Get the reviews that fill next week: 10 AI skills for your reputation

reputation-rota

who answers what, within how long, in writing

How the two work together

Claude thinks it through. Paste the Claude prompt into Claude Code, or drop the folder into your skills folder. Claude does the judgement: what to look for, what is worth doing, what is right.

Codex gets it done. At the hand-off point Claude runs Codex on your machine with one command and passes it the Codex prompt. Codex does the mechanical part and hands the result back. Claude checks it before you see it.

No API key to set up: Claude calls the Codex you already have installed. If Codex is not installed, Claude does that half itself and tells you.

Prompt for Claude

---
name: reputation-rota
description: Writes the standing rota for who answers reviews and public comments at a hospitality business, what each category of review gets, how fast, in whose name, and which ones stop being a reply and become a food safety, employment or legal matter. Produces the rota, the routing table, the published review policy the law expects a publisher to hold, and a log that proves it was followed. Use when replies are sporadic, when nobody knows who answers a one-star, or when a serious allegation has appeared in public and nobody knew whose job it was.
---

# Who answers what, within how long, in writing

You give this the list of places reviews and public comments arrive, who works there, what they are allowed to say, and any review still unanswered. You get back four things: a rota naming a person and a deputy for every source and every day, a routing table saying what each category of review gets and in what time, the published policy on reviews that a business publishing reviews is expected to hold, and a log that records what was answered, by whom and when. It will not write the replies, and it will not let a review alleging illness, injury, discrimination or a criminal act be handled as a reply at all.

## What it does

1. **List every place a review or a public comment can arrive, and name one person for each.** Google, the booking platforms, the delivery platforms, the venue's own feedback form, the social accounts including direct messages, the comment card box and the voicemail. For each one record who holds the login, who is notified when something arrives, and how. A source nobody is notified about is not covered, and writing "the manager" in the column is what produces a three-week-old one-star that everybody assumed somebody else had seen.

2. **Verify the profiles first, because on some platforms an unverified account cannot reply at all.** Google states it plainly: "Before you can reply to reviews, you must verify your business. After your business is verified, you can reply to reviews on your Business Profile." Record the verification state of every source with a date, and where a platform has no reply function at all, record that too and name what the venue does instead, because a rota that quietly assumes every source can be answered leaves a gap nobody sees.

3. **Sort every review into categories before you set any timing, because the timings are different and the routes are different.** Six categories carry a UK hospitality venue: praise, a service complaint, a factual error about the business, an allegation of illness, injury or an allergic reaction, an allegation about a member of staff including discrimination or assault, and content that breaches the platform's own policies. Only the first three are replies in the ordinary sense. Write the category definitions down and give each one an example from the venue's own history, because a rota sorted by a definition nobody recognises gets sorted wrongly at eleven o'clock on a Saturday night.

4. **Set a target time per category, publish it internally, and count it in the log.** The target is the venue's own operating decision, not a benchmark, and it should be one an understaffed Tuesday can meet rather than one that looks good on paper. What the rota must fix is that the time starts when the review is posted rather than when somebody noticed it, that each category has its own clock, and that a missed target appears in the log as missed rather than being quietly reset. Google's own statement of why it matters is as far as any published claim goes: "When you reply to customer reviews, it shows that you value their feedback."

5. **Route an allergen or illness allegation out of marketing entirely, on the day it appears.** This is the step that justifies the whole document. A review saying a diner declared an allergy and was served the allergen is a food safety matter first. The FSA requires that "Food businesses must make sure that staff receive training on allergens", that a business providing non-prepacked food "must supply allergen information for every item that contains any of the 14 allergens", and that the business knows what is in its food, "You can do this by recording allergen ingredient information in a written format." So the routing table sends such a review to the person responsible for food safety the same day, opens the internal record, checks the dish specification and the allergen matrix, and produces a public reply only after that, written so it neither admits nor denies anything the investigation has not established. The same route covers any report of illness after eating.

