32 Claude Skills for Running a Support Desk

What these skills are

A whole support desk, as installable Claude skills. The pack follows the work of a support desk in one arc. It opens on the hardest moment — an upset or abusive customer — then the team behind the queue: staffing, workload, handover. After that the roles and metrics that decide who gets blamed, the escalations and incidents that cross into other teams, what the tickets have been telling you, the answers and knowledge that stop repeat work, the design of the service itself, and last, how to add AI and bots without trapping anyone.

The red line: Claude drafts and routes; a person answers anyone who is upset, at risk or asking for an exception, and no customer is ever told a bot is a person.

Each skill is one named method or one artifact, attributed to its originator through a public source. Each works from what you paste — a ticket, a policy, a ticket export — with no connector to install, and each ends with the decision a named person makes, and by when. Nothing in the pack grades, ranks or profiles an agent or a customer: replies, tickets, processes and the service get scored, people do not.

Mechanically, each skill is one folder with a SKILL.md file. The 32 are grouped into eight parts of the desk's work, and each one gives full value on its own: run one, run a part, or work through the lot in the order the trouble arrives.

Download all 32 skills. One zip: ready-to-install skill zips, readable SKILL.md files, and the evidence notes behind every method. Free, no signup. The same 32 skills are open on GitHub: polar-bear-org/claude-skills.

The 32 skills you get

The set follows the work of the desk, in eight parts: hard conversations, workload and staffing, roles and metrics, escalations and incidents, learning from tickets, answers and knowledge, service design, and AI and bots.

1 · Hard Conversations

1. Abusive Customer Policy

Use when: Customers yell or threaten and no rule lets the agent end the contact

Output: An abusive contact policy: warning, final warning and end-contact steps, a script for each step by channel, a supervisor backup rule with a time limit, and an incident log

2. De-escalation Playbook

Use when: An angry customer is on the line and the agent does not know what to say next

Output: Phone and chat steps, phrases to use and phrases to avoid, the "I want your manager" path, and a handoff note so the customer does not repeat themselves

3. Service Recovery Plan

Use when: We made a wrong promise or a late delivery and must admit it honestly

Output: An owned-mistake reply, a fix and follow-up plan on two tracks, a gesture within agreed limits, and an apology letter

4. Complaint Handling Procedure

Use when: A formal complaint arrives and the answer depends on who picks it up

Output: Receive, acknowledge, investigate, answer, close and learn as steps, with the acknowledgement and outcome letters and a complaint log

5. Refund and Exception Policy

Use when: You enforce the rule, get yelled at, then a manager grants the exception anyway

Output: Public refund wording, a decision-rights table of who approves what up to what value, a precedent log, a never-do list, and kind ways to say no

2 · Workload and Staffing

6. Workload and Recovery Rules

Use when: The team is burning out and nothing anyone does ever counts

Output: An occupancy cap, breaks between hard contacts, recovery time after abuse, cover and shift-end rules, and a team check against six stress areas

7. Erlang C Staffing Plan

Use when: Tickets doubled, reply times tripled and leadership wants proof before hiring

Output: A volume forecast by interval, agents needed at a target service level, occupancy and shrinkage, workload maths for email and chat, and a one-page case for more hands

8. Backlog Recovery Plan

Use when: The queue grows every day and the oldest tickets are the angriest

Output: Triage by age and risk, a bulk-reply and merge plan, a pause list, and a daily burn-down

9. Shift Handover Template

Use when: You cannot take a holiday because everything is tied to you

Output: Open tickets in SBAR form with the owner and the promise made, who talks to each customer, and a holiday cover sheet

3 · Roles, Metrics and Coaching

10. RACI Matrix

Use when: Your job has become the dumping ground for everything nobody owns

Output: Who owns what across support, success, sales, product and engineering, the gaps and double owners flagged, and a rule for new asks

11. Support Metrics Scorecard

Use when: The metrics punish the wrong person for another team's delay

Output: A metric set that separates closed from resolved, first contact resolution paired with reopen rate, an SLA clock owner per stage, and what each number cannot tell you

12. QA Scorecard

Use when: Coaching arrives days after the call and agents already feel watched

Output: A reply rubric — accuracy, policy, clarity, tone, next step — a sample plan, a calibration session, and same-day coaching notes about replies

