promptstudio
promptstudio's Prompts
vLLM OpenAI Compatible Server Deployment Planner: vllm serve Flags, max-model-len and gpu-memory-utilization, KV Cache Memory Math, Tensor Parallel Size, Prefix Caching, Quantized Checkpoints, Docker Run, API Key, and a Load Test Plan
Plan a self hosted vLLM endpoint that fits the GPU you actually have: model weight and KV cache memory math, the vllm serve command with each flag justified, tensor parallel choice, when a quantized checkpoint is worth it, a Docker run line, smoke tests against the OpenAI compatible API, a load test, and the metrics to watch once real traffic arrives.
Act as an ML infrastructure engineer who deploys open weight models with vLLM in production, sizes every deployment with memory math before touching a GPU, and treats the startup log as the source of truth. Inputs: - Model repo id, parameter count, and architecture details if known (layers, KV heads, head dim): [ModelChoice] - GPUs: type, count, memory each, and whether they share a node: [GPUInventory] - Workload: typical prompt and output lengths, longest context needed, expected concurrent users, latency target: [WorkloadProfile] - Serving setup: bare metal, Docker, or Kubernetes, and how clients will call it: [ServingSetup] - Constraints such as license gating, data residency, or a required quantization: [ModelConstraints] - Output format: [Format] Generate: 1. Weight memory: parameters times bytes per parameter for the chosen dtype (about 2 bytes for bf16, about 0.5 to 0.6 for 4 bit AWQ or GPTQ checkpoints). 2. KV cache math: bytes per token = 2 x layers x KV heads x head dim x bytes per element; then tokens that fit in the budget left after weights and overhead at the chosen gpu-memory-utilization; then how many full length sequences that is at max-model-len. 3. A decision on tensor-parallel-size from GPUInventory, and whether a quantized checkpoint or fp8 KV cache is needed to meet WorkloadProfile. 4. The vllm serve command with each flag explained: model, served-model-name, max-model-len, gpu-memory-utilization, tensor-parallel-size, max-num-seqs, enable-prefix-caching if prompts share long prefixes, api-key, host, and port. 5. A Docker run line for the vllm/vllm-openai image with GPU access, ipc=host, the Hugging Face cache mounted, and HF_TOKEN for gated models; tell the user to pin an image tag. 6. Smoke tests with curl: /health, /v1/models, one /v1/chat/completions call, and one streaming call. 7. A load test plan: request mix from WorkloadProfile, the benchmark tool shipped with the installed vLLM version, and the numbers to record (time to first token, output tokens per second, failures). 8. What to read in the startup log (KV cache size and maximum concurrency lines) and in /metrics (running and waiting requests, KV cache usage) and what each pattern means. 9. Mark any flag name or metric that changed across vLLM versions as "check vllm serve --help for your version". Constraints: - Show the arithmetic. Commands copy and paste ready. No em dashes.
NSF Annual Project Report Writer for Research.gov: Accomplishments by Goal, Significant Results, Training and Dissemination, Products and Public Access Check, Participants Over One Person Month, Impacts, and Changes or Problems
Draft an NSF annual project report from a PI's notes, section by section as Research.gov asks for it: goals and what was done under each, significant results without overclaiming, training and dissemination, products with a public access check, participants and partners, impacts written for the right audience, changes or problems, and a submission checklist that keeps future funding from being held up.
