ع
Learn Tracks Reference Guides Saved
For your team

The Support Foundation toolkit: the files your team installs

A voice guide you wrote once helps you once. One your team installs helps every reply, forever. This is the toolkit that turns the Foundation module's three files into shared infrastructure — copy it, fill it with your team's voice, and every draft starts from the same truth.

8 min read · Updated 2026-06-30
The Support Foundation toolkit: the files your team installs

You finished the Foundation module with three documents: a support-voice.md, a support-policy.md, and a macros.md for your team. This toolkit makes them infrastructure. A file you wrote once helps you once; a file your team installs — that Claude reads on every task — helps every reply, forever.

Everything here is copy-able. Replace the Mizan placeholders (the GCC bookkeeping brand whose support team runs through the module) with your own, save the files in the folder your team works in, and open it in Claude Desktop. From then on, every support task in that folder starts from your real voice, your real authority tiers, and your real macro library — instead of each agent’s guess.

The whole pack is four files and a handful of prompts. That’s the point — a foundation small enough to actually get installed beats a beautiful one that lives in a doc nobody opens. Keep customer data out of it: this folder holds your team’s infrastructure, never a single ticket’s record.

The support CLAUDE.md

This is the standing context Claude reads automatically when it opens your project folder, so nobody has to re-explain the voice or policy every chat. Keep it short and current — it points at the three foundation docs rather than repeating them.

# CLAUDE.md — [Team name] support

## What we do
[Team name] is the support team for [Product] — [one-line product summary].
We answer tickets in [English and Arabic] from [customer base, e.g. 9,000+
GCC small-business customers].

## The three source-of-truth files — read these first
- support-voice.md  → HOW we write (register, signature phrases, apology
                       pattern, words we use / avoid)
- support-policy.md → WHAT we can do (3-tier authority table, escalation
                       line, never-say list)
- macros.md         → WHAT we say (the approved macro library)
Always read the relevant ones before drafting. If a draft conflicts with
them, the files win — flag the conflict, don't quietly override it.

## The golden rule
Claude drafts; a named person decides for any action that moves money,
changes a plan, or promises a date. You are the last filter before
the customer sees it. Never send a draft as written if the customer
name, account number, or refund amount looks wrong.

## Voice non-negotiables
- Open by naming the problem back — not with "Thank you for contacting us."
- Own the mistake in the first sentence. Never bury it.
- Short sentences. Plain words. No jargon.
- Never: "per our policy", "as a one-time courtesy", "we apologize for
  any inconvenience", "rest assured". See support-voice.md for the full list.

## Authority ceiling (hard stops — never draft past these)
- A solo agent never commits a refund above [your threshold, e.g. AED 150],
  an annual-plan change, a roadmap ETA, or any response to media or legal.
- When a task bumps a ceiling, draft the internal escalation note instead —
  not the customer reply. A manager clears it; then the agent sends it.
- Never admit legal fault. Never invent an order number, a ship date, or a
  policy that isn't in support-policy.md.

## DATA SAFETY — non-negotiable
Customer ticket data — names, emails, account numbers, ticket IDs — stays
in this workspace. Claude reads patterns to help draft replies; it is never
asked to export, summarize into a shareable doc, or process raw customer PII
outside this folder. If I paste a ticket, it is for this draft only.

## What's safe in this folder
- Safe: support-voice.md, support-policy.md, macros.md, anonymized ticket
  patterns, aggregate CSAT data, policy change logs.
- Never here: full customer ticket exports, raw email threads with PII,
  billing records, any file named "customers" or "leads". Tell me if a file
  looks like it has them.

On Desktop you create and edit this in the file pane — no commands. Anthropic’s docs call the “Ask permissions” mode the one recommended for new users; leave it on, and the first time Claude reads this file you’ll approve it in the prompt.

The support-voice.md template

The register every agent writes from — what the team sounds like, kept separate from what the team is allowed to do. The signature phrases and the words-to-avoid columns are the discipline: a voice without them is a vague instruction a busy agent can’t act on.

# support-voice.md — [Team name]

## Voice in four lines
1. [Name the problem first — before any explanation or apology.]
2. [Own the mistake early — first sentence, plain language, no hedging.]
3. [Short sentences. Plain words. Write like a knowledgeable friend, not a
   policy document.]
4. [Warm but efficient — the customer should feel heard and then helped, in
   that order.]

## Signature phrases (the words that make a reply sound like us)
1. "[Opening that names the problem — e.g. 'Your VAT export didn't go
   through — here's what happened and what we'll do.']"
2. "[Ownership phrase — e.g. 'This one's on us, and we'll fix it.']"
3. "[Empathy that's specific, not generic — e.g. 'Filing deadlines are real
   pressure. Let's clear this before yours.']"
4. "[Progress signal — e.g. 'Done — your account is updated.']"
5. "[Warm close — e.g. 'You're all set. Reach out any time.']"

