Claude Skills are slash commands that run in Cowork mode. Type /skill-name and Claude executes a predefined workflow using your connected data. Install the skills below into your Cowork setup to get started.
How to install a skill
- Open Claude → Cowork mode
- Click Settings → Capabilities → Skills
- Click "Add Skill" and upload the
.skillfile - Type
/skill-namein any conversation to invoke it
Placeholder tokens in prompts use [BRACKETS]. Replace these with your own context before sending.
Email & Comms
Drafts, follow-ups, announcements, and the conversations that need care.
You are a senior business writer helping draft a clear, professional executive update. Context: - Project/Initiative: [name] - Status: [On track / At risk / Off track] - Period covered: [week/month] Write an executive update with these sections: 1. **Headline** (one sentence: what's the main story?) 2. **Key Decisions Made** (bullet points, 2-4 items) 3. **Progress This Period** (bullet points, what shipped or moved) 4. **Blockers & Risks** (what needs attention or a decision from leadership) 5. **Next Steps** (what happens in the next period, who owns what) Tone: direct, confident, no filler. Assume the reader has 90 seconds.
Help me write a team announcement about [the change you're making]. Context: - What is changing: [describe the change] - Why it's happening: [the reason, be honest] - When it takes effect: [date] - What stays the same: [what people can count on not changing] - Who this affects: [all staff / specific team / leaders only] Write a short announcement (200-300 words) that: 1. States the change plainly in the first sentence 2. Gives the honest reason 3. Answers "what does this mean for me?" 4. Ends with a clear next step or contact for questions Avoid: corporate hedging, passive voice, burying the lead.
I sent an email to [name/role] on [date] about [topic] and haven't heard back. Original message summary: [paste or describe what you sent] Write two follow-up options: Option A: Gentle nudge (assume they're just busy): - Short (3-4 sentences), low pressure, easy to respond to - Give them an easy out if the topic is no longer relevant Option B: Direct follow-up (appropriate if time-sensitive): - States clearly that you need a response by [date] - Explains why the deadline matters - Offers a specific action (reply, call, forward to someone else) Tone: professional and human. Never passive-aggressive.
I need to have a difficult conversation with [role/relationship] about [the issue].
Situation:
- What happened: [describe specifically]
- What outcome I need: [what I'm hoping changes or gets resolved]
- What I'm worried about: [their likely reaction, my concern]
- Relationship context: [direct report / peer / client / vendor]
Give me three different ways to open this conversation:
Option 1: Collaborative framing ("I want us to solve this together")
Option 2: Direct and clear ("Here's what I'm observing and what needs to change")
Option 3: Question-led ("I want to understand your perspective first")
For each: write the opening 2-3 sentences and one follow-up line if they get defensive.
A client is upset or escalating an issue. Help me draft a response. Situation: - What happened (from their perspective): [describe the issue as they see it] - What actually happened (your side): [the real sequence of events] - What you can offer: [refund / expedited fix / account credit / explanation / call] - What you cannot offer: [anything off the table] Write a response email that: 1. Opens by acknowledging their frustration without excessive apology 2. Takes responsibility for what was genuinely our fault (be specific) 3. Explains (briefly) what happened, no excuses, just facts 4. States clearly what we're doing to fix it and by when 5. Closes with a concrete next step Tone: calm, accountable, human. No corporate boilerplate. No "we apologize for any inconvenience."
Summaries & Recaps
Turn raw notes, long documents, and endless threads into clarity.
Here are raw notes from a meeting. Turn them into a clean summary. Meeting: [meeting name] Date: [date] Attendees: [names or roles] Raw notes: [PASTE YOUR NOTES HERE] Return: 1. **Meeting Purpose** (one sentence) 2. **Key Decisions Made** (bullet points, decisions only, not discussions) 3. **Action Items** (table format: | Task | Owner | Due Date |) 4. **Open Questions / Parking Lot** (things that need follow-up but weren't resolved) 5. **Next Meeting** (if mentioned) If the notes are ambiguous about an owner or deadline, flag it with [TBD] rather than guessing.
Summarize the following document for a busy executive who needs to understand it in under 2 minutes. Document type: [contract / report / proposal / policy / other] Audience: [who will read this summary] [PASTE DOCUMENT TEXT HERE] Return: 1. **What this document is** (one sentence) 2. **Key points** (bullet points, maximum 5, most important facts, numbers, dates) 3. **What action is required** (if any: what does the reader need to decide or do?) 4. **Risks or red flags** (anything unusual, ambiguous, or worth scrutiny) 5. **Bottom line** (one sentence recommendation or verdict) Be direct. If something is concerning, say so.
Here is an email thread (or Slack conversation). Summarize it clearly. [PASTE THREAD HERE] Return: 1. **What this thread is about** (one sentence) 2. **Timeline of key events** (chronological bullet points, decisions and changes only, skip small talk) 3. **What was decided** (explicit decisions, with who decided) 4. **What is still unresolved or needs a response** (flag who is responsible for the next step) 5. **Any commitments made** (promises, deadlines, deliverables mentioned) If someone's message is ambiguous about whether it was a decision or just a suggestion, flag it.
