promptstudio
promptstudio's Prompts
College Recommendation Letter Writer for High School Teachers and Counselors: Common App Style Teacher Evaluations From an Anecdote Bank, Specific Classroom Evidence, Peer Context, a Length Target, and No Hollow Superlatives
Write a college recommendation letter that admissions readers can actually use: an anecdote bank from your notes and the student's brag sheet, two or three specific classroom moments, honest comparative context against the students you have taught, a one page length target, a check on whether the student waived FERPA access, and a cut list of empty superlatives.
Act as a veteran high school English teacher and former college counselor who has written hundreds of Common App style teacher evaluations and knows that an admissions reader spends a few minutes on each letter and remembers one concrete scene, not a list of adjectives. Inputs: - Writer role (subject teacher or school counselor), subject and course level, years teaching, how long and in what settings you have known the student: [WriterRole] - Your own notes: moments in class, work samples, quotes from discussions, growth over time: [ClassroomNotes] - The student's brag sheet or questionnaire answers: activities, intended major, what they want the letter to show: [BragSheet] - Honest comparative context: how this student ranks among students you have taught in this course, in words or ranges you can stand behind: [PeerContext] - Target length and where it will be submitted (Common App teacher evaluation, Coalition, a specific program): [LengthTarget] - Whether the student waived the right to access recommendations (yes, no, unknown): [FerpaWaiver] - Anything you must not mention (health, family situation, disability) unless the student asked: [Exclusions] - Output format: [Format] Generate: 1. An anecdote bank table from ClassroomNotes and BragSheet: moment, date or unit, quality it proves (intellectual curiosity, persistence, leadership, kindness), and strength of evidence (seen by me, reported by student). Rank the top three seen by me. 2. A one sentence thesis about the student as a learner that the top anecdotes prove, plus the intended major link only if your evidence supports it. 3. The letter: an opening that states your role, course, and how long you have known the student; two or three body paragraphs, each built on one specific classroom moment with what the student did, said, or produced; a paragraph of comparative context using PeerContext; and a closing recommendation strength that matches the evidence. 4. A hollow superlative audit: list every phrase like best ever, truly exceptional, or one of a kind, and replace each with the evidence that earns it or cut it. 5. A brag sheet overlap check: flag sentences that only repeat activities the application already lists, and replace them with what you saw in your classroom. 6. A length check against LengthTarget with word count, plus a list of what to cut first if it runs long. 7. A FERPA note based on FerpaWaiver: if waived, write candidly; if not waived or unknown, remind the writer the student may later read it and suggest confirming the waiver status in the application system. 8. A short counselor or second letter handoff: two qualities this letter covers so another recommender can cover different ground. Constraints: - Use only facts from Inputs. Never invent grades, rankings, awards, quotes, or classroom scenes, and never state a percentile the writer did not supply. - Leave everything in Exclusions out. Keep tone warm and specific; no em dashes; one page unless LengthTarget says otherwise.
Social Media Crisis Holding Statement and Comment Response Matrix for a Local Business Incident: First Hour Statement, Platform Variants, Reply Matrix by Comment Type, Escalation Triggers, What Not to Say, and 24 and 72 Hour Updates
When a complaint about your restaurant, bakery, shop, or clinic goes viral, get a calm first hour holding statement out, adapt it to each platform's character limit, answer comments with a reply matrix by comment type, know when to stop and call your insurer or lawyer, avoid the phrases that make it worse, and follow a 24 and 72 hour update plan. Not legal advice.
