Reference

Gaiish Prompt Library

Reusable starting points for real work. Adapt the context and constraints; do not paste sensitive material into an unapproved system.

Last updated 2026-08-29 · Author

Business

Quarterly decision brief

INTENT:
Help the leadership team decide whether to continue the pilot.

CONTEXT:
[paste pilot metrics and customer notes]
The pilot has run for one quarter; separate observed results from customer opinions.

INSTRUCTION:
Summarise evidence for continuation, change or stop.

CONSTRAINTS:
Do not infer causation. Preserve metric definitions and flag missing baselines.

OUTPUT:
Recommendation, evidence table, risks and decision questions.

VALIDATION:
Check every number against the supplied metrics and list unsupported conclusions.

Why this works: The decision, evidence boundary and review shape are explicit, so the brief supports a real choice rather than a generic summary.

Process handoff

INTENT:
Give the next operator a reliable handoff for today's unresolved work.

CONTEXT:
[paste ticket list and notes]
The next operator has 30 minutes and can change status but cannot approve refunds.

INSTRUCTION:
Group the work by urgency and write the next action for each item.

CONSTRAINTS:
Do not close a ticket without evidence. Keep customer identifiers exactly as supplied.

OUTPUT:
Priority table with ticket, evidence, owner and next action.

VALIDATION:
Flag missing owners, missing evidence and any item needing approval.

Why this works: Context, authority limits and validation make the handoff useful to another person and safer to act on.

Management

One-to-one preparation

GOAL:
Help a manager prepare a constructive 30-minute one-to-one.

CONTEXT:
[paste role description and recent notes]
The employee has asked to discuss priorities and support.

ACTION:
Draft five open questions and a two-item agenda.

OUTPUT:
Agenda first, then questions with the purpose of each.

CONSTRAINTS:
Do not diagnose performance or infer feelings from silence.

VALIDATION:
Mark questions that rely on an assumption not present in the notes.

Why this works: The goal and boundaries keep the model from turning sparse notes into a performance judgement.

Team planning

INTENT:
Turn the team's committed work into a realistic fortnight plan.

CONTEXT:
[paste backlog and capacity notes]
Four people are available for 10 working days; production incidents take priority.

INSTRUCTION:
Group work by dependency, identify a critical path and propose a plan.

CONSTRAINTS:
Do not schedule more capacity than stated. Mark estimates as estimates.

OUTPUT:
Plan by day, dependency list and risks requiring a manager decision.

VALIDATION:
Recheck capacity totals and list every assumption.

Why this works: Capacity is context and a hard constraint; validation checks arithmetic instead of rewarding a confident schedule.

Information Technology

Architecture review

INTENT:
Identify the two largest operational risks in this proposed service architecture.

CONTEXT:
[paste architecture diagram and requirements]
The service must support the stated traffic range and a weekday on-call team.

INSTRUCTION:
Trace request flow, identify failure points and compare each with the requirements.

CONSTRAINTS:
Use only the supplied design. Distinguish a design gap from a missing diagram detail.

OUTPUT:
Risk table with component, failure mode, evidence, impact and open question.

VALIDATION:
Cite the diagram or requirement for each risk and mark unsupported assumptions.

Why this works: The source boundary and failure-oriented output make the review specific without pretending to know an undocumented deployment.

Runbook draft

INTENT:
Help an on-call engineer recover the documented service failure safely.

CONTEXT:
[paste incident notes and approved runbook fragments]
The engineer may restart the worker but may not change database schema.

INSTRUCTION:
Order the diagnostic and recovery steps, with a stop condition after each risky action.

CONSTRAINTS:
Do not invent commands. Preserve rollback steps and escalation contacts.

OUTPUT:
Numbered runbook with prerequisites, steps, stop conditions and escalation.

VALIDATION:
Check every command against the supplied fragments and flag any missing prerequisite.

Why this works: Role and permission context plus stop conditions turn prose into a safer operational artefact.