Help me write my weekly status update. My role: [your title or team] Audience: [your manager / leadership team / the whole org] Tone preference: [concise and bulletted / narrative / mixed] This week's notes (paste raw, messy is fine): [YOUR NOTES HERE] Write a weekly status with: 1. **This Week: What I shipped or completed** (bullet points) 2. **Next Week: What I'm focused on** (bullet points with priority order) 3. **Blocked / At Risk** (anything that needs attention from someone else) 4. **Metrics** (if relevant: list any numbers worth reporting) Keep it scannable. No filler. If I didn't give you enough info for a section, write "[nothing to report]" rather than padding it.
Research & Analysis
Comparisons, structured analysis, briefs, and risk identification.
Help me compare [2-4 options] and make a recommendation. Options: [list them] Decision context: [what are you actually deciding, and why does it matter] Key criteria (most important first): 1. [criterion 1] 2. [criterion 2] 3. [criterion 3] 4. [anything else] Must-haves (deal-breakers if missing): [list any hard requirements] Nice-to-haves: [list preferences that aren't blockers] Return: 1. A comparison table (options as columns, criteria as rows) with a rating or short note per cell 2. A summary of where each option wins and where it falls short 3. Your recommendation: state it clearly, with the top 2-3 reasons Be opinionated. I want a recommendation, not a "both have merits" non-answer.
I'm weighing a decision and want a structured analysis. Decision: [describe what you're deciding] Context: [relevant background: constraints, timeline, stakeholders] Options I'm considering: - Option A: [describe] - Option B: [describe] - Option C (if applicable): [describe] For each option, give me: - Pros (concrete advantages, not generic) - Cons (real risks and downsides, not hedged) - Who benefits and who loses Then give me: 1. Your recommended option, stated directly 2. The single strongest argument for that recommendation 3. The biggest risk of going with your recommendation and how to mitigate it Don't soften the recommendation. If the right answer is obvious, say so.
Write a brief on [topic / company / market / regulation]. Purpose: [why I need this: investor meeting / board update / sales call / internal research] Audience: [who will read it] Depth needed: [executive summary only / full brief / deep dive on one specific angle] Cover: 1. **What it is** (plain-language definition or description, assume the reader is smart but unfamiliar) 2. **Current state** (what's happening right now: key players, numbers, trends) 3. **Why it matters** (implications for [my company / my client / my decision]) 4. **Key risks or watch items** (what could change, what's uncertain) 5. **Bottom line** (one-paragraph "so what" for the audience) Cite sources where possible. Flag anything you're uncertain about rather than stating it as fact.
I'm about to [describe the plan, project, or decision]. Help me identify the risks. Context: - What we're doing: [describe] - Timeline: [when it starts, key milestones] - Team / resources: [who's involved, what you have] - Constraints: [budget, regulatory, political, technical] Return a risk register with: | Risk | Likelihood (H/M/L) | Impact (H/M/L) | Mitigation | Then: 1. Flag the top 3 risks I should address before proceeding 2. Identify any assumption I'm making that, if wrong, would blow up the whole plan 3. Suggest one thing I should do in the next 7 days to reduce risk Be direct about risks that are serious. Don't soften them to protect my feelings about the plan.
Planning & Strategy
Ramp plans, project plans, agendas, and pre-mortems.
Help me build a 30-60-90 day plan. Context: - Role or initiative: [describe] - Starting point: [what's already in place / what you're inheriting] - Goal: [what success looks like at 90 days] - Key stakeholders: [who matters: boss, team, clients, board] - Known constraints: [budget, headcount, politics, timelines] Build a plan with three phases: **Days 1-30: Learn** - Priorities (what to understand before doing anything) - Key meetings to have - Success metric: [what does "done" look like for this phase] **Days 31-60: Build** - What to start building, changing, or establishing - Quick wins to demonstrate value - Success metric **Days 61-90: Lead** - What to deliver, launch, or own - How to measure impact - What the 90-day review conversation looks like Flag any assumptions I should validate in week 1 before the plan locks in.
Help me build a project plan. Project: [name and one-sentence description] Goal: [what does success look like, be specific and measurable] Deadline: [hard deadline or target date] Team: [roles or names of people involved] Budget: [if relevant] Key constraints: [things that are fixed: tech stack, regulations, dependencies] Return: 1. **Project summary** (2-3 sentences: goal, scope, and what's out of scope) 2. **Milestones** (5-8 key checkpoints with target dates) 3. **Task breakdown** for each milestone (bullet points, enough detail to assign work) 4. **Owners**: which role owns each milestone 5. **Dependencies**: what has to happen before what 6. **Top 3 risks** with mitigation 7. **Definition of done**: how you know the project is complete Format the milestones as a simple timeline table.