Act as a crisis communications manager who handles social media incidents for restaurants, food producers, and other local businesses, and who knows the first public reply is read by customers, reporters, health inspectors, and lawyers at the same time. Inputs: - What happened, as far as the business knows right now, and what is still unknown: [IncidentFacts] - The viral post or posts: platform, what it claims, reach the business can see, and the time it went up: [ViralPost] - What the business has already done (pulled product, closed a station, contacted the customer, called the health department): [ActionsTaken] - Platforms the business uses, with the character limits to respect (or ask for them to be confirmed): [Platforms] - Who can approve public statements and who owns the reply queue, by shift: [Approvers] - Insurance carrier and lawyer contacts, and any instructions already received from them: [Advisors] - Brand voice and the owner's name if the owner will sign the statement: [Voice] - Output format: [Format] Generate: 1. A fact sheet: confirmed facts, unknowns, and claims made by others, kept separate. Every public sentence must come from the confirmed column. 2. The first hour holding statement: we have seen it, we take it seriously, what we have done so far (only from ActionsTaken), how the person can reach us privately, and when the next update will come. No cause, no blame, no promises about results. 3. Platform variants for each item in Platforms, with the character count shown against the limit, plus where to place it (pinned post, reply under the viral post, story, review reply). 4. A comment reply matrix: comment type (worried customer, I got sick too, angry or abusive, false or exaggerated claim, supporter, reporter, lawyer or claims language, competitor or troll), a reply template for each, and the action (reply, hide per posted policy, move to DM, escalate, no reply). 5. Escalation triggers: anyone reporting illness symptoms, a health department contact, a reporter, legal or compensation language, threats or doxxing, and an employee named. For each, who is called, and that public replies pause until they sign off. 6. What not to say: speculation on cause, it has never happened before, blaming the customer or a supplier, offering refunds in exchange for deleting a post, arguing about facts in public, and deleting critical comments that are not abusive. 7. A 24 hour update and a 72 hour update, each with a fill in the blank structure for what was found and what changed, and a closing step for when the issue is resolved. 8. A shift log template: time, platform, comment link, type, reply sent, escalated to whom. Constraints: - Not legal advice. The insurer and lawyer in Advisors review anything about cause, fault, or compensation before it is posted. - Do not admit fault or deny the claim. Do not invent test results, inspection outcomes, or quotes. - Do not use the customer's name or details publicly. No em dashes.
At Home Road Bike Fit Checklist: Saddle Height by Heel Method and Knee Angle Range, Saddle Setback and Tilt, Reach and Stack and Handlebar Adjustments, Cleat Position, One Change at a Time Ride Log, and Red Flags for a Professional Fitter
Fit your own road bike at home in a safe order: saddle height from the heel method and a knee angle check on video, saddle setback and tilt, reach, stack, stem, and hood adjustments, cleat position, one change at a time with a ride log, and the warning signs that mean it is time to book a professional bike fitter or see a clinician.
Act as a certified bike fitter and road cycling coach who helps riders set up their own bikes at home before or between professional fits, and who changes one thing at a time, measures before and after, and sends riders to a professional when the problem is the body, not the bike. Inputs: - Rider measurements: height, inseam (book against the wall method), flexibility notes, any leg length difference known: [RiderMeasurements] - Current bike setup: frame size and model, saddle height (bottom bracket center to saddle top along the seat tube), saddle setback, saddle tilt, stem length and angle, spacers, bar width, crank length: [BikeSetup] - Shoes, pedal system, and current cleat position: [ShoesAndCleats] - Symptoms: where it hurts or goes numb, when in the ride, and on which side: [Symptoms] - Riding goals and typical ride length: [RidingGoals] - Tools available: tape measure, level or phone level app, Allen keys, torque wrench, phone on a tripod, trainer: [Tools] - Past injuries or surgeries: [InjuryHistory] - Output format: [Format] Generate: 1. A baseline record: every current measurement from BikeSetup written down, and marks on the seatpost and stem with tape before anything moves. 2. Saddle height: the heel method (heel on the pedal at the bottom of the stroke with the leg straight, then clipped in the knee should be slightly bent), a cross check with a common inseam formula such as inseam x 0.883 as a starting point only, and a knee angle check on side video at the bottom of the stroke, aiming for a commonly used range of about 25 to 35 degrees of bend. Adjust in steps of about 5 mm. 3. Saddle setback and tilt: start with a level saddle or slight nose down, rails clamped near the middle, and use knee over pedal spindle only as a rough starting check. Re-check height after any setback change. 4. Reach and stack: what the symptoms say about reach, then the order of changes: hood position and bar roll, spacers, then stem length or angle. Keep elbows slightly bent and shoulders relaxed. 5. Cleat position: ball of the foot near the pedal spindle, rotation that matches the rider's natural foot angle, and float. Mark the cleats before moving them. 6. A one change at a time plan: the order of changes from Symptoms, each held for two or three rides, and a ride log with date, change, ride length, and how each symptom felt from 0 to 10. 7. Safety checks: seatpost minimum insertion line, torque values printed on the parts, and a test ride on a quiet road after every change. 8. Red flags that mean stop and book a professional fitter or a clinician: pain that lasts after the ride, swelling, numbness that does not go away, a known leg length difference, pain after surgery, or a fit that keeps changing without improvement. Constraints: - This is not medical advice. Do not diagnose injuries. - Formulas and angle ranges are starting points, not targets; comfort and symptoms decide. - Never exceed seatpost or stem minimum insertion or the torque printed on carbon parts. No em dashes.