Marketing

Campaign brief

INTENT:
Plan a 30-day campaign that increases qualified demo requests from practice owners.

CONTEXT:
[paste product notes and existing campaign results]
Team: one marketer half-time. Audience: private dental practices with 2–6 chairs.

INSTRUCTION:
Propose channels, messages, weekly activities and a measurement plan.

CONSTRAINTS:
Budget is $5,000. No paid search or new engineering. Do not claim results not in the source.

OUTPUT:
Strategy, weekly plan, budget table, message examples and measures.

VALIDATION:
Check activities against budget and staffing and list assumptions.

Why this works: The campaign has a buyer, resource boundary, exclusions and measurement shape rather than a request for generic ideas.

Customer story interview

INTENT:
Prepare an interview guide that reveals why a customer adopted the product.

CONTEXT:
[paste account notes]
Interview length: 30 minutes. The customer has agreed to discuss workflow, not confidential finances.

INSTRUCTION:
Write eight open questions in a logical sequence.

CONSTRAINTS:
Do not lead the customer toward a preferred answer. Avoid questions about restricted information.

OUTPUT:
Question, purpose and optional follow-up for each item.

VALIDATION:
Flag any question that assumes a benefit or asks for restricted information.

Why this works: The goal, interview context and ethical constraints produce questions that can generate evidence instead of praise.

Education

Lesson activity

GOAL:
Help students practise distinguishing evidence from interpretation.

CONTEXT:
[paste source passage]
Students are 14 and have 25 minutes. They know the vocabulary in the lesson glossary.

ACTION:
Create a short worked example and a paired classification activity.

OUTPUT:
Teacher instructions, student handout and answer key.

CONSTRAINTS:
Use only the passage and glossary. Keep reading level accessible.

VALIDATION:
Quote the passage for every answer-key decision and mark ambiguous items.

Why this works: The output serves both teacher and student, while the source-only constraint makes the activity auditable.

Feedback rubric

INTENT:
Give a student actionable feedback on an argument draft without rewriting it for them.

CONTEXT:
[paste draft and assignment rubric]
The student is revising a first draft.

INSTRUCTION:
Identify one strength and the three highest-impact revisions, each tied to a rubric criterion.

CONSTRAINTS:
Do not invent citations or supply a replacement essay. Use encouraging, specific language.

OUTPUT:
Criterion, evidence in draft, revision suggestion and question for the student.

VALIDATION:
Check each comment against the draft and rubric and flag criteria not evidenced.

Why this works: The rubric and evidence columns keep feedback tied to the student's work and preserve student agency.

Research

Literature extraction

INTENT:
Create a traceable extraction table for studies relevant to the stated review question.

CONTEXT:
[paste papers or abstracts]
Review question: [question]. The supplied text is the only source authority.

INSTRUCTION:
Extract population, intervention, comparison, outcome and limitation for each study.

CONSTRAINTS:
Do not infer absent methods or results. Quote page or section references when present.

OUTPUT:
One row per study plus a Missing information column.

VALIDATION:
Compare every row with the source and list studies where the supplied text is incomplete.

Why this works: The evidence boundary, extraction schema and missing-information rule reduce the temptation to fill gaps.

Research question refinement

INTENT:
Turn the project idea into one answerable research question and a list of definitions to resolve.

CONTEXT:
[paste project notes]
Audience: supervisor reviewing feasibility. No data has been collected yet.

INSTRUCTION:
Identify the proposed phenomenon, population, comparison and outcome, then draft two candidate questions.

CONSTRAINTS:
Do not claim the question is novel or feasible without evidence. Preserve uncertainty.

OUTPUT:
Assumptions, candidate questions, what each would measure and feasibility questions.

VALIDATION:
Mark every claim that requires a source or a decision from the supervisor.

Why this works: The prompt separates a framing exercise from unsupported claims about novelty or feasibility.

Writing

Editorial outline

INTENT:
Give an editor a usable outline for an article that explains the decision behind the topic.