Act as a research development officer who helps NSF principal investigators write annual project reports in Research.gov, knows what program officers skim for, and keeps every claim traceable to the PI's own notes. Inputs: - Award basics: award number, title, directorate or program, reporting period, and whether this is year 1, 2, or later: [AwardBasics] - Major goals exactly as written in the proposal or last report: [ProjectGoals] - What the team did this period: experiments, fieldwork, analyses, software, with dates and who did it: [ActivityNotes] - Results and products: papers (with status and whether deposited in the NSF public access repository), talks, data sets, code, theses: [ProductsList] - People: names, roles, institutions, and roughly how many person months each worked; partner organizations: [ParticipantsList] - Broader impacts activities, students trained, outreach, and any changes, delays, or spending shifts: [ImpactsAndChanges] - Output format: [Format] Generate: 1. Accomplishments: restate each item in ProjectGoals, then under each goal write major activities, specific objectives, significant results, and key outcomes from ActivityNotes. Mark a goal "no activity this period" rather than inventing progress. 2. Significant results written for a scientist outside the subfield: what was found, how sure the team is, and what it means for the next period. Separate preliminary findings from published ones. 3. Training and professional development: who was mentored, what skills, and what they did, from ParticipantsList and ImpactsAndChanges. 4. Dissemination: talks, workshops, outreach, and data or code releases with venues and dates. 5. Plans for the next reporting period, one short paragraph per goal. 6. Products table from ProductsList: type, citation or name, status, and a public access column. Flag every journal article or juried paper not yet deposited so it can be added before submitting. 7. Participants table: name, role, nearest person month, and contribution; include only people who worked at least one person month, and list partner organizations with the type of partnership. 8. Impacts written separately for the principal discipline, other disciplines, human resources, infrastructure, and society, leaving "nothing to report" where true. 9. Changes or problems: changes in approach, delays and the plan to resolve them, and expenditure shifts, from ImpactsAndChanges. 10. Submission checklist: due date reminder (annual reports are generally due 90 days before the end of the budget period; confirm in Research.gov), co-PI review, and the note that an overdue report can delay funding actions. Constraints: - Never invent results, citations, DOIs, or people; use [PI to add]. - Past tense for completed work, plain language, no em dashes.
Todoist Filter Query Builder: Saved Filters with today, overdue, p1, #Project and ##Subprojects, @Labels, /Sections, assigned to, created before, AND OR NOT Operators, Comma Split Lists, and a Weekly Review Set
Turn how you actually work into a small set of Todoist filters that answer one question each: what to do now, what is waiting on others, what is stale, and what needs a weekly review, with each query written in Todoist's filter syntax, explained piece by piece, tested against sample tasks, and paired with a label plan so the filters keep working.
Act as a Todoist power user and productivity coach who sets up task systems for professionals, writes filter queries in Todoist's own syntax, and keeps the number of saved filters small enough to use every day. Inputs: - Role and the kinds of work that compete for attention: [WorkContext] - Current projects and sub projects as named in Todoist, and sections inside them: [ProjectTree] - Labels in use or wanted (contexts, energy, waiting, people): [LabelPlan] - Shared projects and who assigns tasks to whom: [Collaboration] - Questions the filters must answer, such as "what now", "waiting on others", "stale tasks", "this week by area": [FilterQuestions] - Output format: [Format] Generate: 1. A label plan from LabelPlan: keep or rename each label, mark the one waiting label and any context labels, and drop labels that no filter will use. 2. One saved filter per item in FilterQuestions: name, query, color, and a plain reading of the query. Use Todoist syntax such as today, overdue, no date, 7 days, p1 to p4, #Project, ##Project to include sub projects, @label, /Section, assigned to: me, assigned to: others, due before: +3 days, created before: -30 days, search: word, and the operators & (and), | (or), ! (not), and parentheses. 3. Where one view should show stacked lists, use commas to split the query into separate sections, for example (today | overdue) & #Work, @waiting. 4. A test table: five sample tasks built from ProjectTree and LabelPlan, and which filters each task should appear in. Point out any task that shows up nowhere. 5. Edge cases: tasks with no date, recurring tasks, sub tasks, tasks in shared projects from Collaboration, and how each filter handles them. 6. A daily and weekly routine: which filter to open first each morning, and a weekly review sequence of three or four filters with what to do in each. 7. Setup steps: where to create filters, how to favorite them, and where to put them in the sidebar. 8. If a query relies on syntax you are not sure Todoist supports, label it "test in Todoist" and give a simpler fallback. Constraints: - No more than seven saved filters. Each name under four words. - Exact query strings in code style. No em dashes.
Short Term Rental Permit and Lodging Tax Compliance Planner for Hosts: City Permit, State Sales and County Occupancy Tax Registration, What Airbnb and Vrbo Collect vs What You Remit, and a Filing Calendar
Map every rule and tax that applies to one short term rental before the first guest arrives: the permit and license stack by city, county, and state, which lodging taxes the platforms collect and which the host must file, direct booking tax math, a filing calendar, and a checklist of listing, safety, and record keeping items, with each item marked confirmed or still to verify.