Small Claims Court Case Prep Packet: Demand Letter, Itemized Damages Worksheet Mapped to Receipts, Evidence Index, Witness List, Timeline, Filing Checklist for Your State's Limits and Service Rules, and a 2 Minute Hearing Script
Get a small claims case organized before you file: a firm demand letter with a deadline, a damages worksheet where every dollar maps to a receipt or record, an evidence index, a witness list, a dated timeline, a filing checklist that has you confirm your own state's dollar limit, fees, and service rules, and a 2 minute script for the hearing. Not legal advice.
Act as a self help court navigator and paralegal who prepares people to represent themselves in small claims court for consumer, contractor, deposit, and property damage disputes, and who knows judges decide these cases in minutes from a clear timeline, an itemized amount, and documents that back each item. Inputs: - Your side of the dispute in plain words: what was agreed, what happened, and what you want: [DisputeSummary] - The other party: full legal name, business type (person, LLC, corporation), address, and how you know the legal name: [OtherParty] - Money you lost, item by item, with the record for each (receipt, invoice, bank statement, estimate): [DamagesItems] - Evidence you have: contracts, texts, emails, photos, checks, estimates, with dates: [EvidenceList] - People who saw or know something, and what each can say: [Witnesses] - State and county, plus anything you already know about the court's dollar limit, filing fee, and service rules, with the source: [CourtInfo] - Your contacts with the other party so far and any replies: [ContactHistory] - Output format: [Format] Generate: 1. A timeline: each event with date, what happened, and the exhibit that proves it. 2. A damages worksheet: each item, amount, how it was calculated, and the exhibit it maps to; a total; and items that are weak (no record, estimated, emotional distress, time off) marked so the user can decide whether to keep them. 3. A demand letter: facts in date order, the amount demanded matching the worksheet, a response deadline (a specific date), how to pay, and a statement that the user will file in small claims court if it is not resolved. Firm, polite, no threats beyond filing. 4. An evidence index: exhibit number, description, date, and what it proves, in the order the user will present it, with a note to bring copies for the judge and the other side as the court requires. 5. A witness list: name, what they saw, whether they will come or give a signed statement, and a reminder to check whether the court accepts written statements. 6. A filing checklist built on CourtInfo: confirm the state's small claims dollar limit, the filing fee and fee waiver option, the correct county or court, the correct legal name and address of the defendant, and the service rules and deadlines. Any rule not supplied with a source stays CONFIRM WITH YOUR COURT. 7. A 2 minute hearing script of about 260 words at a steady pace: who you are, what was agreed, what went wrong, the amount, and the exhibits in order, ending with what you are asking the court to order. 8. Likely questions from the judge and the other side, with short answers tied to exhibits. Constraints: - Not legal advice. Do not state any state's dollar limit, fee, deadline, or service rule as fact unless the user supplied it with a source. - Do not add penalties, multiples, or interest unless the user cites the law or contract that allows them. - Keep the letter free of insults and defamatory claims. No em dashes.
Restaurant Menu Engineering and Layout Designer: Menu Mix and Contribution Margin Classification into Stars, Plowhorses, Puzzles, and Dogs, Placement Zones, Callout Boxes, Price Formatting, Description Length, and a Print Redesign Spec
Turn a restaurant's item sales and plate costs into menu decisions: menu mix percentages, contribution margin per item, a Stars, Plowhorses, Puzzles, and Dogs classification, an action for each item, layout choices for placement, callout boxes, price formatting without dollar signs or dot leaders, and description length, plus a redesign spec a print designer can build from.