6. **Route an allegation about a person out of public view and into the employment process.** A review naming or identifying a member of staff and alleging rudeness is a management conversation; one alleging discrimination, assault or theft is a formal process with the accused person's own rights in it. Neither is answered by a public reply that comments on an individual. The rota names who receives these, what the holding reply says, and the rule that it says nothing about the individual, because a public sentence about a named employee is both an employment risk and, on Google's rules, likely to be personal information: the platform prohibits content containing "full name, or last name, their face in a photograph or a video, or other information which has been reported as having been posted without consent."

7. **Write the reply rules once, so nine people write in one voice without nine conversations.** What every reply does: thank, name the specific thing, say what has been done or will be, and offer a private route. What no reply ever does: argue the facts of a service the writer experienced and the replier did not, reveal that the person is a customer or what they ordered, name a member of staff, offer anything in exchange for changing the review, or repeat a claim about the business that nobody can evidence. Replies are public and reviewed: Google notes that "If your reply is approved, it will be publicly posted under the customer's review", and that it checks replies against its content policies. Whose name they go out in is a decision to record, and the rota records it per source.

8. **Write the published review policy, because a business that publishes reviews is expected to have one.** The CMA's guidance is direct: "Publishers should all have published policies which prohibit fake reviews. Such policies should also set out the publisher's approach to incentivised reviews and consumer review information", and those policies "should be easily accessible, with signposting from relevant parts of the medium used (for example, a website) and not tucked away in a hard-to-find place. They should also be written in plain English." The same chapter sets three prohibitions that go straight into the rota as rules, because a well-meaning manager breaks them daily. Publishers should not "try to persuade consumers to submit a complaint, rather than leave a review for publication", should not "try to dissuade consumers from leaving a review of their experience even if their initial problem has been resolved through the complaints process", and should not "treat a negative review intended for publication as a complaint and not publish it".

9. **Log every review and every reply, and review the log monthly.** One row per review: source, date posted, category, rating, who it was assigned to, the date it was answered, whether the target was met, the reply in full, and where it was routed if it left the reply process. The monthly read is not a report to anybody, it is four questions: which category is being missed, whether one source has no coverage on particular days, whether the same complaint has appeared three times and is therefore an operations problem rather than a reply problem, and whether anything was routed to food safety or employment and then never closed. The CMA expects publishers to "regularly evaluate the effectiveness of these steps" and, where they find inadequacies, to address them. Date the rota, name who owns it, and set the date it is next rewritten.

## Then it checks

1. Every source of reviews and public comments has a named person, a named deputy and a stated notification route, with no entry reading a job title alone.
2. The verification state of every source is recorded with a date, and any source that cannot be replied to at all is named with what the venue does instead.
3. Every category has a written definition, an example from this venue, a target time and a named route, and every route ends at a person rather than a department.
4. Allergen, illness, injury, discrimination and criminal allegations are routed out of the reply process on the day they appear, with the internal record opened before any public reply is written.
5. No rule anywhere in the file permits a reply that names a member of staff, reveals what a customer ordered, offers anything in exchange for a review being changed or removed, or discourages anybody from leaving a review.
6. The published review policy exists, is linked from the page where reviews appear, prohibits fake reviews, states the venue's position on incentivised reviews, and is written in plain English.

Any check fails: name it, redo that step once. Failed twice: say what is wrong and stop.

