quote calculator, a quiz].
Deliverable: a single self-contained HTML/JS app I can open and use, with state kept in the browser (localStorage) so my data persists.
Context: the core user job is [JOB]; the main screen shows [WHAT]; key actions are [ACTIONS].
Constraints: make it actually functional, not a mockup; handle empty and error states; keep the UI clean and responsive. End with a short list of features you'd add next and how.

Goal: build the tool described in this spec: [PASTE SPEC — inputs, logic, outputs].
Deliverable: working code (single file if possible) that implements the spec, plus a couple of example runs proving it behaves correctly.
Context: intended user is [USER]; the tricky part is [EDGE CASE / RULE].
Constraints: implement exactly the spec — if something is under-specified, choose a sensible default and tell me what you chose. Include input validation, test it against the edge case above, and show the test output. Don't add features I didn't ask for.

Goal: [BIG GOAL — e.g. launch-ready package: landing page + deck + one-pager for PRODUCT].
Before you build anything, produce a plan: break the goal into sub-tasks, say which sub-agent handles each and in what order, and list the deliverables. Show me the plan.
Then execute the plan end to end and hand me every artifact.
Context: [CONTEXT / BRAND / AUDIENCE]. Constraints: [FORMAT, TONE, LIMITS]. Definition of done: [WHAT "FINISHED" MEANS].
Rule: don't stop for approval mid-run unless a decision would materially change the output — in that case ask one focused question, then continue.

Goal: produce [DELIVERABLE] that requires both research and building.
Split the work: one track gathers and cites the facts (Deep Research), another builds the artifact from those facts, and a final pass reconciles them so every claim in the artifact traces to a source.
Context: topic is [TOPIC]; the artifact is [WEBSITE / DECK / REPORT / DASHBOARD]; audience [AUDIENCE].
Constraints: run the research and drafting in parallel where you can, cite every factual claim with a dated link, and don't let the build get ahead of the evidence. Deliver the artifact plus the source list it's built on.

Goal: complete [TASK] and deliver a finished artifact I can use without rework.
After you build it, run a self-review pass: check the output against my requirements one by one, test anything testable, fix what fails, and only then deliver.
Context: requirements are [REQUIREMENTS]; the artifact is [ARTIFACT]; "good enough to ship" means [CRITERIA].
Constraints: show me the requirement-by-requirement checklist with pass/fail, list every fix you made in the review, and flag anything you still couldn't satisfy so I know exactly what's left.