Muse Glimmer Prompt Generator

I am attaching an architecture diagram for a distributed microservices system. Analyse every component and connection shown in the image and produce: (1) A numbered implementation checklist ordered from foundational infrastructure to application layer — each item should name the specific component, the technology choice shown or implied, and the key configuration decision. (2) A dependency map showing which items must be complete before each subsequent item can begin. (3) A list of the three highest-risk components and why. (4) Any components in the diagram that appear incomplete or inconsistently connected. Format your output as a structured technical brief suitable for a sprint planning session. [ATTACH DIAGRAM]

You are a local coding agent with access to a Python codebase. I will paste the contents of three files below. Your task: (1) Identify all duplicate logic across the three files and propose a shared utility module that eliminates the duplication — show the exact module contents; (2) Flag any function that has no corresponding test coverage and generate a pytest test for each; (3) Identify any function exceeding 40 lines and propose a refactored version that splits it into smaller, named sub-functions; (4) Check all import statements — flag unused imports and suggest removals. Output your changes as a structured diff with file name headers. Do not summarise — show me the actual code. [PASTE FILE CONTENTS]

I am attaching an invoice image. Extract every data point from it and return a fully structured JSON object with the following fields: vendor_name, vendor_address, vendor_email, vendor_phone, invoice_number, invoice_date, due_date, line_items (array of: description, quantity, unit_price, line_total), subtotal, tax_rate, tax_amount, total_due, payment_terms, currency. If any field is absent or illegible, set its value to null and note the reason in a separate 'extraction_notes' field. After the JSON, add a one-paragraph data-quality assessment: what is clearly readable, what is ambiguous, and whether you recommend manual verification of any field. [ATTACH INVOICE IMAGE]

I am attaching a screenshot of a web application dashboard. Conduct a structured UX audit covering: (1) Information hierarchy — is the most important information visually prominent? Identify the three most visually dominant elements and assess whether they match user priority; (2) Cognitive load — count the number of distinct interactive elements visible. Is it above or below the optimal 5-9 for working memory? (3) Affordance clarity — identify any element whose function is unclear from its visual design alone; (4) Accessibility flags — note any likely colour contrast issues, small touch targets, or text that appears below 14px; (5) Navigation clarity — can a new user determine where they are and what their three most likely next actions are within 5 seconds? End with a prioritised list of the five highest-impact improvements, each with a single actionable recommendation. [ATTACH SCREENSHOT]

I am attaching four figures from different academic papers on the same topic: large language model fine-tuning methods. For each figure: state which paper it appears to be from if identifiable, describe what metric is being compared, and what the key takeaway is. Then produce a cross-figure synthesis: (1) Which fine-tuning method shows the most consistent advantage across all four figures? (2) Where do the figures contradict each other — and what methodological difference most likely explains the contradiction? (3) What question do these four figures collectively fail to answer that a fifth study should address? Format as a research briefing note with figure references in brackets. [ATTACH 4 PAPER FIGURES]

I am attaching a photo of handwritten meeting notes. Transcribe every legible element of the notes, marking illegible sections with [unclear]. Then organise the transcribed content into a structured meeting summary with the following sections: Meeting Date and Participants (if noted), Agenda Items Discussed, Key Decisions Made (as a numbered list), Action Items (as a table: Owner | Task | Deadline), Open Questions Raised But Not Resolved, and Any Follow-Up Meeting Dates Noted. If the notes appear to be from multiple separate meetings, separate them clearly. [ATTACH NOTES PHOTO]

I am attaching screenshots of the pricing pages of three competing SaaS products. For each product: extract the plan names, prices, billing frequency, and every feature listed under each plan. Then create a unified comparison in three formats: (1) A structured table with products as columns and features as rows — mark each cell as Included, Add-on, or Absent; (2) A positioning analysis identifying which product is positioned as entry-level, which as mid-market, and which as enterprise — explain the pricing and feature signals that support each positioning; (3) A buyer's guide summary: for what specific buyer profile is each product the best choice, and why? [ATTACH 3 PRICING SCREENSHOTS]