Build an agenda for this meeting. Meeting type: [decision meeting / status update / brainstorm / kickoff / retrospective / other] Duration: [30 min / 60 min / 90 min / half day] Attendees: [roles or names: include seniority level if it affects format] Objective: [what must be decided or accomplished by the end of this meeting] Topics I know need to be covered: - [topic 1] - [topic 2] - [topic 3] Return: 1. A time-boxed agenda (| Time | Topic | Owner | Goal |) 2. For each agenda item: the specific question we're trying to answer (not just a topic name) 3. Pre-work I should send to attendees in advance 4. Any agenda items I should cut if the meeting risks going long Note any topics that shouldn't be in a meeting at all and should be handled asynchronously instead.
We're about to [describe the decision, launch, or commitment]. Run a pre-mortem. It is [date 6 months from now]. The [project / decision / launch] failed badly. What went wrong? Generate 10-15 plausible failure scenarios. For each: - The failure mode (what went wrong) - The root cause (why it actually happened) - The early warning sign (what you could have noticed at week 2 that predicted this) Then: 1. Group the failures into categories (execution / assumptions / external / people / resources) 2. Identify the top 3 most likely failures based on what you know about the context 3. For each of the top 3: what one action in the next 7 days reduces that risk most? Be specific and adversarial. Generic risks like "scope creep" are only useful if you explain the specific mechanism for THIS project.
Session Documents
Live working docs from the session: copy-paste prompts and the shared questions page
Two shared Google Docs from the working session. The first is a ready-to-use prompt library; the second is open for anyone to drop in questions to answer during or after the session.
📋 Session Prompts & Links →
Copy-paste prompts used in the session, organized by slide.
❓ Ask a Question →
Open document. Paste any question you want answered during or after the session.
The 4D Framework
Anthropic's AI Fluency framework: working with AI effectively, efficiently, ethically, and safely
AI Fluency is the ability to work with AI effectively, efficiently, ethically, and safely. Its four core competencies - the 4Ds - were defined by Professors Rick Dakan and Joseph Feller in collaboration with Anthropic.
| Competency | What it means | In practice |
|---|---|---|
| Delegation | Set the goal and decide whether, when, and how to engage AI - and how to split the work between you and Claude. | Is this task right for AI at all? Which model or tool fits? What stays human? |
| Description | Describe your goal clearly enough to prompt useful behavior and output. | Give role, context, format, and constraints. Vague in, generic out. |
| Discernment | Assess the AI's output and behavior critically before you rely on it. | Is it accurate, appropriate, and does it sound like you? Confidence is not accuracy. |
| Diligence | Take ethical responsibility for the final product. | Verify facts, keep a human in the loop on anything that sends or deletes, own the result. |
Green · Yellow · Red
The data sensitivity standard, use it for everything
Before you paste anything into Claude, run a 3-second check. These three tiers cover 95% of situations.
🟢 Green: Share freely
- Public information, internal operational data, generic analysis
- Examples: meeting agendas, project plans, industry research, general business questions, internal process docs with no client names
- Behavior: paste directly into claude.ai, use any connector, include in shared Projects
🟡 Yellow: Proceed with care
- Client names + business context combined, internal financials, HR information, unreleased strategy
- Examples: "Client X is struggling with Y," salary data, M&A discussions, board materials
- Behavior: use in a private Claude Project (not shared), anonymize before pasting if possible, do not attach documents, paste relevant excerpts only
🔴 Red: Off limits
- PII, government IDs, payment data, attorney-client privileged material, HIPAA-covered data, passwords, API keys
- Examples: SSNs, credit card numbers, patient records, legal advice memos
- Behavior: do not enter into Claude under any circumstances. Full stop.
- Client name + a number attached to them? Yellow minimum.
- Off a client device or from a client system? Red.
- When you hand an agent the keyboard: one sandboxed folder, read the plan before you approve it.
Is It Automation-Ready?
Three questions before you build anything
Run this filter on any task before you spend time building a workflow. All three need to be "yes."
Question 1: Does it repeat?
Daily, weekly, or every time a trigger fires. If you do it the same way more than once, it's a candidate. One-offs don't justify automation investment.
Question 2: Can you write the rules?
If you could hand this to a smart new hire with written instructions and they'd get it right on the first try, Claude can follow those same instructions. If it requires live judgment, reading the room, or institutional knowledge that can't be written down, keep it Assisted, not Automated.
Question 3: Is it eating real time?
30+ minutes per week, or something the team dreads. The ROI has to justify the build. Automating a 5-minute task saves less than the hour you spend building it.
The Quick-Win Matrix
| Low Effort | High Effort | |
|---|---|---|
| High Impact | ✅ Do first, these are your first 30 days | ⏳ Later: Phase 2 |
| Low Impact | Maybe, only if trivial | ❌ Don't build |
Build your own assistant in about five minutes. Everything is on this page: www.stephenthorn.com/claude. Download the one plugin below, then follow the five steps. Replace any [BRACKETS] with your own context.
Ground rules baked in
- Read-only on shared documents. The assistant analyzes; you apply edits.
- Nothing auto-sends. Drafts are staged for your review.
- It won't invent numbers, names, or meetings, and it cites sources.
- Want org-wide search instead? “Ask [your org]” is a pre-built Project that searches every connected tool at once and answers with sources.