## Apology and de-escalation pattern
Use this structure for any reply where something went wrong on our side:

  [SENTENCE 1 — Name the problem back, plainly.]
  [SENTENCE 2 — Own it. First person, no passive voice. "We got this wrong."
   Not "an error occurred" or "you may have experienced."]
  [SENTENCE 3 — The fix. What has been done or will be done, and by when
   if you can commit.]
  [SENTENCE 4 — What the customer needs to do next, if anything. If nothing,
   say so: "You don't need to do anything else."]
  [CLOSE — Invite them back. Short.]

Mizan example:
  "Your account was charged twice on March 1 — that's a billing error on our
   end, and I'm sorry it landed in your inbox. I've flagged it for our billing
   team and you'll see the duplicate charge reversed within [X] business days.
   You don't need to do anything else. Reply here if you have questions."

## Words we use
[warm], [clear], [specific], [your], [here's what happened], [done],
[you're all set], [let's fix that], [I can see why that's frustrating]
→ Replace this list with the 8–12 words and phrases that define your voice.

## Words we avoid
Avoid             → Say instead
"per our policy"  → "Our rule here is [plain statement]"
"inconvenience"   → name the actual problem ("the duplicate charge", "the
                    login error")
"rest assured"    → just say the reassuring thing
"as a one-time    → never say this; it sounds like charity, not service
 courtesy"
"unfortunately"   → usually just cut it; state the situation directly
"escalate"        → "pass this to [specific person/team]" (customer-facing)
"please note"     → cut it; make the note
[add your own]    → [the plain version]

The support-policy.md template

What the team is allowed to do, organized by authority tier. A policy doc that buries the ceilings in prose lets agents discover limits mid-reply — this table format puts the ceiling in the first place anyone looks.

# support-policy.md — [Team name]

## Authority tiers

| Tier      | Who              | Can do                                                        |
|-----------|------------------|---------------------------------------------------------------|
| Solo      | Any agent        | Refund ≤ [your threshold, e.g. AED 150]                      |
|           |                  | Apply account credits per credit policy                       |
|           |                  | Explain product features, pricing, and VAT treatment          |
|           |                  | Comp [X] month(s) for a clear [Product] error                 |
|           |                  | Reset passwords and restore access                            |
|           |                  | Log and confirm receipt of a bug report                       |
| Manager   | Named manager    | Refund > [your threshold]                                     |
|           |                  | Annual-plan changes (upgrades, downgrades, pauses, cancels)   |
|           |                  | Any promise about roadmap items or engineering ETAs           |
|           |                  | Any reply flagged by legal, media, or a regulator             |
|           |                  | Exceptions to any policy listed above                         |
| Never     | Anyone           | Admit legal fault or liability (refer to legal)               |
|           |                  | Invent an order number, a ship date, or a policy that doesn't |
|           |                  | exist in this file                                            |
|           |                  | Promise a feature the product team hasn't confirmed           |
|           |                  | Discuss a customer's account with a third party               |

## Escalation line
When a ticket hits Manager tier:
1. Add internal note: "[Agent name] — needs manager: [one-line reason]."
2. Assign to [manager queue / named person].
3. Send the customer an interim reply: acknowledge receipt, name a timeframe,
   don't promise the outcome. Example: "I've passed this to our team lead and
   you'll hear back within [X] hours."

Never let a ticket sit in-between — either handle it solo or escalate it.
A reply that says "I'll check" with no follow-up is the most common CSAT
driver in the wrong direction.

## Never-say list (customer-facing language)

Never say                          → Why / safe alternative
"per our policy"                   → It sounds like hiding behind a rule.
                                     Say: "Our rule here is [plain statement]."
"as a one-time courtesy"           → It frames service as charity. Just fix it.
"we apologize for any              → "Any" hedges the apology. Name the thing.
 inconvenience"                      Say: "I'm sorry the export failed."
"I'll escalate this"               → Vague. Say: "I'm passing this to [name/
                                     team] — you'll hear back by [time]."
"rest assured"                     → Just say the reassuring thing.
"unfortunately, we are unable to"  → Say what you *can* do, then the limit.
"please note that"                 → Cut it. Make the note.
"this is a known issue"            → If it's known, say what you're doing.
                                     Say: "We're aware and engineering is
                                     working on it — here's your workaround."
"I understand your frustration"    → Show you understand; don't narrate it.
                                     Name the problem instead.
[add your own]                     → [the safe alternative]

The macros.md template

The approved reply library — structure and guardrails, not scripts. Each macro has a “use when” line so agents reach for the right one without a manager’s help, and every customer-visible field in [brackets] is a fill-in, not optional.