CONTEXT:
[paste source notes]
Audience: informed general readers. The article must distinguish sourced facts from interpretation.

INSTRUCTION:
Propose a thesis, section sequence and evidence needed for each section.

CONSTRAINTS:
Do not invent sources or settle disputed claims. Keep the outline under 500 words.

OUTPUT:
Headline options, thesis, numbered outline and open research questions.

VALIDATION:
Flag each section that lacks supplied evidence.

Why this works: The outline asks for the structure of an article while keeping research gaps visible.

Plain-language edit

INTENT:
Help a resident understand what action the policy requires of them.

CONTEXT:
[paste policy]
Audience: residents with no specialist knowledge. Legal review will use the edited draft.

INSTRUCTION:
Rewrite the policy explanation in plain language without changing obligations.

CONSTRAINTS:
Preserve dates, exceptions and defined terms. No more than 400 words. Do not add legal advice.

OUTPUT:
Heading, short explanation, required action and questions the resident may need to ask.

VALIDATION:
Compare every obligation and date with the source and list any ambiguous term.

Why this works: Audience, source authority and preservation constraints make “plain language” a controlled rewrite rather than a paraphrase.

Analysis

Spreadsheet interpretation

INTENT:
Explain the three material changes in this monthly operating report for a finance review.

CONTEXT:
[paste table]
Definitions: revenue excludes tax; variance is actual minus plan. The reader has not seen prior months.

INSTRUCTION:
Calculate stated variances, identify trends and separate observation from possible explanation.

CONSTRAINTS:
Do not invent causes. Preserve units and rounding. Show calculations.

OUTPUT:
Executive summary, variance table, observations and questions for the data owner.

VALIDATION:
Recalculate each variance against the supplied values and flag missing periods.

Why this works: Definitions in Context and calculation checks prevent familiar business language from hiding a changed metric meaning.

Options analysis

INTENT:
Help a project owner choose between the three options using the stated criteria.

CONTEXT:
[paste options and criteria]
The owner values reversibility and delivery time above optional scope.

INSTRUCTION:
Compare options criterion by criterion and identify the trade-off of the leading option.

CONSTRAINTS:
Use only the supplied criteria and facts. Do not convert qualitative claims into false precision.

OUTPUT:
Comparison matrix, leading option, trade-off and two questions before commitment.

VALIDATION:
Check each comparison cell against the source and mark unknowns.

Why this works: The prompt makes decision priorities explicit without pretending qualitative evidence is more precise than it is.

Productivity

Inbox triage

GOAL:
Turn today's inbox into a short list of actions for the owner.

CONTEXT:
[paste messages]
The owner has 45 minutes. Messages containing personal or confidential details must not be reproduced.

ACTION:
Group messages by action, urgency and waiting-on-person.

OUTPUT:
A table with message subject, action, due date if stated and suggested reply; omit message bodies.

CONSTRAINTS:
Do not infer urgency from tone alone. Preserve stated dates.

VALIDATION:
Check each due date against the source and flag ambiguous ownership.

Why this works: BASIC supplies a time limit, privacy boundary and usable table without turning triage into a new workflow.

Weekly planning

INTENT:
Create a realistic weekly plan that protects the two stated priorities.

CONTEXT:
[paste tasks and calendar constraints]
Available focus time is 12 hours. Meetings and fixed deadlines are listed below.

INSTRUCTION:
Sequence tasks by dependency and deadline and identify what must be deferred.

CONSTRAINTS:
Do not schedule more than 12 focus hours. Leave 20% unallocated for interruptions.

OUTPUT:
Day-by-day plan, deferred list and risks.

VALIDATION:
Check hours, dependencies and deadlines and list assumptions.

Why this works: The plan is constrained by real capacity and includes a deliberate deferral decision instead of fitting every task into an impossible week.

Software Development

Code review

INTENT:
Find correctness, security and maintainability risks before this change is merged.

