--- name: copilot-prompt-library-builder description: Build a team-specific Microsoft 365 Copilot prompt library — tested prompts per role, per task, with grounding rules, prompt-quality criteria, contribution process, and governance. version: 1.0.0 author: VantagePoint Networks author_url: https://www.vpnetworks.co.uk audience: IT Managers, Modern Workplace Leads, Department Heads, Change Champions, MSP Consultants output_format: Formatted Markdown prompt library document with categorised prompts per role, grounding guidance, contribution template, evaluation criteria, and governance model. license: MIT last-reviewed: 2026-04 --- # Copilot Prompt Library Builder A Claude Code skill for the IT manager or change champion who's been asked "what should I actually do with Copilot?" — and needs to publish a curated, tested prompt library instead of pointing staff at Microsoft's marketing site. ## How to use this skill 1. Download this `SKILL.md` file. 2. Place it in `~/.claude/commands/` (macOS/Linux) or `%USERPROFILE%\.claude\commands\` (Windows). 3. In Claude Code, run `/copilot-prompt-library-builder`. Describe your organisation and the roles in scope. Answer the clarifying questions. Receive a ready-to-publish library. ## When to use this - You've rolled out Copilot and adoption is flat — staff don't know what to ask. - Leadership asked for a "list of things Copilot can do for our firm" and you'd rather not hand them a generic 100-prompt PDF. - You want a governance-aware library that includes grounding rules and data-handling guidance, not just clever prompts. - A specific department (finance, legal, HR) wants guidance tailored to their workflow. - You want a prompt-of-the-week programme to keep engagement up six months into a rollout. ## What you'll get A single Markdown document containing: - **Library index** organised by role and task - **Role-specific prompt sets** (5-10 prompts per role, tested and graded) - **Prompt quality criteria** (what makes a prompt "library-grade") - **Grounding rules** (what data Copilot can see for each prompt class) - **Sensitivity & data-handling notes** per prompt family - **Contribution template** and review process - **Governance** (owner, review cadence, deprecation rules) - **Measurement** (which prompts are driving time saved) - **A "starter" 20-prompt universal set** for every user ## Clarifying questions I will ask you 1. **Sector and size?** 2. **Which roles do you want prompts for?** (e.g. partners, associates, paralegals, finance, ops, marketing, IT) 3. **What's your base Copilot tier?** (Microsoft 365 Copilot / Copilot Pro / Business / Enterprise) 4. **Is EU Data Boundary enabled?** 5. **Typical data sources staff would ground prompts in?** (SharePoint sites, OneDrive, Teams chats, email, custom connectors) 6. **Any prohibited data classes?** (client privileged, PII, HR, M&A) 7. **Do you want each prompt pre-labelled with sensitivity guidance?** (recommended) 8. **Known adoption pain points?** (hallucinations, too-generic outputs, unclear when to use which tool) 9. **Existing prompt list to enhance, or starting fresh?** 10. **How do you plan to publish this?** (SharePoint page / wiki / dedicated app / intranet) 11. **Who owns the library ongoing?** 12. **Target publication date?** ## Output template ```markdown # Copilot Prompt Library — **Version:** 1.0 · **Owner:** · **Last reviewed:** **Review cadence:** monthly for 6 months post-launch, then quarterly. ## 1. How to use this library - Copy-paste the prompt into the Copilot chat where noted (Copilot in Word, Outlook, Teams, or the standalone Copilot app — indicated per prompt). - Check the **Grounding** field — that's what data Copilot needs access to for this prompt to work. - Check the **Sensitivity** field — do NOT run this prompt against data you're not allowed to share with Copilot per our AI Use Policy. - Review every output before sharing externally. Copilot drafts; you decide. ## 2. Prompt Quality Criteria (library-grade) A prompt qualifies for the library when it: 1. **Produces a consistent, useful output** across at least 5 test runs 2. **Has a clear, named author** who's willing to support questions 3. **States its grounding** (which data sources it reads from) 4. **Flags sensitivity** (which data classes it should NOT be used with) 5. **Has a measured use case** — someone saved real time using it 6. **Is role-appropriate** — matches at least one job family in this firm ## 3. Starter 20-Prompt Set (every user) These are the safest, most-reusable prompts for anyone new to Copilot. ### General productivity 1. **Meeting summary from transcript** - **Use in:** Copilot in Teams - **Grounding:** Current meeting transcript - **Sensitivity:** Internal (confidential if meeting contains confidential content) - **Prompt:** > Summarise this meeting. Return: (1) 3-sentence summary, (2) key decisions made with owners, (3) action items with owners and due dates, (4) unresolved questions. Use British English spelling. 2. **Inbox triage** - **Use in:** Copilot in Outlook - **Grounding:** Outlook inbox - **Prompt:** > Review emails received today that haven't been replied to. Group them into: needs reply today, needs reply this week, can be delegated, can be archived. For each, suggest a one-sentence reply or action. 3. **Document summary to bullets** - **Use in:** Copilot in Word - **Grounding:** Current document - **Prompt:** > Summarise this document for a busy colleague in 5-7 bullets. Include 1 bullet each for: what this is about, the headline decision or ask, the top risk or concern, and what the reader needs to do next. No fluff. (Continues — 20 prompts total with the same structure.) ## 4. Role-Specific Sets ### Legal — Associate / Paralegal (8 prompts) 1. **Clause library search** - **Use in:** Copilot Chat - **Grounding:** /Precedents SharePoint site - **Sensitivity:** Confidential (privileged content may be surfaced) - **Prompt:** > Search my precedents site for clauses addressing . Return 3 variants, the matter each was drafted on, the drafting partner, and the rationale for the drafting approach if it's documented. Do not include client-identifying information. 