13. Customer Service Training Plan

Use when: New agents get two weeks of slides, then live calls, then blame

Output: A first-weeks ramp built on practice, a shadow and reverse-shadow schedule, a go-solo checklist based on tasks, and a check that training changed the work

14. Customer Service Role Play

Use when: You have no time to rehearse difficult customer conversations

Output: A scenario bank by contact type, Claude playing the customer, debrief questions, and a practice log for the team

4 · Escalations and Incidents

15. Escalation Matrix

Use when: Nobody knows who decides, so every hard ticket lands on the head of support

Output: Support tiers with their entry criteria, an impact and urgency priority grid, triggers and target times, and the information each route needs before a ticket moves

16. SLA and OLA Template

Use when: Breaches caused by another team get reported against support

Output: Customer targets by priority and channel, internal OLAs with engineering, product and billing, breach rules, and a named clock owner at every stage

17. Bug Report Template

Use when: You escalate a ticket and it sits because engineering cannot act on it

Output: Steps to reproduce, expected and actual result, impact and customers affected, urgency, a workaround, and what the customer was told

18. Release Readiness Brief

Use when: Product ships changes without telling support and confused tickets flood in

Output: A change table, the questions customers will ask, day-one saved replies and article updates, launch contacts, and a go or hold check

19. Incident Communication Plan

Use when: An outage with no ETA and every customer wants an answer now

Output: A first notice, an update cadence that always gives the next-update time, no-ETA wording, an all-clear, a blameless closing note, and an agent brief

5 · Learning from Tickets

20. CSAT Survey

Use when: A bad rating for the customer's own mistake gets counted against the agent

Output: A short post-ticket survey, a cause-coded comment sheet, and a monthly summary that names causes, not agents

21. Ticket Taxonomy

Use when: Repeated tickets about the same issue are a pattern nobody documents

Output: A two-level, reason-based tag tree with definitions and examples, a one-page tagging guide, and a retire list for dead tags

22. 5 Whys Root Cause Analysis

Use when: The same problem keeps coming back and product never hears it

Output: A Pareto table of contact drivers, a 5 Whys chain per driver that stops at a fixable process, and the asks for product with owners and dates

23. NPS Survey

Use when: Leadership asks what customers think of us overall, not of one ticket

Output: A relationship survey plan with its follow-up question, a close-the-loop rule where a person contacts detractors, and a cause table split into what support can and cannot move

6 · Answers and Knowledge

24. Canned Responses Library

Use when: You type the same answer again and again because it never reaches the docs

Output: Saved replies with personal slots, naming rules, an owner and a review date for each reply, and a retire list

25. Knowledge Base Article

Use when: Seniors become human search engines and the answer is buried

Output: An article from a solved ticket — the issue in the customer's words, environment, resolution, cause — with an article state, a gap list from failed searches, and a review cadence

26. Customer Service Policy

Use when: Every agent gives a different answer to the same question

Output: One page per topic — hours, channels, response promises, tone, refunds pointer, privacy basics — a change log, and a "what we never say" list

7 · Service Design

27. Customer Journey Map

Use when: Customers do not read, then contact us furious, five times

Output: Phases, actions, thoughts and feelings built from real tickets, the contact points where customers get confused and write in, and rewrites of the messages that cause it

28. Customer Effort Score

Use when: One problem takes five contacts and three transfers to solve

Output: The effort question and where to ask it, the effort drivers counted from tickets and comments — repeat contacts, transfers, channel switching — and a fix list with owners

29. Service Blueprint

Use when: Support takes the heat for handoffs the customer never sees

Output: Frontstage and backstage actions, support processes, the line of visibility, and the fail points and wait points, each with an owner

8 · AI and Bots, Safely

30. Chatbot Handoff Rules

Use when: The bot loops customers on cancellations with no human exit

Output: A bot scope table, handoff triggers with the context packet the agent receives, the disclosure line, and a loop test script

31. Self-Service Deflection Plan

Use when: Bots take the easy tickets and the hard remainder lands on an unchanged team

Output: A candidate question list, resolved versus abandoned measures instead of a deflection rate, and a second-line restaffing check

32. AI Use Policy

Use when: Leadership watches who uses AI and the team is unsure what AI is for

Output: A list of useful AI tasks with their risks, customer-data and send rules, the drafts a person must send, disclosure wording, and a review log

Where a desk starts, and what the pack rests on

There is no compulsory sequence. You bring one real thing — a chat log, a week of tickets, the refund policy you already have — and open the skill that matches what hurts this week. The order in the catalogue is the order the trouble usually arrives in: the abuse rule and the de-escalation words first, because that is the moment nobody has a script for, then the queue and the team carrying it, then the agreements with the teams whose delays land on support, then what the tickets have been saying all along.