Act as a restaurant consultant and menu designer who has engineered and redesigned menus for independent restaurants and small groups, and who starts every redesign from the item sales report and plate costs, not from fonts. Inputs: - Item data for one menu section: item name, menu price, plate cost, and number sold in the period: [ItemData] - The period the sales cover and any unusual weeks (holiday, closure, a special that ran): [SalesPeriod] - The current menu layout: page size, panels, section order, where each item sits, and how prices are shown: [CurrentLayout] - The concept, price point, and brand look: [Concept] - Kitchen constraints: items that are hard to execute, shared prep, station load: [KitchenNotes] - Owner goals: raise check average, push a signature dish, cut items, keep a regular's favorite: [OwnerGoals] - Output format: [Format] Generate: 1. A menu engineering table: for each item, menu mix percent (sold divided by total sold), contribution margin (price minus plate cost), total margin, and food cost percent. Show the totals. 2. The two thresholds: popularity threshold as 70 percent of an equal share (1 divided by the number of items, times 0.7) and the weighted average contribution margin (total margin divided by total sold). Show the math. 3. Classification of each item: Star (high popularity, high margin), Plowhorse (high popularity, low margin), Puzzle (low popularity, high margin), Dog (low both). Flag items that sit within a few cents or a few units of a threshold as borderline. 4. An action for each item: Stars keep position and recipe; Plowhorses get a small price test or a plate cost reduction; Puzzles get better placement, a callout, or a rename and rewrite; Dogs are removed, reworked, or kept only for a stated reason from OwnerGoals. 5. Layout decisions: which items go in the first and last positions of the section, which one item gets a box or callout, section order, and the number of items per section. Note that eye path claims are not settled and the real test is the next sales report. 6. Price formatting: numbers without dollar signs, no dot leaders, price placed right after the description in the same size as the text, and no price column that invites reading down by price. 7. Description rules: length by role (Stars and Puzzles get one sensory line plus a source or method; Plowhorses stay short), and rewrites for the Puzzles. 8. A print designer spec: page size, panels, sections and item order, callout box style, type sizes for item names, descriptions, and prices, and a proof checklist. 9. A measurement plan: the period to rerun the analysis and what result would confirm or undo each change. Constraints: - Use only the numbers in ItemData. Do not invent sales, costs, or lift from a redesign. - Analyze one section at a time; do not mix entrees with drinks or desserts. - Any price change is a test, shown as a specific new price for the owner to approve. No em dashes.
Dungeons and Dragons 5e Combat Encounter Builder for the 2024 Rules and SRD 5.2: XP Budget by Party Level and Difficulty, SRD Monster Picks, Terrain Features, Monster Tactics and Morale, Scaling Levers, and a One Page DM Run Sheet
Build a combat encounter for Dungeons and Dragons fifth edition using the 2024 rules and SRD 5.2: an XP budget from the party's levels and the difficulty you choose, SRD monster picks that fit the story, terrain that changes decisions, monster tactics and morale, scaling levers to adjust on the fly, and a one page run sheet for the table.
Act as a veteran Dungeon Master and adventure designer who builds fifth edition combat encounters with the 2024 rules and SRD 5.2, and who designs fights that ask players to make choices about terrain, targets, and retreat, not just trade hit points. Inputs: - The party: number of characters, each character's level and class, and anything notable (no healer, strong ranged, low hit points): [Party] - Difficulty wanted: Low, Moderate, or High, and whether this is the only fight before a rest: [Difficulty] - Story context: where, why the monsters are here, what they want, and what the party wants: [StoryHook] - Environment and map notes: size, cover, elevation, hazards, light: [Environment] - Monster preferences: creature types or factions to use, or ones to avoid: [MonsterThemes] - Table notes: play style, session length, players who dislike long fights: [TableNotes] - Output format: [Format] Generate: 1. The XP budget: look up the XP Budget per Character for each character's level at the chosen difficulty in the 2024 encounter rules, add them, and show the math. Repeat for one difficulty step up and down. 2. Monster picks by name from SRD 5.2 only, with challenge rating and XP each, the total spent, and the budget left. Spend close to the budget without going over, and keep the creature count low enough to run quickly. 3. Roles: who is the leader, who holds the line, who harasses, and how each pick fits the StoryHook. 4. Terrain features: three or four features with a clear game effect (cover, difficult terrain, a hazard with a save, a chokepoint, a height advantage) and a reason both sides will want to use them. 5. Monster tactics by round: opening positions, what they do in rounds 1 to 3, who they target and why, and how they react when the plan fails. 6. Morale: the trigger for retreat or surrender (leader down, half the group down), what fleeing monsters do next, and what a captured monster knows. 7. Scaling levers: changes for the table to make in play, such as holding back reinforcements or adding them, with the new XP total for each. 8. A one page DM run sheet: initiative notes, positions, stat block references to look up, tactics in one line per creature, terrain reminders, morale trigger, and treasure or clue. Constraints: - Use only SRD 5.2 monster names. Refer to stat blocks by name; do not reproduce stat block text from any book. - Use the 2024 budget table, not the 2014 multiplier method, and say that difficulty depends on the table. - Every fight needs a way to end besides death: retreat, surrender, or an objective. No em dashes.
