prompting-practice
/prompting-practice
Uczy pisać skuteczne prompty do Microsoft Copilot bezpiecznie: podawać cel, kontekst, źródła i oczekiwany wynik oraz sprawdzać cytowania i liczby.
- Repozytorium
- CopilotForge ↗
- Licencja
- CC-BY-4.0
- Ostatnia zmiana repozytorium
- 03.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 2
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 03.11.2026, bez odwołań do wycofanych modeli i API.
CC-BY-4.0, otwarta.
Co robi i kiedy
Uczy pisać skuteczne prompty do Microsoft Copilot bezpiecznie: podawać cel, kontekst, źródła i oczekiwany wynik oraz sprawdzać cytowania i liczby.
Oryginalny opis (EN)
Teaches people to prompt Microsoft Copilot effectively and safely - stating goal, context, sources and expected output, iterating, choosing what not to enter - and to verify outputs by checking citations, numbers and omissions before use. Produces a prompting guide, a verification checklist and prompt library standards. Use when users get generic or wrong answers, when a prompt library is being built, or when verification habits are weak.
Jak działa
- Purpose
- When to use
- When not to use
- Questions to ask first
- Required inputs
- Decision framework
- Practitioner approach
- Outputs
- Microsoft platform considerations
- Adoption considerations
Podgląd SKILL.md
--- name: prompting-practice description: Teaches people to prompt Microsoft Copilot effectively and safely - stating goal, context, sources and expected output, iterating, choosing what not to enter - and to verify outputs by checking citations, numbers and omissions before use. Produces a prompting guide, a verification checklist and prompt library standards. Use when users get generic or wrong answers, when a prompt library is being built, or when verification habits are weak. license: CC-BY-4.0 metadata: department: m365-copilot lifecycle: Enable, Improve maturity: explorer, adopter, scaler, ai-enterprise risk-tiers: 0, 1 platforms: copilot-chat, m365-copilot edition-support: starter, scale, enterprise reviewers: none sources: ms-copilot-overview, ms-copilot-adoption-hub, ms-copilot-privacy, ms-responsible-ai status: draft authors: Simon Foote --- # Prompting practice ## Purpose Most disappointment with Copilot comes from two habits: asking vague questions and trusting the answers without checking. This skill builds the opposite habits across the workforce. It teaches a simple, durable way to ask Copilot for help, what to keep out of a prompt, how to improve an answer by iterating, and how to verify an output before relying on it. It also sets the standards for any shared prompt library. It goes wrong when prompting is taught as a bag of tricks or magic phrases, which age quickly and leave people no better at judging the output. ## When to use - Users report that Copilot gives generic, shallow or wrong answers. - Someone is building a shared prompt library and needs standards for what goes in it. - Verification is weak: outputs are forwarded without checking sources, numbers or omissions. - Champions, trainers or managers need a consistent way to teach prompting. ## When not to use - Writing the instructions that define an agent's behaviour - use `m365-agents:agent-instructions`. - Engineering and versioning system prompts for a production Foundry solution - use `ai-foundry:foundry-prompt-engineering`. - Designing role-specific, app-specific scenarios - use `m365-copilot:copilot-role-scenarios`. - Building formal role curricula and learning pathways - use `learning-and-skills:ai-role-curriculum` or `learning-and-skills:ai-learning-pathway`. - Facilitator materials for running sessions - use `learning-and-skills:facilitator-guide`. ## Questions to ask first 1. Which audiences are in scope, and which Copilot experiences do they have (Copilot Chat, licensed Microsoft 365 Copilot or both)? 2. What does the approved-use guidance say must never be entered into a prompt or referenced? 3. Where are people going wrong today: vague prompts, missing context, no iteration, or no checking? 4. Is there an existing prompt library, and who owns its quality? 5. Through which routes will people learn (onboarding, Champions, office hours, short videos)? ## Required inputs - The organisation's approved-use guidance and data classification. - Which Copilot experiences each audience has, confirmed by IT. - Examples of real prompts and outputs that went wrong, shared voluntarily and anonymised. - Any existing prompt library and the scenarios from `copilot-adoption:scenario-discovery`. If approved-use guidance is missing, the "what not to enter" section cannot be written; flag it to the policy owner and do not invent rules. ## Decision framework Teach a small set of habits and check each one in practice. | Habit | What good looks like | Signal it is missing | |---|---|---| | State the goal | The prompt says what the output is for and who will read it | Generic answers that fit any situation | | Give context and sources | The prompt names the files, meetings or threads to use, or attaches them | Answers drawn from the wrong material, or invented detail | | Set expectations | Format, length, tone and structure are specified | Repeated reformatting by hand | | Iterate | The person refines with follow-up prompts instead of starting again | One-shot use, then abandonment | | Keep out what must stay out | Prohibited content is never entered or referenced | Sensitive data appears in prompts against guidance | | Verify before use | Citations opened, numbers recalculated, omissions considered | Outputs forwarded unchanged with errors | Prompt library entries must meet a standard: a named scenario and audience, editable placeholders explained in words, the sources it expects, a verification step, an owner and a review date. Entries without a verification step are not published. ## Practitioner approach Teach prompting as a conversation with a capable but uninformed colleague. Copilot does not know the purpose, the audience or which of a person's hundreds of files matter unless told. A simple frame works across apps: goal, context, sources, expectations. "Draft a two-paragraph update for the steering group on the supplier delay, using the notes from Tuesday's meeting and the latest risk log, in a neutral tone" beats "write an update" every time. Avoid naming the frame after a product feature; it should still work when the interface changes. Show iteration explicitly. Many people abandon Copilot after one disappointing answer. Demonstrate asking for a shorter version, a different audience, a table instead of prose, or "what