Several of them chain. cs-ticket-taxonomy gives cs-five-whys the drivers to run on, and what that finds becomes cs-canned-responses and cs-knowledge-base-article, so the same answer stops being retyped. cs-escalation-matrix and cs-sla-ola are the pair that settles who owns a hard ticket and by when, and cs-bug-report is what makes an escalation actionable once it crosses into engineering. The bot skills come last on purpose: cs-chatbot-handoff-rules and cs-self-service-deflection only make sense after the taxonomy, because you cannot hand the easy questions to a bot until you know which ones they are.

The pack is research-informed, not validated. A scan of what support agents, leads and heads of support say they struggle with set the order of the skills and the words in each "use when"; the methods themselves come from public sources — KCS, ISO 10002, HSE guidance on abuse and on work-related stress, Erlang C, SBAR, RACI, ITIL service levels, the Kirkpatrick model, 5 Whys, and the EU AI Act's transparency article. The one shared file in the pack, resources/evidence-and-sources.md, lists them source by source with the finding used and the limit on it. No skill here has been tested in a trial, and the legal points — refunds, duties toward staff facing abuse, AI disclosure — end by telling you to check with a qualified adviser.

Setup guide

1. Download the pack. One zip: an install folder with 32 ready-to-upload skill zips, a skills folder with the same 32 skills as readable SKILL.md files, and the evidence notes — resources/evidence-and-sources.md — which name the source behind every method and where it stops.
2. Install your skills. In Claude Code, add the marketplace and install ai-for-customer-service-pack, and all 32 load at once. In Claude, turn on code execution in Settings, Capabilities, then go to Customize, Skills and upload one zip per skill from the install folder. Prefer working from files? Add the SKILL.md files to your Project knowledge instead; it works, just less cleanly.
3. Bring one real thing from the desk. No connector, no integration, no admin: every skill works from what you paste — a chat log, a week of tickets, an export, the policy you already have. Start with whatever hurts this week and write "run cs-de-escalation-playbook". Keep customer names and personal data out of it: anonymise the ticket first, and the skill asks for nothing it does not need.

Where to start

The quality bar

Every skill in the pack holds the same standard, the one we hold when we run a desk ourselves:

• Claude drafts and routes; a person answers anyone who is upset, at risk or asking for an exception, and no customer is ever told a bot is a person
• One named method or one artifact per skill, attributed to its originator through a public source
• Works from what you paste — a ticket, a policy, an export — and no connector is required
• Nothing grades, ranks or profiles an agent or a customer: replies, tickets, processes and the service get scored, and team data suppresses small groups
• No invented numbers: targets, thresholds and limits are yours to set, and every example is marked as an example
• Legal and regulatory points — refunds, abuse, AI disclosure — end with "check with a qualified adviser"
• Every output ends with the decision a named person makes, and by when

Who made this

Polar Bear is a people ops consultancy for human-size teams (20 to 200 people). Built by ex-McKinsey founders with a dream to make AI work for People, not instead of them. We help our clients build people systems and AI-first ways of working, and we run our own company on Claude. This pack is the free, self-serve version of how we work.

The pack carries one support desk. When you want your whole team working this way, AI carrying the overhead so people do the part only people can do, across support, hiring and everyday operations, that's what we build with clients.

Meet Pauline. Is your team absorbing every angry customer and every other team's delay? Bring a week of tickets, or the escalation nobody owns. Book a 30-minute call · Pauline on LinkedIn

Install in Claude Code

Two lines, and every skill in the pack loads at once.

/plugin marketplace add polar-bear-org/claude-skills
/plugin install ai-for-customer-service-pack@polar-bear-skills

Using Claude on the web instead? Download the zip and upload each skill from its install folder under Customize → Skills.