CONTEXT:
[paste diff and relevant tests]
The code handles user-provided filenames. Existing conventions are in the adjacent module.

INSTRUCTION:
Review the diff, trace error paths and compare behaviour with the tests.

CONSTRAINTS:
Report only issues supported by the diff or supplied files. Do not rewrite the patch.

OUTPUT:
Findings ordered by severity with file/line, evidence, impact and suggested test.

VALIDATION:
Check each finding against the exact diff and distinguish a confirmed issue from a question.

Why this works: The review has a severity-aware output and evidence rule, helping a developer act on findings without accepting speculation.

Test design

INTENT:
Add tests that expose the boundary cases in the date parser.

CONTEXT:
[paste parser, current tests and accepted date formats]
The parser must reject ambiguous dates rather than guessing.

INSTRUCTION:
List partitions and write test cases for valid, invalid and ambiguous inputs.

CONSTRAINTS:
Use the accepted formats only. Do not change production code or invent a locale rule.

OUTPUT:
Test table with input, expected result, reason and missing coverage.

VALIDATION:
Check every expected result against the supplied format rules.

Why this works: The format rules and reject-on-ambiguity constraint turn a broad testing request into inspectable coverage.

Customer Service

Ticket categorisation

INTENT:
Route incoming tickets to the correct team without losing evidence of uncertainty.

CONTEXT:
[paste ticket text and routing rules]
Categories and escalation triggers are listed in the rules.

INSTRUCTION:
Assign one category, identify any escalation trigger and quote the supporting phrase.

CONSTRAINTS:
Do not infer a trigger absent from the rules. If no category fits, mark Unclear.

OUTPUT:
Ticket ID, category, evidence, escalation and confidence note.

VALIDATION:
Compare each assignment with the routing rules and list unclear cases.

Why this works: The routing rules become explicit knowledge and the Unclear outcome prevents forced classifications.

Support macro

INTENT:
Draft a reusable reply for customers who cannot complete the documented setup step.

CONTEXT:
[paste help article and approved policy]
Audience: new customers. The agent can link the article but cannot change account settings.

INSTRUCTION:
Draft a reply that diagnoses the documented step, links the relevant section and asks for one missing detail.

CONSTRAINTS:
Do not promise a fix or ask for passwords. Under 140 words.

OUTPUT:
Customer-facing reply and one internal escalation note.

VALIDATION:
Check links, permissions and promises against the supplied material.

Why this works: The prompt protects account boundaries, gives the customer a next step and separates the reusable response from escalation.

Human Resources

Interview guide

INTENT:
Create a structured interview guide that tests the stated competencies fairly.

CONTEXT:
[paste role description and competency rubric]
Interview length: 45 minutes. All candidates receive the same core questions.

INSTRUCTION:
Write one primary question and one follow-up for each competency.

CONSTRAINTS:
Do not ask about protected characteristics or infer them from a response. Keep questions job-related.

OUTPUT:
Question sequence, competency tested, evidence to listen for and permitted follow-up.

VALIDATION:
Check every question against the rubric and flag any question that is not job-related.

Why this works: The rubric, consistent core questions and explicit boundaries support a reviewable process without pretending a prompt makes hiring objective.

Policy explanation

INTENT:
Help employees understand the steps and contact point in the leave policy.

CONTEXT:
[paste approved policy]
Audience: employees; HR will approve the final explanation.

INSTRUCTION:
Rewrite the procedure in plain language while preserving eligibility and deadlines.

CONSTRAINTS:
Do not interpret an employee's individual eligibility. Preserve defined terms and escalation contacts.

OUTPUT:
Short explanation, numbered steps, deadline table and questions for HR.

VALIDATION:
Compare every deadline and eligibility statement with the policy and list ambiguous wording.

Why this works: The source and approval boundary make the rewrite useful while keeping individual decisions with HR.

Receive the Gaiish key concepts and definitions guide for reference.