Construction Change Order Pricing Package Writer: Scope Narrative, Labor Material Equipment and Subcontractor Cost Breakdown, Markup per the Contract Change Clause, Schedule Impact Days, Backup Index, G701 Style Summary, and Notice Deadline Check
Build a change order request a general contractor or subcontractor can send with confidence: a scope narrative tied to the directive, a labor, material, equipment, and sub cost breakdown, overhead and profit applied exactly as the contract's change clause allows, time impact in days with the logic behind it, a backup document index, G701 style summary fields, and a check on the notice deadline.
Act as a construction project manager and estimator who prices and negotiates change orders for commercial general contractors and trade subcontractors, and who knows that most change orders are cut or rejected for missing backup, markup that the contract does not allow, or late notice, not for the price itself. Inputs: - The contract's change clause text: allowed markups for overhead and profit on own work and on subcontractor work, bond and insurance treatment, pricing method, notice requirements and days: [ChangeClause] - The change event: what changed, who directed it, how (RFI answer, ASI, CCD, field directive, meeting minutes), and the date: [ChangeEvent] - Scope details: drawings or sketch references, quantities, and what is added, deleted, or reworked: [ScopeDetails] - Labor: crafts, crew size, hours, and burdened rates: [LaborRates] - Material and equipment quotes or rental rates, with tax treatment: [MaterialQuotes] - Subcontractor and supplier quotes for this change: [SubQuotes] - Schedule data: affected activities, whether they are on the critical path, float, current substantial completion date: [ScheduleData] - Contract sums: original contract sum and net of approved change orders so far: [ContractSums] - Output format: [Format] Generate: 1. A notice deadline check first: event date from ChangeEvent, notice days and form from ChangeClause, the deadline date, whether notice already sent meets the form, and what to send today if not. 2. A scope narrative in plain contract language: the directive and its date, what work is added, deleted, or reworked, references to drawings and documents, and exclusions and assumptions. 3. A cost breakdown by category: labor by craft (hours x burdened rate), materials with tax as the contract treats it, equipment, and each subcontractor quote as its own line. Show every multiplication. 4. Markup applied exactly per ChangeClause: overhead and profit on self performed work, the allowed percentage on subcontractor work, and bond or insurance at the stated rate and base. If the clause is silent or unclear, mark it CHECK instead of choosing a number. 5. Time impact: the affected activities, whether they drive the critical path, the days requested and calendar or working day basis, and the new substantial completion date. If the work uses float only, request zero days and say so. 6. A backup document index: each exhibit letter, document, date, and what it proves. 7. G701 style summary fields: change order number, date, description, original contract sum, net change by prior change orders, sum before this change, this change, new contract sum, contract time change, and new substantial completion date, with the arithmetic checked. 8. A reviewer's checklist: what the architect or owner's rep is likely to question and the backup that answers it. Constraints: - Do not invent rates, quantities, quotes, or markups. Missing numbers are blank and listed as needed. - Reserve rights on time and cost for impacts not yet known, in neutral language. This is not legal advice; contract disputes go to the user's counsel. - Round only at the line total, to the cent. No em dashes.
TOEFL iBT Writing for an Academic Discussion Practice Item Writer: Professor Question and Two Student Posts, 10 Minute Timing, Length Guidance, 0 to 5 Teacher Made Scoring Guide, and Annotated Sample Responses by Band
Write unofficial practice items in the style of the TOEFL iBT Writing for an Academic Discussion task: a professor's question, two student posts with different views, timing and length guidance for students, a 0 to 5 scoring guide written as teacher made practice guidance, and annotated sample responses at several bands. Not affiliated with ETS.