## Rules
- Public information only.
- Never invent a fact, a number or a quote.
- Anything sent in someone's name says whose name it is. A reply going out as the landlord is approved by the landlord, and the rota records whose name each source replies in.
- Never write, approve or suggest any message that offers a refund, a voucher, a free item, a discount or any other benefit in exchange for a review being changed, softened or removed. The CMA names this as commissioning a banned review, and it is a banned practice under Schedule 20 to the Digital Markets, Competition and Consumers Act 2024.
- Never treat a review alleging illness, an allergic reaction, an injury, discrimination or a criminal act as a reply to be drafted. Route it, open the record, and let the public wording follow the investigation rather than lead it.
- Never draft a public reply that names, identifies or comments on an individual member of staff, and never confirm in public what a reviewer ate, drank, paid or booked.
- Never discourage a review, steer a complaint away from the review box, or ask somebody to take a review down as the condition of putting something right.
- Never state a target response time as an industry standard, and never state a figure for the bookings, calls or ratings that replying produces. No such figure exists for a single independent venue, and a target borrowed from a chain with a contact centre will be missed every week until the rota is abandoned.
- Where a review involves food safety, an employee, a possible discrimination claim or a threat of legal action, this document prepares the matter for the owner's environmental health officer, HR adviser or solicitor to handle. It is a routing document, not the advice, and it does not settle whether anything alleged is true.
- This output is a working document prepared for the owner to check against their own staffing, their food safety management system and their employment procedures before it is relied on. It applies published regulatory guidance and platform rules to one business's process. It is not legal advice and it is not a finding that the business complies with anything.

## Built from
- Competition and Markets Authority, "Fake reviews", CMA208, guidance on the prohibition under paragraph 13 of Schedule 20 to the Digital Markets, Competition and Consumers Act 2024, https://assets.publishing.service.gov.uk/media/67eeb64fe9c76fa33048c790/CMA208_-_Fake_reviews_guidance.pdf, published 4 April 2025, read 16 September 2026: the published-policy requirement and its accessibility and plain English conditions in step 8, the three prohibitions on steering complaints away from reviews in step 8, the refund-for-removal example in the Rules, and the duty to evaluate the process regularly in step 9.
- Digital Markets, Competition and Consumers Act 2024, Schedule 20, paragraph 13, https://www.legislation.gov.uk/ukpga/2024/13/schedule/20, read 16 September 2026: the banned practice covering commissioning a review that no longer reflects a genuine experience, and the positive duty to take "reasonable and proportionate steps" to prevent and remove banned reviews, behind step 8 and the Rules.
- Food Standards Agency, "Allergen guidance for food businesses", GOV.UK, https://www.gov.uk/government/publications/allergen-guidance-for-food-businesses/allergen-guidance-for-food-businesses, published 2 April 2020 and updated 17 July 2026, read 16 September 2026: the 14 allergens, the duty to supply allergen information for non-prepacked food, the staff training requirement and the written recording of allergen ingredient information, all in step 5.
- Google, "Manage customer reviews", Google Business Profile Help, https://support.google.com/business/answer/3474050, no publication date shown on the page, read 16 September 2026: the requirement to verify before replying in step 2, and the public posting and policy review of replies in step 7.
- Google, "Prohibited & restricted content", Maps User Generated Content Policy Help, https://support.google.com/contributionpolicy/answer/7400114, no publication date shown on the page, read 16 September 2026: the personal information prohibition behind step 6, and the conflict of interest and incentive prohibitions behind the Rules.
- Google, "Tips to improve your local ranking on Google", Google Business Profile Help, https://support.google.com/business/answer/7091, no publication date shown on the page, read 16 September 2026: the only published statement the skill makes about why replying matters, in step 4.

Prompt for Codex

# reputation-rota

## You are given
One UK hospitality business and the raw material for its review rota. The list of every source where reviews or public comments arrive, each with its type, the account holder, the notification route and the verification state with a date. The staff list with, for each person, their role, the days and hours they work, whether they are authorised to publish a reply, and whose name they reply in. The category definitions Claude has written, with an example from this venue against each, a target response time in hours and a named route. The named holders of the food safety, employment and legal routes with their deputies. The backlog of unanswered reviews, each with its source, date posted, rating and full text, and the category Claude has assigned to each. The reply rules Claude has written. The published review policy text Claude has drafted, and the page it is to be linked from. You build the rota, count the coverage and record the log. You never assign a category, never write a reply, never write policy text and never set a target.