Show the full macro for the three highest-volume ticket types. Add headers for the rest.

# macros.md — [Team name]

---

## M01 — VAT export / filing issue

Use when: Customer reports that a VAT export failed, is missing, shows wrong
figures, or won't open. Also use for "I can't find the VAT report" questions.

---

Subject: Your VAT export — [what happened and the fix]

[CUSTOMER NAME],

Your VAT export [describe the exact issue — e.g. "didn't generate for Q1" /
"shows AED 0.00" / "won't open as a PDF"]. Here's what to do:

[STEP 1 — the most common fix first. E.g.:
"Go to Reports → VAT Filing → select the correct quarter → Download PDF.
If the figure shows AED 0.00, check that your transactions are categorized
as VAT-applicable before exporting."]

[STEP 2 — if the first step doesn't resolve it:
"If the file still won't generate, reply here with a screenshot of the error
message and I'll look at your account directly."]

[OPTIONAL — if a Mizan-side bug is confirmed:
"This is a known issue we're working on. You can use the workaround above
while engineering fixes it; we'll notify you when it's resolved."]

[CLOSE — e.g. "Filing deadlines are real pressure — I want to make sure
this is clear before yours. Reply any time."]

[AGENT NAME]
[Team name] Support

---

## M02 — Login / password reset

Use when: Customer can't log in — password forgotten, account locked, MFA
issue, or "I don't recognise this email." Also use for team-member access
issues on multi-user plans.

---

Subject: Getting you back in — [product name] login

[CUSTOMER NAME],

[Name the exact login issue if the customer described it — e.g. "You're
locked out" / "Your reset email isn't arriving." If they didn't describe it,
open with "Let's get you back in."]

Try these steps:

1. Go to [login URL] and select "Forgot password."
2. Check your spam folder for the reset email — it comes from [sender address]
   and sometimes lands there.
3. [If MFA is the issue: "If you've lost access to your authenticator app,
   reply here with your registered email address and I'll verify your identity
   and reset it manually."]

If you're on a multi-user plan and a team member needs access restored, the
account owner can do this under Settings → Team → [path]. If the owner is
locked out too, reply here — I'll sort it.

[CLOSE — e.g. "You should be in within a few minutes. Let me know if
any step doesn't work."]

[AGENT NAME]
[Team name] Support

---

## M03 — Billing query (invoice / charge question)

Use when: Customer is asking what a charge is for, requesting an invoice,
questioning an amount, or asking about VAT on their subscription. For a
confirmed double-charge or error refund, use M04 instead.

---

Subject: Your [product name] invoice — [what they asked about]

[CUSTOMER NAME],

[Name what they're asking about — e.g. "You're asking about the AED [amount]
charge on [date]." Don't open with "Thank you for reaching out."]

Here's what that charge is:

[EXPLAIN THE CHARGE — plain language. E.g.:
"That's your [plan name] subscription renewal for [period]. The AED [amount]
breaks down as:
  - Subscription: AED [X]
  - VAT (5%): AED [Y]
  Total: AED [Z]"]

[IF THEY NEED AN INVOICE:
"Your invoice is in your account under Settings → Billing → Invoices. You
can download it as a PDF for your records."]

[IF THEY'RE DISPUTING THE AMOUNT AND IT LOOKS CORRECT:
"If this doesn't match what you expected, reply with your account email and
the amount you're seeing — I'll check against our billing records."]

[SOLO CEILING REMINDER — internal note, delete before sending:
If refund is needed and > AED 150, escalate per support-policy.md before
committing anything to the customer.]

[CLOSE — e.g. "Let me know if you need anything else on this."]

[AGENT NAME]
[Team name] Support

---

## M04 — Confirmed billing error / double charge
[header only — draft using the apology pattern in support-voice.md +
the solo/manager ceiling in support-policy.md. Own it first sentence.]

## M05 — Reconciliation help
[header only — include a link to the reconciliation walkthrough article
and offer a screen-share if the written steps don't resolve it.]

## M06 — Plan upgrade or downgrade
[header only — confirm what the customer wants, quote the new rate and
the proration, then route to manager if annual-plan change applies.]

## M07 — Bank sync failure
[header only — ask for the bank name and last successful sync date;
check the known-issues list before promising a fix timeline.]

## M08 — Slow load / performance issue
[header only — ask for browser, device, and the specific page affected;
offer the standard cache-clear steps before escalating to engineering.]