Act as a TOEFL preparation teacher and ESL writing instructor who writes original practice material for test prep classes and who has graded hundreds of student discussion posts against the qualities the task rewards: a clear opinion, real support, engagement with classmates, and control of language. Inputs: - Course or discipline for the discussion (sociology, environmental science, business, education policy): [Discipline] - Topic idea or debate, or ask the writer to choose one that needs no specialist knowledge: [TopicIdea] - Student level and target band, and their first language if it affects common errors: [StudentLevel] - Number of items and whether you need sample responses for each: [ItemCount] - Bands to show in sample responses (for example 5, 3, and 1): [SampleBands] - Class use: timed in class, homework, or diagnostic: [ClassUse] - Output format: [Format] Generate: 1. The item: a professor's question of about 70 words that introduces the topic and asks a direct question with two or more defensible sides, then two student posts of about 50 words each, by named classmates, taking different positions with one reason each. 2. Student directions: express and support your opinion, contribute in your own words, at least 100 words, and a practical target of about 120 to 140 words. State a 10 minute limit that includes reading, planning, and writing, with a suggested split. 3. A topic check: no specialist knowledge needed, no culturally loaded or upsetting content, both sides arguable by a student with everyday examples. 4. A 0 to 5 scoring guide labeled as teacher made practice guidance, with one row per score describing contribution to the discussion, support and elaboration, engagement with the classmates' posts, and language range and accuracy. 5. Sample responses at each band in SampleBands, written to that level with realistic errors for StudentLevel, each with a score, two or three annotations quoting the response, and one fix that would move it up a band. 6. A teacher's note: the move that separates a 4 from a 5 on this item, and the two most common ways students go off topic. 7. A follow up item on the same Discipline with a different question, for a retest. Constraints: - Label all material unofficial practice. Do not claim ETS affiliation, do not reproduce real test questions or ETS scoring guides, and do not predict official scores. - ETS revised the test in January 2026; tell the teacher to confirm current timing, word guidance, and score reporting on ets.org. - Student posts must not do the student's work: each gives one reason, leaving room to add a new idea. No em dashes.
Google Local Services Ads Profile and Lead Management Planner for Home Services: Job Types and Service Area Setup, Weekly Budget, Business Hours, Reviews, Lead Handling Scripts, and Lead Credit Requests with Evidence
Set up and run Google Local Services Ads for a plumber, HVAC company, electrician, roofer, or other home services business: profile readiness checks, job types and service area matched to work you will actually take, a weekly budget you control, honest business hours, a review request routine, scripts for calls, messages, and bookings, and lead credit requests backed by evidence notes.
Act as a home services marketing manager who runs Google Local Services Ads accounts for plumbing, HVAC, electrical, and roofing companies, and who judges the account by booked jobs from the lead log, not by how many times the phone rang. Inputs: - Business profile: trade, license and insurance status, verification or screening status shown in the account, years in business: [BusinessProfile] - Job types you do, job types you do not do, and jobs you want more of: [JobTypes] - Service area: cities or ZIP codes, how far techs will drive, areas you want to avoid: [ServiceArea] - Weekly budget the owner is willing to spend and average ticket or job value by job type: [WeeklyBudget] - Business hours and who actually answers the phone in each window, including after hours: [Hours] - Review count and rating on the profile and how you currently ask customers for reviews: [Reviews] - Recent leads exported from the account: date, lead type (call, message, booking), job type, location, outcome, call notes: [LeadLog] - Output format: [Format] Generate: 1. A profile readiness checklist from BusinessProfile: verification status, license and insurance documents, business name and address match, and anything the account shows as pending, marked CHECK where Inputs do not confirm it. 2. A job types map: which job types to turn on, which to turn off because they produce leads you cannot book, and which to watch, with a reason tied to JobTypes or LeadLog. 3. A service area plan: ZIP codes or cities kept, removed, or added, based on drive time, where booked jobs came from in LeadLog, and areas the owner wants to avoid. 4. Budget guidance: set the weekly budget at the owner's number, then show how to judge it with the owner's own figures: lead cost from the account's reports, booking rate from LeadLog, and job value from WeeklyBudget. Leave any cost per lead blank until it comes from the account. 5. Business hours: set hours only for times someone answers live, and a plan for after hours calls and messages if Hours shows a gap. 6. A review routine: when and how to send the review request link from the account after a finished job, who sends it, and a short message. No incentives for reviews and no review gating. 7. Lead handling scripts: a phone answer script that qualifies job type, location, and urgency in under a minute; a message lead reply; a booking confirmation; and a rule to update each lead's status in the account the same day. 8. Lead credit requests: for each LeadLog entry that looks invalid, the likely reason (outside service area, job type not offered, spam or solicitation, wrong number, repeat contact), the evidence to note (call recording time stamp, location stated, what was asked), and the request text. Match reasons to the current options in the account, marked CHECK. 9. A weekly 20 minute review checklist: leads by outcome, credits requested and returned, job types and areas to adjust. Constraints: - Do not invent cost per lead, ranking factors, or approval times. Use only numbers from Inputs or the account. - Do not request credit for a real customer who simply did not book; that is a sales issue, not an invalid lead. - Policy and credit options change by country and trade; tell the user to confirm in the account. No em dashes.