Act as a short term rental compliance consultant who sets up vacation rental hosts, reads city ordinances and county tax pages for a living, and refuses to state a tax rate or rule the host has not sourced. Inputs: - Property address area (city, county, state), property type, and whether it is a primary residence, a second home, or an investment unit: [PropertyLocation] - Rules found so far: ordinance excerpts, permit page notes, HOA or condo rules, zoning limits: [LocalRules] - Tax notes with sources: state sales or lodging tax, county or city occupancy tax, tourism district fees, and what each platform's help page says it collects there: [TaxNotes] - Booking channels and rough mix (Airbnb, Vrbo, direct bookings, other): [BookingChannels] - Nightly rate, cleaning fee, average stay, and expected nights per year: [RateCard] - Output format: [Format] Generate: 1. A jurisdiction stack table: level (state, county, city, HOA), requirement (permit, registration, business license, inspection, occupancy limit, night cap, primary residence rule), source from LocalRules, status (confirmed, needs verification), and deadline or renewal date. 2. A tax responsibility matrix: each tax from TaxNotes as a row, each channel from BookingChannels as a column, and in each cell "platform collects and remits", "host collects and remits", or "verify". Never fill a cell without a source in TaxNotes. 3. Registration steps in order: which accounts to open with the state and local tax offices, what ID numbers to put on the listing if the city requires it, and what to keep on file. 4. Direct booking tax math from RateCard: which charges are taxable (nightly rate, cleaning fee), a sample invoice line by line, and the total tax to collect. 5. A filing calendar: each return, frequency, due date pattern, and whether a zero return is still required for months the platform covered everything or nobody stayed. 6. A listing and safety checklist: permit number display, occupancy limit, parking and quiet hours, smoke and carbon monoxide alarms, fire extinguisher, emergency contact posting, and insurance that covers short term rental use. 7. Records: what to save each month (platform tax reports, direct booking invoices, filed returns) and for how long, per TaxNotes or "check with your tax office". 8. A risk list ordered by penalty exposure, with the first three actions for this week. Constraints: - No invented rates, thresholds, or deadlines; write [verify] gaps. Not legal or tax advice. - Tables first, short sentences, no em dashes.
Real Estate License Exam Practice Item Writer for Prelicensing Instructors: National and State Portion Topic Map, Four Choice Items, Proration and Commission Split Math, Agency and Fair Housing Scenarios, and Distractor Rationales
Write original practice questions for a real estate prelicensing course: items mapped to the national and state portions of the exam outline your students will sit, four choice stems with one defensible key, worked math for prorations, commissions, and loan to value, agency and fair housing scenarios, a rationale for every wrong answer, and a tally that keeps the set balanced to the blueprint.
Act as a real estate prelicensing instructor and item writer who builds practice exams from the candidate handbook outline, writes every question from scratch, and checks each math item twice by hand. Inputs: - State, exam vendor if known (for example Pearson VUE or PSI), and whether the set covers the national portion, the state portion, or both: [ExamPortion] - Content outline topics and weights pasted from the candidate handbook: [ContentOutline] - State law notes you teach from (agency disclosure form name, license law rules, deposit handling, contract forms): [StateLawNotes] - Number of items, difficulty mix, and how many math items: [SetSpec] - Conventions to state in math stems: 360 or 365 day year, who owns the day of closing, rounding rule: [MathConventions] - Output format: [Format] Generate: 1. A topic map from ContentOutline with the number of items per topic so the set matches the weights within one item. 2. Items, each with: ID, portion (national or state), topic, difficulty, a stem of one to three sentences, four choices A to D, the key, and a one line source note pointing to the outline topic or StateLawNotes line. 3. Stem rules: test one idea, avoid "all of the following except", no trick wording, no answers that depend on unstated local custom. State items must rely only on StateLawNotes. 4. Math items using MathConventions written into the stem: prorations, commission splits between brokerages and agents, loan to value and down payment, price per square foot, and simple capitalization rate. Show the worked solution step by step and make each distractor a real mistake (wrong day count, wrong split order, using sale price instead of loan amount). 5. Agency, disclosure, and fair housing scenarios that describe what a licensee says or does and ask what is required or prohibited. 6. A rationale for the key and for each distractor, one sentence each. 7. A blueprint tally table: topic, target, written, and difficulty spread, flagging gaps. 8. A note that the items are original practice items, not recalled exam content, and that state rules should be checked against the current candidate handbook and license law. Constraints: - Never reproduce or imitate recalled live exam items. - Plain US English, one correct answer, no em dashes.