## Produce
Write into a `./reputation-rota-output/` folder:

1. `sources.csv` with these columns in this order: `source_ref`, `source_name`, `source_type`, `account_holder`, `notification_route`, `verified`, `verification_date`, `reply_function_available`, `alternative_if_no_reply`. `source_ref` is S001 upward. `source_type` is exactly one of `search profile`, `booking platform`, `delivery platform`, `own feedback form`, `social account`, `social direct message`, `comment card`, `voicemail`. `verified`, `reply_function_available` are `yes`, `no` or `not applicable`.
2. `rota.csv` with columns: `source_ref`, `day_of_week`, `named_person`, `named_deputy`, `hours_covered`, `replies_in_name_of`, `authorised_to_publish`. Seven rows per source, one per day. `authorised_to_publish` is `yes` or `no`. Any cell holding a job title with no personal name is listed in `gaps.md`.
3. `categories.csv` with columns: `category`, `definition_verbatim`, `venue_example_verbatim`, `target_hours`, `route_to_named_person`, `route_deputy`, `is_reply_or_route`, `internal_record_required`. `category` is exactly one of `praise`, `service complaint`, `factual error about the business`, `illness injury or allergic reaction`, `allegation about a member of staff`, `platform policy breach`. `is_reply_or_route` is `reply` or `route`. The last three categories must read `route` and `internal_record_required` must read `yes` for each of them, or the row is listed in `gaps.md`.
4. `coverage-check.csv` with columns: `source_ref`, `day_of_week`, `covered`, `deputy_present`, `gap_reason`. `covered` is `yes` where a named person and named deputy both appear for that source and day. One row per source per day, 7 times the source count in total.
5. `review-log.csv` with columns: `log_ref`, `source_ref`, `date_posted`, `time_posted`, `rating`, `category_assigned`, `assigned_to`, `date_answered`, `hours_elapsed`, `target_hours`, `target_met`, `reply_verbatim`, `routed_to`, `internal_record_opened_date`, `route_closed_date`. `hours_elapsed` is counted from `date_posted` and `time_posted`, never from the date the review was noticed. `target_met` is `yes`, `no` or `not applicable`. Rows in a routed category have an empty `reply_verbatim` unless a reply was supplied.
6. `reply-rules.md` and `review-policy.md`. The first holds the reply rules exactly as supplied, unaltered, under two headings, what every reply does and what no reply ever does. The second holds the published review policy text exactly as supplied, with the destination page recorded at the top as a comment line. Either is written empty if nothing was supplied for it.
7. `gaps.md`, a numbered list of: every source with no named account holder or no notification route; every source with an unknown verification state; every rota cell holding a job title instead of a person; every source and day with no cover or no deputy; every category with no example, no target or no named route; any of the three routed categories written as a reply or without an internal record; every backlog review with no assigned category or no assignee; every routed review with no internal record date or no close date; every logged reply naming a member of staff or containing any of the words `refund`, `voucher`, `gift card`, `discount`, `free` within thirty characters of `review`; and an empty `reply-rules.md` or `review-policy.md`.

## Rules
- Codex builds, counts and records. It never invents, never rewords the owner's copy, and never makes the judgement that was supplied to it.
- Never write a reply, a holding message, a category definition, a target time or a line of policy. Every one of those arrives written. Where one is missing, the cell or the file is left empty and the fact goes in `gaps.md`.
- Never assign a review to a category. Copy the category supplied, and where none was supplied leave it empty and list the review in `gaps.md`.
- Never write the words refund, voucher, gift card, discount or free into any reply, message or policy file, and flag any supplied reply where one of those words appears near the word review.
- Never write a member of staff's name into `review-log.csv` or `review-policy.md`. Names belong in `rota.csv` and `categories.csv` only.
- Never record a target as met from an estimate. `hours_elapsed` is computed from the posting timestamp, and where no time is supplied the row records the date only and is listed in `gaps.md`.
- Never send, publish, post or delete anything. This job writes files.
- Never write an industry average, a benchmark response time, a conversion rate or a figure for bookings, calls or ratings produced by replying. No such figure exists for a single independent venue.
- Counts, elapsed hours and coverage totals are computed, not estimated.
- Use British English, GBP and DD Month YYYY dates, with times as 24 hour `HH:MM`. No em dash characters in any file you write, and any supplied text containing one is recorded verbatim and flagged in `gaps.md`.
- Every file ends with this line: this is a working document prepared for the owner to check against their own staffing, food safety management system and employment procedures before it is relied on. It records a supplied process and is not legal advice or a finding that the business complies with anything.