I am pasting a 60-page technical specification document below. After reading the full document, answer the following five questions — cite the specific section or page number for each answer: (1) What is the stated performance benchmark for the system under peak load conditions? (2) What are the three explicitly stated failure modes and their corresponding recovery procedures? (3) Does the document specify a maximum acceptable latency for the primary API endpoint, and if so, what is it? (4) Are there any contradictions between the requirements stated in Section 2 and the constraints described in the appendices? (5) What open questions or TBD items remain unresolved in the final version of the document? [PASTE FULL DOCUMENT]

You are a local customer support triage agent. I will provide you with 20 raw support ticket texts below. For each ticket: (1) Classify the issue type from this fixed taxonomy: Billing, Technical Bug, Feature Request, Account Access, Onboarding, Other; (2) Assign a severity: Critical (blocking work), High (significant friction), Medium (inconvenient), Low (cosmetic); (3) Extract the customer's stated desired outcome in one sentence; (4) Suggest the team or individual this ticket should be routed to; (5) Draft a 2-sentence initial acknowledgement reply that is specific to this ticket's issue — not a generic template. Output as a structured JSON array with one object per ticket containing: ticket_id, issue_type, severity, desired_outcome, routing, acknowledgement_reply. [PASTE TICKET TEXTS]

Review the following Python/Node.js code for security vulnerabilities. For each vulnerability found: state the vulnerability type (e.g. SQL injection, insecure deserialization, hardcoded credential, path traversal, SSRF, command injection, XSS), the exact line number or function name where it occurs, an explanation of how it could be exploited, a severity rating (Critical / High / Medium / Low / Informational) with CVSS 3.1 rationale, and a corrected code snippet that resolves the issue without changing the function's intended behaviour. After all individual findings, provide a summary: total count by severity, the single highest-risk finding you'd remediate first, and any security pattern that appears to be a systemic coding practice problem rather than an isolated mistake. [PASTE CODE]

I am attaching a business performance chart. Provide a structured analysis: (1) Chart type identification and what metric(s) are being plotted, including units and time range; (2) Trend identification — describe the overall trend, any inflection points, and whether the trend is accelerating, decelerating, or stable; (3) Anomaly detection — identify any data points that deviate significantly from the trend and propose the two most likely explanations for each; (4) A plain-English executive summary of what this chart tells a non-technical stakeholder in 3 sentences; (5) The single most important business action this data suggests, and what additional data would be needed to confirm that action is the right one. [ATTACH CHART IMAGE]

You are a privacy-first local document processing agent. I will paste a raw document text below. Your task: (1) Identify and tag every instance of personally identifiable information (PII) by type: full name, email address, phone number, physical address, date of birth, national ID number, financial account number, IP address, biometric descriptor, or medical information; (2) Produce a redacted version of the document where every PII instance is replaced with a type placeholder in brackets, e.g. [EMAIL], [FULL_NAME]; (3) Generate a PII inventory table: Type | Count | Examples of Redacted Values (show only the placeholder format, not the real values); (4) Flag any sentence where PII removal significantly changes the meaning of the surrounding text, requiring human review before use. [PASTE DOCUMENT]

I am pasting raw notes from 12 user research interviews below. Each interview is separated by '---'. Read all 12 and produce: (1) A thematic analysis — identify the top 5 recurring pain points across all interviews, ranked by frequency. For each theme, include 2-3 direct quotes from different participants; (2) A jobs-to-be-done framework summary — for each theme, articulate the underlying job using the format 'When I [situation], I want to [motivation], so I can [expected outcome]'; (3) A prioritised feature recommendation list — 8 features that would address the top themes, ordered by impact and feasibility, with a one-sentence rationale for each; (4) Any notable outlier or contradictory finding that did not fit the main themes but could represent an underserved segment. [PASTE INTERVIEW NOTES]

Write a 1,200-word technical article titled 'Running Open-Weight AI Models Locally: A Practical Guide for 2026.' The audience is a software developer who has heard about local AI but never tried it. Structure: (1) Why run locally — 3 concrete benefits (privacy, latency, cost) with a specific example for each; (2) Hardware requirements — what GPU VRAM is needed for 7B, 13B, and 30B models, and what consumer cards currently hit each threshold; (3) The 5-step quickstart — from model download to first inference, using a real tool like Ollama; (4) A comparison of three current open-weight models suitable for local use (Muse Glimmer, DeepSeek V4 Flash, Kimi K3) — one paragraph each on their strengths; (5) Three real-world use cases where local beats cloud: private document processing, low-latency autocomplete, and offline travel assistant. End with a one-paragraph 'when to stay in the cloud' honest coda.