Nonprofit Year End Fundraising Appeal Builder: Donor Segments (Current, LYBUNT, SYBUNT, Lapsed), Ask Strings from Last Gift, Print Letter and Reply Device, December Email Series, Matching Gift Deadline, and IRS Acknowledgment Language
Build a nonprofit's year end appeal from the donor file: segments with their own message, ask strings calculated from each donor's most recent or largest gift, a one page letter and reply device, a short December email series, honest matching gift language, and gift acknowledgment wording that meets the written acknowledgment rule for gifts of $250 or more.
Act as a nonprofit development director who runs year end appeals for small and mid sized charities, segments the donor file before writing a word, and writes to one reader about one result. Inputs: - Organization, mission in one sentence, and one specific story or result from this year with permission to share: [ImpactStory] - Donor file summary: counts and typical gifts for current year donors, LYBUNT (gave last year but not this year), SYBUNT (gave some earlier year), lapsed, and new prospects: [DonorSegments] - Goal amount, match details if any (who, cap, deadline, terms), and campaign dates: [GoalAndMatch] - Channels and budget: print mail volume, email list size, send tool: [Channels] - Gift options: online form link, check address, monthly giving, DAF and stock gifts: [GiftOptions] - Output format: [Format] Generate: 1. Segment plan: one row per segment in DonorSegments with the message angle (thank and upgrade for current donors, "we missed you this year" for LYBUNT, renewal for SYBUNT and lapsed, introduction for prospects), channel, and whether they get print. 2. Ask string rules: base the first ask on most recent gift (or highest previous gift for major donors), then roughly 1x, 1.5x, and 2x rounded to clean numbers, plus Other. Show the math for three sample donors and a default string for donors with no history. 3. Print letter of one page from ImpactStory: open with one person or moment, state the need, make one clear ask with the donor's first ask amount, mention the match only as GoalAndMatch states it, and add a P.S. with the deadline. 4. Reply device copy: name and address block, the personalized ask string, monthly giving checkbox, gift options from GiftOptions, and the note that mailed gifts count for the tax year when postmarked by December 31. 5. December email series: about five sends from Giving Tuesday to December 31, each with subject, preview text, one paragraph, and one button, ending with two short sends on December 30 and 31. 6. Match language that is exactly true: name the matcher, the cap, and what happens to gifts after the cap. If there is no match, use none. 7. Acknowledgment template: thank you within 48 hours, gift amount and date, and for gifts of $250 or more a statement of whether any goods or services were provided in return. 8. A results sheet: response rate, average gift, and dollars by segment, plus the date to pull LYBUNT donors for a January follow up. Constraints: - No invented statistics or beneficiaries; use [add result] gaps. - Warm, plain, second person. No guilt language, no em dashes.
Django 4.2 LTS to 5.2 LTS Upgrade Planner: Python 3.10+ Floor, Stepwise 5.0 and 5.1 Hops, python -Wa Deprecation Sweep, django-upgrade Codemods, STORAGES Setting, index_together and pytz Removals, and a Third Party Package Matrix
Plan a Django upgrade from 4.2 LTS to 5.2 LTS as a series of safe hops: environment floors, a deprecation sweep with warnings turned on, django-upgrade codemods, the removals that actually break 4.2 code, a package compatibility matrix, migration checks, and a deploy and rollback plan for each step.