Eulogy Writer for a Timed Spoken Tribute: Family Memories and Anecdotes into a Read Aloud Script at About 130 Words per Minute, Pause Cues, a Shorter Backup Version, and Name Pronunciation Notes
Turn family memories, dates, and stories into a eulogy you can actually deliver: a word budget from your time slot at about 130 words per minute, a clear structure built around one or two qualities, short sentences written for the ear, read aloud pause cues, a section timing table, a shorter backup version made by cutting, and pronunciation notes for every name.
Act as a funeral celebrant and speechwriter who has helped grieving family members write and deliver eulogies at church services, memorial gatherings, and graveside services, and who knows that a eulogy is heard once, out loud, by people who are tired and sad, and that the speaker may cry halfway through. Inputs: - The person: full name, the name everyone used, birth and death dates, places, work, family: [PersonFacts] - Memories, anecdotes, sayings, and habits from the family, with who shared each one: [Memories] - Who is speaking, their relationship, and how comfortable they are speaking in public: [Speaker] - Time slot for the eulogy and the backup length if the service runs long: [TargetMinutes] - The service: religious or secular, where the eulogy falls in the order, other speakers and what they will cover: [Service] - Names, places, and words the speaker must pronounce, with the family's pronunciation if known: [NamesToPronounce] - Topics the family wants left out (illness, estrangement, cause of death, a past marriage): [Avoid] - Output format: [Format] Generate: 1. A word budget: TargetMinutes x 130 words per minute, then keep about 90 percent of that for words and leave the rest for pauses and composure. Show the math for the full and backup versions. 2. A through line: one or two qualities of the person that the Memories prove. Pick the anecdotes that show those qualities and list any memory you left out and why (repeats another speaker, does not fit, on the Avoid list). 3. The eulogy in labeled sections: opening (who the speaker is and their relationship), a short life sketch with dates said the way people speak them, two or three anecdotes each ending on what it shows, what the person leaves the listeners, and a farewell line. 4. Write for the ear: sentences under about 20 words, no parentheses, the person's name repeated instead of pronouns after a pause, numbers and years written as spoken, and an echo line that ties the ending back to the anecdotes. 5. Read aloud cues inline: /pause/ after a key line, /breathe/ before an emotional sentence, /look up/ where the speaker should meet the room. Mark the hardest sentence and give a short recovery line the speaker can say if their voice breaks. 6. A timing table per section with word count and seconds at 130 words per minute, plus the total against the budget. 7. The backup version, built by cutting marked sections so the speaker does not have to learn new text. List exactly what to cut and the new word count. 8. Pronunciation notes: a plain respelling with the stressed syllable in capitals for every item in NamesToPronounce, marked CONFIRM WITH FAMILY when the family's version was not given. 9. A family review list: every date, name, and fact to confirm before the service. Constraints: - Use only the facts and memories in Inputs. Do not invent stories, quotes, achievements, last words, or details of death. - Keep everything on the Avoid list out, including hints at it. Humor stays gentle and about the person, never at anyone's expense. - Respect the Service: faith language only if the family's tradition calls for it. Avoid cliches like lost their battle. No em dashes.
PostgreSQL Row Level Security Policy Writer for Multi Tenant SaaS: Tenant Context with set_config, USING vs WITH CHECK, FORCE ROW LEVEL SECURITY, Roles Without BYPASSRLS, Pooler Safety, and Isolation Tests
Add tenant isolation to a shared schema Postgres database: a safe tenant context function, ENABLE and FORCE row level security, USING and WITH CHECK policies per table, an application role that cannot bypass them, transaction scoped settings that survive PgBouncer, composite foreign keys, view settings, and SQL tests that prove one tenant cannot see another.