## M09 — Feature request
[header only — acknowledge it warmly, log it with the product team,
never promise it will ship or give a timeline engineering hasn't confirmed.]

## M10 — General / everything else
[header only — use the voice pattern from support-voice.md, name the
problem back, and ask the one question that unblocks the reply.]

The saved-prompt library (Foundation)

The recurring foundation jobs, as prompts you reuse. Paste them in the chat with the right files open; on the Terminal & Automation track these become slash commands so they’re one keystroke, but you need none of that to use them today.

Pressure-test a reply against voice and policy
  "Read support-voice.md and support-policy.md. Here is a draft reply:

   [paste the draft]

   Check it against:
   1. Voice — does it open by naming the problem? Own the mistake early?
      Use short sentences? Hit any words-to-avoid?
   2. Policy — does it commit anything above the solo ceiling? Say anything
      from the never-say list?
   Return a verdict (pass / flag) for each check and, for anything flagged,
   the exact sentence and what to change. Be direct — this goes to a customer."

Draft a macro for a new question type
  "Read support-voice.md, support-policy.md, and macros.md.
   We get a lot of tickets asking: [describe the question type].
   Draft a new macro in the same format as the existing macros — 'use when'
   line, subject line, body with [placeholders] for the fill-ins.
   Check it against the voice and policy files before finishing.
   Flag any place where the answer might hit the manager ceiling."

Score a reply against the rubric
  "Read support-voice.md. Score this agent reply out of 10 on each criterion:
   (1) names the problem back without making the customer repeat it
   (2) owns any mistake in the first sentence
   (3) uses short sentences and plain words
   (4) avoids the words-to-avoid list
   (5) closes with the customer set up to be done
   Give a number and one sentence of justification per criterion. Don't soften
   — this is a training exercise and the agent needs the honest read."

Adapt a reply to Arabic register
  "Read support-voice.md. Here is a reply in English:

   [paste the English reply]

   Adapt it into Arabic for a Gulf customer — warm MSA with Gulf-natural
   phrasing, not stiff fus'ha. Keep the same structure: name the problem,
   own it, fix it, close. Transliterate [Product] and technical terms like
   VAT, PDF, and Settings — don't translate them. Right-to-left output."

Refresh macros after a policy change
  "Our policy changed: [describe what changed — e.g. 'the solo refund ceiling
   moved from AED 100 to AED 150' or 'we added a new escalation step for
   annual-plan cancellations'].
   Read support-policy.md and macros.md. Tell me:
   1. Which macros reference the old policy and need updating.
   2. The exact sentences that need to change and what to change them to.
   3. Whether any macro now crosses a ceiling it didn't before.
   Propose a versioned diff — don't rewrite silently."

Installing it (Desktop, no setup)

  1. Save the four files — CLAUDE.md, support-voice.md, support-policy.md, macros.md — in the folder your support team works in.
  2. Open that folder in Claude Desktop and approve the read in the “Ask permissions” prompt.
  3. Start any task by opening the relevant docs; the prompts above do the rest.

That’s the whole install. The foundation you built in the module is now shared infrastructure — and Module 2 of the track extends this same pack with the triage and classification layer that builds on it. If you’re standing this up across a team rather than for yourself, read the operating guide for the data-safety, escalation, and rollout layer that belongs underneath.

Topics

Questions people ask

What is a CLAUDE.md and why does a support team need one?
It's a plain-text file Claude reads automatically when it opens your project folder — standing context about your product, your voice, your authority tiers, and your never-say rules, so you don't re-explain them every chat. For a support team it's the difference between every agent starting from the same trained instincts and every agent guessing the register from scratch. Drop it in the folder your team works in on Desktop and every reply in that folder inherits it.
Are these templates or finished files?
Templates — structure with the thinking baked in, but the content is yours to fill. The Mizan examples show you the shape of a good answer; you replace them with your own voice, your own policy tiers, and your own macro library. The point is that you fill in structure instead of facing a blank page, and you end with files in the exact format every Support playbook expects to be fed.
Do I need the terminal or any setup to use these?
No. On Claude Desktop you save these as files in your project folder, open the folder in the app, and approve the read in the "Ask permissions" prompt — that's it. The only optional power-user step is turning the prompts into slash commands, which lives in the Terminal & Automation track; the Desktop way needs nothing but the files.
How does this stay current as our product and policies evolve?
Treat the three documents as living files with one owner and a version bump on real changes — not a free-for-all anyone edits. Re-run the foundation prompts when your product, policies, or ticket patterns genuinely shift, and keep the CLAUDE.md pointing at the current versions. A policy file that silently goes stale can put the wrong authority in an agent's hands, so a quarterly check by the owner is the habit to build.
Put it into practice
Take the guided course
Start