Act as a senior Django engineer who has moved several production apps from one LTS release to the next, upgrades one minor version at a time, and lets the deprecation warnings drive the work list. Inputs: - Current versions: Django, Python, database engine and version, OS image: [CurrentStack] - requirements or pyproject dependency list with pinned versions: [Dependencies] - Settings excerpts and code patterns you know you use (storage settings, timezone handling, custom auth, model Meta options, test helpers): [SettingsAndPatterns] - Test suite size, CI setup, and how long a full run takes: [TestSuite] - Deploy method and tolerance for downtime: [DeployContext] - Output format: [Format] Generate: 1. Floors table: Django 5.x needs Python 3.10 or newer, and each release raises database minimums (for PostgreSQL, 5.2 expects 14 or newer). Compare against CurrentStack and list what must move first. 2. Hop plan: 4.2 to 5.0, then 5.1, then 5.2, each its own branch and deploy. Never jump straight to 5.2. 3. Warning sweep for each hop: run the suite with python -Wa manage.py test (or PYTHONWARNINGS=always with pytest), collect RemovedInDjango warnings, and fix all of them before bumping the version. 4. Codemods: run django-upgrade with --target-version for the hop, and pyupgrade for the Python floor, then review the diff by hand. 5. Breaking removals mapped to SettingsAndPatterns, for example: DEFAULT_FILE_STORAGE and STATICFILES_STORAGE replaced by STORAGES, index_together replaced by Meta.indexes with a migration, pytz support and USE_DEPRECATED_PYTZ removed in favor of zoneinfo, USE_L10N removed, logout over GET removed, assertQuerysetEqual renamed to assertQuerySetEqual, CheckConstraint check renamed to condition. For each: file, fix, test. 6. Package matrix from Dependencies: package, current pin, first version that supports the target Django, blocker yes or no, and an alternative if abandoned. 7. Migration safety: makemigrations --check after each hop, review any generated migration, and run migrations against a production sized copy. 8. Deploy and rollback for each hop from DeployContext, including what to watch in logs for the first day. 9. Mark anything you cannot confirm from the official release notes as "check the 5.x release notes". Constraints: - Ordered, checkable steps. Show commands exactly. No em dashes.
Health Insurance Claim Denial Appeal Letter Writer for Patients and Caregivers: Denial Reason Quote, ACA vs ERISA Plan Type, Internal Appeal Deadlines, Medical Necessity Evidence Packet, Claim File Request, and External Review Follow Up
Turn a denial letter (EOB or adverse benefit determination) into a firm, organized first level appeal: the exact denial reason quoted back, the plan type and deadlines that apply, a point by point rebuttal tied to the policy language and the doctor's letter, an evidence checklist, a claim file request, and a short external review request to send if the internal appeal fails.
Act as a patient advocate who writes health insurance appeal letters for patients and caregivers, reads denial notices line by line, and builds each appeal around the plan's own wording instead of emotion. Inputs: - Denial notice text: claim or reference number, date of notice, service, and the stated denial reason or code: [DenialNotice] - Plan type if known: employer self funded (ERISA), employer fully insured, ACA marketplace, or individual off exchange: [PlanType] - Policy or coverage language the insurer cited, pasted from the plan document or medical policy: [PolicyLanguage] - What the treating clinician says and what records exist (letter of medical necessity, notes, test results, prior treatments tried): [ClinicalEvidence] - Whether care is still pending or already received, and whether waiting could seriously harm health: [Urgency] - Output format: [Format] Generate: 1. A one line restatement of the denial reason in the insurer's own words, sorted into a type: medical necessity, experimental or investigational, out of network, prior authorization missing, coding or billing error, or benefit exclusion. Each type needs a different argument. 2. A deadline box: date of notice, the appeal window stated in the notice (for most ACA era plans 180 days from the notice), the decision clock to expect (pre service, post service, or urgent expedited), and the target send date with a margin of at least two weeks. 3. A routing note: if the denial is a coding or billing error, call the provider's billing office first and ask for a corrected claim, since an appeal may not be needed. 4. The appeal letter itself: member name, ID, claim number, date of service, a clear request (overturn and pay or authorize), a point by point rebuttal that quotes PolicyLanguage and shows how ClinicalEvidence meets each criterion, prior treatments tried and why they failed, and a request for review by a clinician in the same specialty. 5. A claim file request paragraph asking, free of charge, for the documents, records, and criteria used to decide the claim. 6. An enclosure checklist numbered to match the letter, with what to ask the clinician to add if the evidence is thin. 7. A send and track plan: method with proof of delivery, a call log template, and the date to escalate if no decision arrives. 8. A short external review request to use only if the final internal decision is still a denial, with the window to file stated as "check the final notice; often four months". 9. One honesty line: mark any deadline, law, or plan rule you could not confirm from DenialNotice or PolicyLanguage as "confirm with the plan or your state insurance department", and do not present the letter as legal advice. Constraints: - Never invent clinical facts, test results, or diagnosis codes; leave a [ask clinician] gap instead. - Calm, specific, first person. No threats, no em dashes.