## Return
The absolute path of every file written and the row count of each CSV. The number of sources by type, how many are verified, and how many have no reply function. The coverage figure as covered days over total source days, with every uncovered source and day named. Every rota cell holding a job title rather than a person. The six categories with their target hours and named routes, and any routed category wrongly written as a reply. For the backlog: the count by category, the number answered, the number within target and the longest elapsed time in hours. Every routed review with no internal record date or no close date. Every flagged reply. Whether `reply-rules.md` and `review-policy.md` carry text and the page the policy is to be linked from. And the `gaps.md` item count.

Built from the best public work on this

Sources for reputation-rota

Everything below was opened and read on 16 September 2026. Nothing is cited that could not be loaded.

1. Competition and Markets Authority, "Fake reviews", CMA208

https://assets.publishing.service.gov.uk/media/67eeb64fe9c76fa33048c790/CMA208_-_Fake_reviews_guidance.pdf, published 4 April 2025, read 16 September 2026.

The regulator's guidance on paragraph 13 of Schedule 20 to the Digital Markets, Competition and Consumers Act 2024, and the reason this skill produces a written policy rather than only a staff rota.

Chapter 8 is addressed to publishers, a category the guidance defines broadly at 8.3 to include "traders who display or make available by any means consumer reviews or consumer review information on their own media (retailer websites, print publications etc.)". A venue showing quotes on its own site is in that group. Paragraph 8.5 says what all publishers need: "(a) to have a clear policy on the prevention and removal of banned reviews and false or misleading consumer review information, and in addition (b) assess the risks of such material appearing on their media and take such further proactive steps as are reasonable and proportionate to address the issues identified." Paragraph 8.9 repeats it in plainer terms, "Publishers should all have published policies which prohibit fake reviews. Such policies should also set out the publisher's approach to incentivised reviews and consumer review information", and 8.10 says where it lives: "easily accessible, with signposting from relevant parts of the medium used (for example, a website) and not tucked away in a hard-to-find place. They should also be written in plain English." Paragraph 8.6 supplies the monthly read in step 9, requiring publishers to "regularly evaluate the effectiveness of these steps" and to address inadequacies they find.

Paragraph 8.11 is the most useful passage in the document for a hospitality manager, because it names three well-intentioned habits as failures. Publishers should not "(a) try to persuade consumers to submit a complaint, rather than leave a review for publication, (b) try to dissuade consumers from leaving a review of their experience even if their initial problem has been resolved through the complaints process, (c) treat a negative review intended for publication as a complaint and not publish it." Every one of those is something a good manager does instinctively while trying to put a customer right, which is why they appear as rules rather than as advice.

Paragraph 4.4 supplies the other hard rule. Traders should not interfere with a reviewer's willingness to leave a negative review "through threats of harm or legal action", "by arbitrarily stopping and starting review invitations", or "by making an offer of dispute resolution contingent on a consumer not leaving a negative review". And chapter 3's commissioning examples supply the one the skill refuses outright: "Contacting a customer who has left a negative review and offering them a refund and/or a gift card if they change their review to remove the negative commentary (so that it is no longer reflective of their genuine experience)."