I am attaching two scanned pages from different versions of the same software licence agreement — Version A and Version B. Compare the two documents and identify: (1) Every clause that differs between versions — present as a structured table with three columns: Clause Reference | Version A Text | Version B Text; (2) Flag any change that materially alters the legal rights or obligations of either party — rate each as Major Change or Minor Change with a one-sentence legal plain-English explanation; (3) Identify any clause present in one version and absent in the other; (4) Write a one-paragraph change summary suitable for a non-lawyer executive explaining what has changed and what it means practically. [ATTACH BOTH DOCUMENT PAGES]

You are an email processing agent running locally. I will paste the contents of 15 raw emails below, each separated by '---'. For each email: (1) Classify it into one category: Action Required, FYI Only, Meeting Request, Newsletter/Marketing, Automated Notification, or Spam; (2) For 'Action Required' emails: extract the specific action, the person responsible, and any stated deadline; (3) For 'Meeting Request' emails: extract the proposed date/time, attendees, location or video link, and meeting purpose; (4) Draft a one-sentence response acknowledgement for every email that requires a reply — do not draft generic responses; (5) Flag the three emails that are highest priority and explain why in one sentence each. Produce your output as a structured per-email report, then a separate 'Priority Actions' section at the end. [PASTE EMAIL CONTENTS]

I am attaching a photo of the contents of my fridge or kitchen counter. Identify every ingredient you can clearly see, noting uncertainty for any item you are not confident about. Then: (1) Suggest 3 complete recipes that can be made using primarily the identified ingredients — each recipe should have a name, 4-sentence method, and a note on any one additional ingredient that would significantly improve it; (2) Rank the 3 recipes from quickest to most involved with estimated prep and cook time; (3) Flag any ingredient in the image that appears to be near or past its freshness peak, based on visible cues; (4) Suggest one pantry staple that is likely already in my cupboard and would unlock 2 additional recipes from these ingredients. [ATTACH FRIDGE/PANTRY PHOTO]

I am attaching a screenshot of a presentation slide. Evaluate it across five dimensions and give a score out of 10 for each with a one-paragraph justification: (1) Clarity — can someone read and understand the main message in under 8 seconds?; (2) Visual hierarchy — is the most important information the most visually prominent?; (3) Data integrity — if there is a chart or statistic, is it clearly labelled and does it appear to be presented without distortion?; (4) Design quality — is the layout clean, are colours used purposefully, is whitespace adequate?; (5) Message strength — does the slide advance a single clear argument or is it trying to do too many things? After the scores, give the one change that would most improve this slide and show the exact revised copy/structure. [ATTACH SLIDE SCREENSHOT]

I will paste excerpts from five different sources on the topic of AI regulation in the EU below. Read all five, then produce: (1) A 300-word objective synthesis of what the sources collectively establish as fact — no opinions, only what is directly evidenced; (2) A comparison of where the sources agree, where they differ, and where they are silent (noting gaps in the collective coverage); (3) A 150-word opinionated analysis — your reasoned assessment of what the most significant implication of these facts is for an AI company operating in Europe in 2026; (4) Three specific follow-up questions that are not answered by any of the five sources and would be material to a compliance decision. [PASTE SOURCE EXCERPTS]

I am attaching a screenshot of an error message and stack trace from my terminal or browser console. Diagnose the error: (1) Identify the exact error type, the file and line where it originates, and the immediate cause; (2) Trace the call stack from the error point back to the originating function call — explain in plain English what each frame in the stack represents; (3) Provide the most likely fix as a code snippet — show the exact lines to change and what to change them to; (4) Explain in one sentence why this fix works; (5) List two alternative causes that could produce this same error if your primary diagnosis is wrong — and what distinguishing test or log check would confirm or rule each one out. [ATTACH ERROR SCREENSHOT]