Security Deposit Itemized Deduction Statement Writer for Landlords and Property Managers: Move-In vs Move-Out Comparison, Wear and Tear Screening, Prorated Repair Charges with Receipts, State Deadline and Delivery, and a Tenant Cover Letter
Prepare the itemized security deposit statement a landlord or property manager sends after move-out: compare move-in and move-out condition room by room, screen out normal wear and tear, prorate replacement costs by remaining useful life, tie every charge to a receipt or quote, do the deposit math, and follow the return deadline and delivery method from your own state research.
Act as a residential property manager who has processed hundreds of move-outs and defended deposit deductions in small claims court, and knows most deposit disputes are lost on missing move-in evidence, charging for normal wear and tear, or missing the state's return deadline. Inputs: - Deposit amounts, lease dates, the date the tenant returned possession, and forwarding address: [LeaseAndDeposit] - Move-in inspection report and photos, room by room: [MoveInReport] - Move-out findings with photo references: [MoveOutFindings] - Invoices, quotes, staff labor rate, and install dates of items being charged: [CostsAndReceipts] - State and local rules copied from the statute or city page with the date checked (return deadline, itemization, receipt rules, interest, delivery method, penalties): [StateRules] - Unpaid rent, utilities, or other lease charges with the lease clause: [UnpaidCharges] - Output format: [Format] Generate: 1. Key dates: possession return date, the deadline counted from StateRules, and a target mailing date with a safety margin. If StateRules has no deadline, write "NOT RESEARCHED" and stop that line. 2. A room by room comparison table: item, move-in condition, move-out condition, photo references, and what changed. Anything already noted in MoveInReport is pre-existing and cannot be charged. 3. A wear and tear screen for every change: normal wear and tear (no charge), damage, cleaning beyond move-in condition, or missing item, each with a one line reason. Faded paint, light scuffs, a few small nail holes, and worn traffic paths are wear. Burns, pet stains, holes, and broken fixtures are damage. 4. Charges: actual cost from CostsAndReceipts, labor at the stated rate and hours, and a remaining useful life proration for items with an install date (carpet, paint, appliances). Show the math on one line. 5. Unpaid lease charges from UnpaidCharges with the lease clause cited. 6. Deposit math: total deposits plus interest if StateRules requires it, minus deductions, equals refund or balance owed. 7. The itemized statement itself, ready for letterhead, with each line numbered and its receipt reference. 8. A neutral cover letter to the tenant: what is enclosed, the refund amount and method, how to ask questions or dispute, and a contact. 9. A delivery and records checklist: method from StateRules, proof of mailing, copies of photos, receipts, and the statement kept in the tenant file. Constraints: - Never state a deadline, interest rule, or penalty that is not in StateRules. This is a worksheet, not legal advice; recommend a local attorney or landlord association for disputes. - No round number fees, no charges without a receipt, quote, or labor log, and no em dashes.
Raycast Script Commands and Snippets Builder for a Mac IT Helpdesk: Scripts with @raycast Metadata, Output Modes, Typed Arguments, Confirmation for Risky Actions, Keychain Secrets, Snippet Placeholders, Quicklinks, and a Shared Repo
Turn the repetitive work of a small Mac support or IT team into Raycast tools: sort each task into a script command, a snippet, a quicklink, or a job for your MDM, then write ready to save Bash or Python script commands with correct @raycast metadata, the right output mode, typed arguments, confirmation on risky actions, tokens read from the macOS Keychain, snippet keywords with dynamic placeholders, and a shared repo with a test checklist.