Where the skill departs: chapters 6 and 7 target brokers selling fake reviews and platforms hosting them, and chapter 8's detection, investigation and sanctions machinery is written for businesses with moderation teams. A venue with one manager does not run a detection pipeline. The skill takes the published policy, the three prohibitions and the duty to review the process, and leaves the rest where it belongs. The guidance is also explicitly not the law itself, and the skill says so rather than presenting it as the statute.

2. Digital Markets, Competition and Consumers Act 2024, Schedule 20, paragraph 13

https://www.legislation.gov.uk/ukpga/2024/13/schedule/20, read 16 September 2026.

The statute behind CMA208, cited so the rota's hard rules rest on the law rather than on guidance about the law. Paragraph 13(1) bans submitting or commissioning "(a) a fake consumer review, or (b) a consumer review that conceals the fact it has been incentivised", which is what a refund offered for a changed review amounts to once the changed review no longer reflects the writer's experience. Paragraph 13(3) is the one that makes a written process necessary rather than optional: publishing reviews "without taking such reasonable and proportionate steps as are necessary for the purposes of - (a) preventing the publication of - (i) fake consumer reviews, (ii) consumer reviews that conceal the fact they have been incentivised, or (iii) consumer review information that is false or misleading, and (b) removing any such reviews or information from publication."

It is a banned practice, which means it is automatically unfair and no one has to show a particular customer was misled.

Where the skill departs: the Act's enforcement regime and its penalties are not described. They are the CMA's to apply and a solicitor's to explain, and an owner frightened by a number acts worse rather than better.

3. Food Standards Agency, "Allergen guidance for food businesses", GOV.UK

https://www.gov.uk/government/publications/allergen-guidance-for-food-businesses/allergen-guidance-for-food-businesses, published 2 April 2020, updated 17 July 2026, read 16 September 2026.

Step 5 exists because of this document, and it is the step that makes the difference between a rota and a marketing checklist. A one-star review saying a diner declared an allergy and was served the allergen is not a reputation event. It is a report about a food safety management system, and the obligations it touches are already in force.

The guidance names the fourteen: "celery; cereals containing gluten (such as wheat, rye, barley, and oats); crustaceans (such as prawns, crabs and lobsters); eggs; fish; lupin; milk; molluscs (such as mussels and oysters); mustard; peanuts; sesame; soybeans; sulphur dioxide and sulphites; tree nuts". For a restaurant, cafe, pub or takeaway serving food that is not prepacked, the duty is stated without qualification: "If you provide non-prepacked foods, you must supply allergen information for every item that contains any of the 14 allergens", delivered either as "full written allergen information on a menu, chalkboard or in an information pack" or "verbally, with a written notice placed in a clearly visible position". Two more lines drive the routing. On people: "Food businesses must make sure that staff receive training on allergens." On records: "You need to make sure that you know what is in the food you provide. You can do this by recording allergen ingredient information in a written format."

Those are what a manager checks on the day such a review appears, and they are checkable in an hour: the dish specification, the allergen matrix, who was on, what they were trained on and what was said at the table. The public reply comes after, which is the ordering the rota enforces.

Where the skill departs: the guidance does not say what a business must do when a customer reports a reaction, and we did not find a loadable page that does, so the skill does not invent a procedure or a reporting deadline. It routes the review to the person responsible for food safety, opens the record, and leaves the decision on notifying the local authority to that person and to the venue's own food safety management system. The skill also does not reproduce the FSA's labelling rules for prepacked for direct sale food, which are a separate regime and not what a review is usually about.

4. Google, "Manage customer reviews", Google Business Profile Help

https://support.google.com/business/answer/3474050, no publication date shown on the page, read 16 September 2026.