have you not covered?". Iteration is where most of the quality comes from. Spend as long on verification as on prompting. Teach people to open the cited sources and confirm the claim is there, to recalculate any number that matters, to ask themselves what is missing, and to be most careful when the output sounds most confident. Give them a short verification checklist they can keep beside them. Explain why: Copilot can only use what the person can access, can miss recent or poorly stored content, and can still be wrong about content it did find. Be concrete about what stays out. Translate the approved-use guidance into examples people recognise in their role, and tell them what to do instead. Rules that only exist in a policy document are not followed. Use real examples. Collect, with permission and anonymised, prompts that worked and prompts that failed in the organisation's own work, and use them in sessions. Pair each with its verification step. For any shared prompt library, apply the entry standard and give it an owner. Libraries rot quickly; review entries on a cycle, retire those nobody uses, and prefer fewer, well-tested prompts linked to scenarios over hundreds of untested ones. Measure capability through demonstrated practice at group level, never by inspecting individuals' prompts. ## Outputs - A prompting and verification guide for the organisation, in plain language with role examples. - A one-page verification checklist. - Prompt library standards and an initial set of reviewed entries, linked to the `scenario-catalogue`. - Short teaching materials that Champions and trainers can reuse. ## Microsoft platform considerations Copilot Chat and licensed Microsoft 365 Copilot differ in the work context they can draw on, and each app offers different ways to reference files, meetings and messages. Capability and interface details change often; keep teaching materials anchored on habits and check `ms-copilot-overview` before showing interface specifics. How prompts and responses are handled is described in `ms-copilot-privacy`. ## Adoption considerations Prompting improves through practice, not slides. Embed the habits in onboarding, office hours and Champion challenges through `copilot-adoption:launch-plan` and `champions:champion-challenges`, and reinforce them with short, regular examples rather than one long course. ## Knowledge considerations Not applicable in depth, because the skill teaches individual habits; where many people repeatedly need the same curated knowledge, route the need to `m365-agents:agent-opportunity-assessment` instead of a longer prompt. ## Security and privacy considerations The "what stays out" section must match the approved-use guidance exactly, and examples must be anonymised. Prompts and responses may be stored and discoverable under the organisation's retention policy; tell people this plainly, based on the current source records and the records management position. ## Responsible AI considerations People remain accountable for what they do with an output. Teach that Copilot drafts and suggests; it does not decide. Prompts that ask Copilot to judge or rank colleagues, or to make decisions about people, are outside approved use unless the organisation has an approved process. ## Evaluation requirements Check the guide with a small group from different roles: can they apply the frame to their own task and catch a planted error in an output using the checklist? Revise until most can. ## Value measures Measure at group level: confidence and reported quality before and after sessions, the share of published library entries with verification steps, and reductions in support questions about poor answers. The adoption lead owns these measures. Never review or score an individual's prompts. ## Scale notes ### Starter One short guide, one checklist, and a monthly drop-in session run by the Copilot lead or a partner. ### Scale Guide and checklist embedded in onboarding, a curated prompt library with owners, and Champions running practice sessions by role. ### Enterprise Localised guides by region and language, a governed library with business-unit curators, and links into formal learning pathways. ## Hand-offs - To `learning-and-skills:ai-role-curriculum` to embed the habits in role curricula. - To `champions:champion-challenges` for practice challenges. - To `m365-copilot:copilot-role-scenarios` for app-specific scenario cards. - To `m365-agents:agent-opportunity-assessment` when a prompt pattern becomes a shared, knowledge-heavy need. ## Reviewer gates Not applicable as a formal gate, because the guide covers individual use at low risk tiers; the "what stays out" section is checked with the `security-privacy` department before publication. ## Failure modes - Prompting is taught as magic phrases that stop working when the product changes. - Verification is mentioned once and never practised, so errors flow into real work. - The prompt library grows into hundreds of untested entries that nobody owns. - Examples use real sensitive content, teaching the opposite of the guidance. ## Sources `references/sources.md` in this skill lists the source records behind this skill, what each is authoritative for, how volatile it is and what you may do with it. Product availability, licensing, limits and preview status change often, so check the current source before stating them, and cite what you used. ## Never - Teach prompting without teaching verification. - Publish library prompts without an owner, a scenario and a verification step. - Inspect, score or rank individuals' prompts. - Use real sensitive content in teaching examples.
Pliki w paczce (2)
- SKILL.md 11 KB
- references/sources.md 2 KB
Wersje
W bazie nie ma innych skilli o nazwie prompting-practice.
Migawka bazy z 13.11.2026 · wersja paczki d76dd28b5321 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 95Prompting
Tworzy dynamiczne prompty z szablonów i wzorców, także przez programowe składanie i optymalizację promptów. - 95Prompting
Generuje zoptymalizowane prompty z rolą, kontekstem i formatem wyjścia na podstawie szablonów i wzorców. - 100copilot-role-scenarios
Zamienia priorytetowe zadanie roli na konkretne scenariusze Copilota w aplikacjach Microsoft 365, z podaniem promptów, sprawdzenia i znanych ograniczeń. - 95Prompting
Tworzy, optymalizuje i składa prompty programowo przez standardy, szablony Handlebars i narzędzia; wynikiem jest zawsze prompt do użycia w innym miejscu. - 95Prompting
Generuje, optymalizuje i składa prompty programowo przy użyciu standardów, szablonów Handlebars i narzędzi.