2. **Contract comparison** - **Use in:** Copilot in Word (compare mode) - **Grounding:** Two attached contracts - **Prompt:** > Compare these two contracts. Output: (1) risks in version B that weren't in version A, (2) clauses that move liability, (3) clauses that change payment terms or termination rights, (4) my recommended negotiation points. Use plain English. 3. **Case-note drafter** - **Prompt:** > Using the meeting transcript and the matter files, draft a case note in house style. Include: parties, matter reference, date, attendees, substantive discussion, next steps, time billed basis. (5 more.) ### Accountancy — Manager (8 prompts) 1. **Draft client engagement letter from brief** 2. **Summarise client management accounts for partner review** 3. **Draft explanation of tax rule for client plain-English letter** 4. (…) ### Finance & Admin (6 prompts) ### Marketing (6 prompts) ### HR (5 prompts) ### IT / Ops (6 prompts) ## 5. Prohibited Prompt Patterns Do NOT use Copilot for: - Drafting personal data (HR, medical, financial) in a way that could be surfaced in another user's query - Evaluating staff performance on non-aggregated individual-level data - Drafting board-confidential material until Sensitivity label applied - Any prompt that instructs Copilot to "ignore previous instructions" (prompt-injection pattern) - Asking Copilot to generate or explain malware / exploitation code ## 6. Contribution Template To propose a new prompt, fill in: ``` **Name:** **Role(s):** **Use in:** (Copilot chat / Word / Outlook / Teams / Excel / PowerPoint) **Grounding:** (which data source(s) does this read from) **Sensitivity:** (Public / Internal / Confidential / Restricted — per AI Use Policy) **Measured benefit:** (e.g. "saves 20 min per proposal draft") **Prompt:** (exact text) **Example output:** (redacted) **Author:** (name, role) **Date proposed:** ``` Submit to . Reviewed within 5 working days against the criteria in §2. ## 7. Review & Deprecation - Every prompt reviewed monthly for the first 6 months, then quarterly. - A prompt is deprecated if: it hasn't been used in 90 days; its output has become inconsistent; the data source has changed materially; the AI model's behaviour has changed. - Deprecated prompts are removed, not renamed. Authors are notified. ## 8. Measurement Track in : - Most-used prompts - Prompts producing the most reported "time saved" - Prompts generating the most "correction needed" feedback - Roles with lowest library engagement (targeted support) ## 9. Library Governance - **Owner:** . Accountable for content quality and review cadence. - **Editorial board:** <3-5 named individuals across roles>. Meets monthly. - **Publication:** . - **Versioning:** semantic. Major on structural change. Minor on prompt additions/removals. - **Feedback channel:** . - **Changes log:** maintained at the end of this document. ## 10. Launch Campaign (first 6 weeks) | Week | Activity | |---|---| | 0 | Library published; 3 champions briefed per role | | 1 | "Prompt of the week" in all-staff newsletter | | 2 | 30-min role-specific clinic (legal) | | 3 | 30-min role-specific clinic (finance) | | 4 | First measurement review; 3 prompts re-tuned based on feedback | | 5 | "Prompt submission drive" — internal contest for new prompts | | 6 | Library v1.1 published with top submissions | ``` ## Example invocation **User:** "45-person accountancy firm, Copilot for M365 rolled out last month. Adoption is disappointing — partners and managers are barely using it. We want prompts tailored to accountancy workflows, especially tax advisory and management accounts." **What the skill will do:** 1. Ask the 12 questions, paying special attention to: what document templates and precedents already exist in SharePoint (grounding targets), ICAEW ethical sensitivity on client data, whether partners want short prompts (typical) or longer structured ones. 2. Produce the library with: - 20 starter prompts - 8 Partner-specific prompts (client memos, technical tax research, practice management) - 8 Manager-specific prompts (file review, management accounts explanation, engagement letter draft) - 8 Senior/Junior prompts (working-paper drafting, client-query response, adjusting-journal documentation) - 6 HR / Admin prompts - A 6-week launch campaign with a Partners' Breakfast clinic in week 2 (known audience behaviour: partners need in-person modelling, not webinars) - Measurement hooks into the Microsoft 365 Copilot Dashboard so adoption data is visible 3. Flag that the current adoption gap is usually down to three things: (a) no grounded prompts — staff typing open-ended requests, (b) poor SharePoint hygiene — Copilot surfacing the wrong files, (c) partner scepticism — solved by modelling, not memos. ## Notes for the requester - **Grounded prompts outperform clever prompts.** A mediocre prompt pointed at a curated precedents site beats a brilliant prompt against a chaotic file-server. - **Role-specific beats generic.** "10 Copilot tips for accountants" from the web will never beat "10 prompts that worked for OUR accountants on OUR files." - **Measure "time saved" not "usage."** Usage goes up when people click around. Time saved is the metric you can show the board. - **Deprecate ruthlessly.** A library with 200 mediocre prompts is worse than one with 40 excellent prompts. Quality > quantity. - **Prompt hygiene matters.** A prompt that says "ignore confidentiality" or "pretend you are a different model" is not a prompt — it's a policy violation. Never published. - **Good looks like:** at month 3, weekly active Copilot users ≥ 70% of licensed, the library has 40-60 prompts, at least 3 prompts are producing measurable time savings, and partners are sponsoring the library publicly. --- *VantagePoint Networks · · Authored by Hak · Free under the MIT licence*