Act as a Mac IT support lead who builds Raycast Script Commands, snippets, and quicklinks for a small helpdesk, and has cleaned up scripts that leaked API tokens into a shared repo or ran a destructive command on one stray keystroke. Inputs: - Team size, who uses the tools, and the tasks repeated every day: [TeamAndTasks] - Systems the scripts may touch, with base URLs and what needs admin rights: [ToolsAndAPIs] - Allowed script languages and installed command line tools (for example bash, python3, jq): [ScriptLanguage] - Where secrets live and what may never be committed: [SecretsPolicy] - Canned replies and text the team retypes: [SnippetLibrary] - Output format: [Format] Generate: 1. A triage table for TeamAndTasks: each task becomes a script command, a snippet, a quicklink, or "not a Raycast job" when it needs admin rights or belongs in MDM or Self Service per ToolsAndAPIs. One line of reason each. 2. For each script command, a complete file: kebab case filename, shebang, then the metadata comment block with @raycast.schemaVersion 1, @raycast.title (verb first), @raycast.mode, @raycast.packageName to group commands, @raycast.icon, and @raycast.description. 3. Pick the mode on purpose: silent for fire and forget, compact when one result line is enough, fullOutput for multi line results, inline only for a status line and only with @raycast.refreshTime. State the reason in a comment. 4. Arguments: at most three @raycast.argument lines with JSON type text, password, or dropdown, a placeholder, and optional when it truly is. Read them as $1 or sys.argv, validate, and on bad input print one plain line and exit 1 so Raycast shows the failure. 5. Risky actions (cancel, delete, restart, post to a system): add @raycast.needsConfirmation true and print exactly what will change before doing it. 6. Secrets per SecretsPolicy: never hard code a token. Read it with security find-generic-password -s name -w, and give each teammate the one line setup with security add-generic-password. If the item is missing, exit 1 with the setup hint. 7. Output discipline: last line is the human message in compact and silent modes, results that agents paste go to pbcopy, and no sudo inside any script. 8. Snippets from SnippetLibrary: a keyword scheme with a prefix such as ;; so words never expand by accident, then each snippet with dynamic placeholders like {cursor}, {clipboard}, or a date with an offset. Note that date offsets count calendar days. 9. Quicklinks for lookups by ID, using an argument placeholder in the URL. 10. Shared repo setup: folder layout, chmod +x, adding the folder with Add Script Directory in Raycast settings, a README line per command, and a test checklist (empty argument, bad argument, offline, missing Keychain item). Constraints: - Only call systems listed in ToolsAndAPIs. Mark any endpoint you did not get as "confirm in the API docs". - Placeholder and metadata syntax can change between Raycast versions; mark anything uncertain "check the current Raycast manual". No em dashes.
Septic System Maintenance Plan for Homeowners: Pumping Interval from Tank Size and Household, Pumper Visit Checklist with Sludge and Scum Measurements, Drain Field Protection, What Not to Flush, Failure Warning Signs, and a Records Binder
Keep a septic system working and avoid a five figure drain field replacement: estimate a pumping interval from tank size and household, know what the pumper should measure and check, protect the drain field from cars, roots, and roof water, change the habits that clog systems, recognize early failure signs, and keep the records a buyer or county inspector will ask for.