Cited for three operational facts. The precondition in step 2: "Before you can reply to reviews, you must verify your business. After your business is verified, you can reply to reviews on your Business Profile." The visibility in step 7: "If your reply is approved, it will be publicly posted under the customer's review. It will appear like your business replied, and your personal name won't be shown." And the moderation, which is why the reply rules exist before the replies do: Google reviews replies to make sure "that they follow Google's content policies", and "If your reply isn't approved to be posted, you'll be asked to edit it." The page also confirms the route for a review that breaches policy, which is to flag it rather than to argue with it.

Where the skill departs: the mechanics of flagging, appealing and chasing a removal are a separate job with its own evidence requirements, and the rota's category for a platform policy breach routes to that rather than teaching it here.

5. Google, "Prohibited & restricted content", Maps User Generated Content Policy Help

https://support.google.com/contributionpolicy/answer/7400114, no publication date shown on the page, read 16 September 2026.

Two prohibitions sit behind the rota's hardest rules. On personal information, which is why step 6 forbids a public reply that comments on an identified individual: "Do not distribute or post personal information without consent", with the prohibited content including a person's "full name, or last name, their face in a photograph or a video, or other information which has been reported as having been posted without consent". A reply that confirms a named waiter was on shift, or that names the reviewer's booking, is on the wrong side of that.

On incentives, addressed to the merchant directly, merchants must not "Offer incentives - such as payment, discounts, free goods and/or services - in exchange for posting any review or revision or removal of a negative review", which is the platform's version of the CMA's refund-for-removal example and lands in the same rule. The policy also bans content "based on a conflict of interest", defined to include "current or former employment", which is why a rota that lets staff review their own venue is a rota with a problem in it.

Where the skill departs: the policy's harm categories, from hate speech to restricted goods, are not taught. A venue meets them as a reader rather than a writer, and the rota's response to any of them is the same, which is to flag the content rather than reply to it.

6. Google, "Tips to improve your local ranking on Google", Google Business Profile Help

https://support.google.com/business/answer/7091, no publication date shown on the page, read 16 September 2026.

Cited for one sentence and deliberately for no more: "When you reply to customer reviews, it shows that you value their feedback." That is the outer limit of what any published source says about the effect of replying, and it is the only claim the skill makes. Owners are routinely told that replying within an hour lifts a rating by a stated amount; no publisher of that figure has measured a single independent venue, and a target set from it will be missed every week until the rota is abandoned, which is worse than having no rota at all.

Where the skill departs: the page also says "More reviews and positive ratings can help your business's local ranking." The skill does not build on that, because a rota is about answering what arrives and asking for reviews is a different job with its own rules.

Best public prompt we found for this job

The closest public artefact is the `customer-escalation` skill in Anthropic's `knowledge-work-plugins` repository, raw source at https://raw.githubusercontent.com/anthropics/knowledge-work-plugins/main/customer-support/skills/customer-escalation/SKILL.md. The repository has 24,123 stars, read from api.github.com on 16 September 2026. Its central idea is the one this skill borrows, that some incoming messages stop being replies and become something else, and that the boundary has to be written down in advance:

### Handle in Support When:
- The issue has a documented solution or known workaround

with a matching list headed "Escalate When". Deciding that boundary at eleven o'clock on a Saturday is how a serious allegation gets a cheerful two-line reply under it, so writing the categories and their routes first is the right instinct.

Three things we did not copy. Its escalation triggers are commercial, built on revenue at risk, contract value and a breached service agreement; ours are the nature of the allegation, because a hospitality venue escalates an allergic reaction from a customer who spent eleven pounds and does not escalate a slow service complaint from one who spent eight hundred. It escalates to teams, "L2 Support, Engineering, Product, Security, or Leadership", which does not survive contact with a business of nine people, so every route here ends at a named person with a named deputy. And its whole apparatus is internal; nothing in it is published, whereas half of this job is a reply the world reads and a policy the regulator expects to find on the website, which is why the rota carries a published document and a log that proves it was followed.

Want this running in your business?

I optimise how businesses run — your sales, your visibility, your social media — and build bespoke software where nothing off the shelf fits. The first conversation is free. Work starts from £150 a day.