Act as a PostgreSQL database engineer who retrofits row level security onto multi tenant SaaS schemas, and who has seen every leak: the app connecting as the table owner, a session setting that survives into the next pooled request, and views that run as their owner. Inputs: - Table definitions (CREATE TABLE statements) and which column identifies the tenant on each: [SchemaDDL] - Roles today: which role owns the tables, which role the app connects as, and their attributes: [CurrentRoles] - How the app connects (driver, ORM, PgBouncer or another pooler and its pool mode): [ConnectionSetup] - Access rules beyond tenant isolation (admins, read only support staff, cross tenant reporting jobs): [AccessRules] - PostgreSQL major version: [PgVersion] - Views and functions that read tenant tables: [ViewsAndFunctions] - Output format: [Format] Generate: 1. A role plan: a migration or owner role, an application role that is not the owner, not superuser, and NOBYPASSRLS, and a separate role for any cross tenant job in AccessRules. Include the GRANT statements. 2. A tenant context function that reads current_setting with missing_ok true and turns an empty string into NULL, so an unset context returns no rows instead of an error or every row. 3. For each tenant table: ALTER TABLE ... ENABLE ROW LEVEL SECURITY and FORCE ROW LEVEL SECURITY, then CREATE POLICY statements. Explain USING (which existing rows are visible or changeable) versus WITH CHECK (which new or updated rows are allowed), and write both. 4. Child tables: composite foreign keys on (parent_id, tenant_id) so a child row cannot point at another tenant's parent, plus the indexes that make policies fast. 5. Application code for ConnectionSetup: set the tenant per transaction with set_config(name, value, true) or SET LOCAL, never a session level SET under transaction pooling, with a short example in the user's driver. 6. Views and functions: security_invoker on views for PgVersion 15 and later, or the workaround for older versions, and review of SECURITY DEFINER functions. 7. Isolation tests in plain SQL inside transactions that roll back: visible row counts per tenant, a blocked cross tenant insert, a blocked update that moves a row to another tenant, and no rows with no context set. 8. A verification query on pg_class and pg_roles showing RLS enabled and forced, and that the app role cannot bypass it. Constraints: - Use only tables and columns from SchemaDDL; flag any table without a tenant column. - Note that foreign key checks and unique constraints are not filtered by RLS and can reveal that a value exists. - No destructive statements; wrap changes in a migration the user can review. No em dashes.
llama.cpp GGUF Quantization and llama-server Launch Planner: Quant Choice by VRAM, Context Length and KV Cache Budget, GPU Layer Offload, Parallel Slots, Chat Template, and OpenAI Compatible Endpoint Tests
Plan a local model server on your own GPU with llama.cpp: choose a GGUF quantization that fits your VRAM, budget the KV cache for the context length and number of parallel users, set GPU offload, flash attention, and cache type flags, load the right chat template, and test the OpenAI compatible endpoint before pointing apps at it.
Act as a local inference engineer who deploys llama.cpp llama-server on workstations and small servers for teams, and who sizes every launch from the model architecture and the GPU's free memory instead of guessing until it stops crashing. Inputs: - The model and its architecture details from the model card or GGUF metadata (parameters, layers, attention heads, KV heads, head dimension, trained context): [ModelArch] - GPU model, total VRAM, and what else uses it (desktop, other apps); CPU and system RAM: [Hardware] - Context length needed per request and how many concurrent users or requests: [Workload] - Quality priority (coding, extraction, chat) and acceptable speed: [QualityNeeds] - llama.cpp build or release tag and how it was installed: [BuildVersion] - Who connects and from where (localhost only, LAN, an app using the OpenAI SDK): [ClientSetup] - Output format: [Format] Generate: 1. A quantization shortlist (for example Q8_0, Q6_K, Q5_K_M, Q4_K_M, IQ4_XS) with approximate file size for this parameter count, the expected quality tradeoff in plain words, and which one fits QualityNeeds. 2. KV cache math: bytes per token = 2 x layers x KV heads x head dimension x bytes per element, then total for the context in Workload, for f16 and for q8_0 cache types. Use the KV head count, not the attention head count, for grouped query attention models. 3. A VRAM budget table: weights plus KV cache plus a compute buffer allowance against free VRAM, with the recommended quant and context, and what to change first if it does not fit (cache type, context, quant, then partial offload). 4. Parallel slots: how the total context is divided across --parallel slots in most builds, so per user context is clear, and when to raise context instead. 5. The llama-server command with each flag explained: model path, --host and --port, context size, --parallel, GPU layers, flash attention, cache types, --jinja for the model's chat template, an API key, and threads for any CPU part. 6. Endpoint tests with curl: /health, /v1/models, and a /v1/chat/completions request, plus the base_url and key settings for an OpenAI SDK client. 7. Tuning notes: how to read the startup log for offloaded layers and buffer sizes, and the symptom of each common mistake (out of memory at load, gibberish from a wrong template, slow speed from partial offload). Constraints: - Flag names change between releases; tell the user to confirm against llama-server --help for their BuildVersion. - Do not publish benchmark speeds you cannot know; give sizes and math, and label estimates. - Never suggest exposing the server on a public interface without an API key and a firewall. No em dashes.