Act as a licensed septic inspector who also explains maintenance to new rural homeowners, and has seen most failed drain fields trace back to skipped pumping, a garbage disposal, or a driveway extended over the field. Inputs: - System details: tank size in gallons, tank material, conventional, mound, or aerobic treatment unit, age, effluent filter if known: [SystemDetails] - Household: number of people, garbage disposal, water softener, home business, guests or rentals: [Household] - History: last pumping date, any sludge and scum measurements, repairs, backups: [History] - Site: where the tank and drain field are, trees, driveway, roof drains, sump pump discharge: [SiteNotes] - Local rules from the county health department or permit records, including any inspection at sale: [LocalRules] - Output format: [Format] Generate: 1. A pumping interval estimate from SystemDetails and Household: typical household tanks are pumped every three to five years, with more people, a disposal, or a small tank pushing toward the short end. State the estimate as a starting point and explain that the pumper's sludge and scum measurements decide the next date. 2. A pumper visit checklist: pump through the main tank lid rather than only a small inspection port, measure the scum and sludge layers before pumping, inspect inlet and outlet baffles or tees, clean the effluent filter, note any water flowing back from the drain field, and leave a written report. 3. Aerobic unit extras when SystemDetails says so: service contract, alarm and air pump checks, and any disinfection component, marked "follow the manufacturer and county schedule". 4. Drain field protection from SiteNotes: no parking, driving, building, or patios over it, grass only, trees a safe distance away, roof drains and sump discharge pointed away from it. 5. Household habit changes: spread laundry over the week, fix running toilets and leaks, and a do not flush or drain list (wipes even when marked flushable, grease, paint, solvents, medications, large amounts of harsh cleaners). Explain why a garbage disposal adds solids and shortens the interval. 6. Additives: explain that routine additives are not a substitute for pumping, and check LocalRules before using any. 7. Warning signs and actions: slow drains through the house, gurgling, sewage odor, soggy or very green patches over the field, backups in the lowest drain. Say who to call and what to stop doing while waiting. 8. A records binder: the as built drawing or site plan from the county, a sketch with measured distances to the tank lids, every pumping report, repairs and permits, and what LocalRules require at sale. 9. A 10 year calendar with pumping estimates, filter cleanings, and reminders. Constraints: - Do not invent county requirements or tank sizes; mark unknowns [ASK COUNTY HEALTH DEPARTMENT] or [CHECK TANK LID OR RECORDS]. - This is maintenance guidance, not a design or repair plan; repairs need a licensed installer and usually a permit. No em dashes.
Spoonflower Seamless Repeat File Spec for Surface Designers: Tile Size Math at 150 DPI, Basic vs Half Drop vs Half Brick Repeats, Photoshop Offset and Illustrator Pattern Tool Setup, Scale Variants, Color Checks, and Proof Swatches
Prepare a surface pattern so it prints seamlessly on Spoonflower fabric and wallpaper: tile size math at 150 DPI, the right repeat type for the motif, step by step seam building in Photoshop or Illustrator, large, medium, and small scale versions, color checks against printed swatches because colors shift by fabric, and a proofing and listing checklist before selling.
Act as a surface pattern designer who sells fabric and wallpaper designs on Spoonflower, builds repeats in Photoshop and Illustrator, and has learned from printed swatches that a seam you cannot see on screen shows up across a full yard. Inputs: - Motif description, style, and how dense the layout is: [Motif] - Repeat you have in mind, or "advise": [RepeatType] - Products you want to offer (quilting cotton, apparel knit, linen, wallpaper, home decor): [Products] - Intended scale in real inches, such as motif height on the fabric: [ScaleTargets] - Palette with hex codes and any colors you love that may print dull: [Palette] - Software you use: [Software] - Output format: [Format] Generate: 1. Tile size math for ScaleTargets at 150 DPI, which Spoonflower recommends: inches times 150 equals pixels. Give the tile in inches and pixels, and how many repeats fit across a fabric width you confirm in the product listing. 2. Repeat choice for Motif and RepeatType: basic (block) for grids and plaids, half drop for organic florals because it hides rows, half brick for horizontal layouts, mirror for symmetrical damask looks, and center for placement prints. Explain whether to build the drop into the tile yourself or use Spoonflower's repeat setting. 3. Seam building in Software: in Photoshop, Filter, Other, Offset with Wrap Around by half the width and half the height, then fill the cross gaps; in Illustrator, Object, Pattern, Make, with Brick by Column at a 1/2 offset for a half drop. Include a check at 25 percent zoom across a 3 by 3 grid of tiles for tangents, gaps, and lines of motifs. 4. Scale variants for Products: a hero scale, a medium, and a small coordinate, each as its own file at the right pixel size, never upscaled from a smaller file. 5. Color from Palette: work in RGB, compare each hex to a printed Spoonflower color reference if you own one, flag very bright or neon colors and large dark areas that may print differently, and adjust on the swatch, not on the monitor. 6. Export: file type, color mode, and a note to "confirm current file size limits in Spoonflower help". 7. Proof plan: order test swatches on each fabric in Products before selling, and list what to check (seams, scale against a ruler, color, white areas on wallpaper). 8. A marketplace listing draft: design title, a short description that names the motif and scale, and tags that describe subject, style, color, and use. Constraints: - Do not invent fabric widths, prices, or limits; ask the user to confirm them on the product page. - Never upscale a tile to change scale; rebuild from vector or high resolution art. No em dashes.