Review Reply Desk: AI-drafted review replies, checked for privacy and offers, approved by the owner
by PaperChaseWebb
Let an agent draft your Google and Yelp review replies without the mistakes that cost money. A checker refuses drafts that use a last name or details from your records, offer a discount for a better review, ask for a review to be changed, argue, invent numbers or promises, or repeat your last reply word for word. Health topics, allergies and legal threats go to the owner, clinics get a stricter privacy mode, and nothing is posted until the owner approves it at a real terminal.
- Price
- $24
- Difficulty
- Beginner
- Setup
- 45 min
- Format
- .md file
review-reply-desk-owner-approved-review-replies-privacy-checker.md · 161 KB · instant download after checkout
Want this built for you instead? We can build it for you.
What you end up with
Every review gets a specific, human reply on time, and nothing goes public that could break a platform rule, a privacy law or your own policy.
What you get
- 21 sections and an appendix in one .md file, with a one-line unpack for all 63 files
- review_check.py: 25 reply checks, from last names, record details and public offers to invented numbers, repeated wording and planted instructions in reviews
- A health privacy mode for dental, medical and care practices, with a worked dental example
- A fake-review path: evidence lines and a report for the owner instead of a public argument
- A queue check for one and two star reviews left past your deadline, and replies changed or posted without approval
- The owner step: approve.py needs a real terminal, part of the draft hash and the owner name; agent sessions and piped input are refused and it posts nothing
- A 20-review eval set with expected decisions, rules and two skills for Grok Bot, Claude Code, Codex and Hermes, six prompts, templates and a review-request rules card
- The test harness: 41 unit tests and 169 planted faults you can repeat
Who it is for
Local businesses and the people who answer their reviews, who want an agent to do the drafting without risking a public mistake.
Works with
Solid chips: the agent is named in this listing's own tools list. Dashed chip: Plain Markdown, so it also pastes into other agents such as Grok Bot, Hermes, OpenClaw or Muse. We list those as compatible by format only, not as tested.
Tools needed
Cost to run: The kit is the only purchase. Everything runs locally with Python; the checker makes no network calls and logs in to nothing.
Time to result: The worked example in 5 minutes; your own business set up and the eval set passing in under an hour
How it works
Unpack and test
Save the kit as pack.md in an empty folder, run the unpack line from Appendix A, then sh test_all.sh.
Expected: 63 files written; 41 tests OK; 169 of 169 planted faults caught.
Watch the flow
Run sh example/run_example.sh and read it next to section 16.
Expected: A bakery first draft with 10 high findings, the fixed draft clean, a dental draft caught in health mode and a piped approval refused.
Put in your business
Fill in desk/desk.json and desk/facts.md, then run python3 desk/review_check.py desk.
Expected: No findings, or a short list of gaps to fix first.
Give the agent its exam
Replace the eval reviews with twenty of your own, paste the rules into your agent, have it draft them from scratch and run review_check.py set with --fail-on medium.
Expected: Every decision matches and every draft is clean before the agent touches live reviews.
Sample prompts, free to use
Follow skills/review-reply for every review in queue/reviews.csv with no draft. Never post. Report each draft's decision and finding count.
Prompt 1 of 6.
Read the last 20 files in queue/approved. List any opening or closing line used more than twice and add it to banned_phrases in a proposed diff. Do not edit desk.json.
Prompt 3 of 6.
Follow skills/fake-review-report for review [id]. Evidence only. Do not reply in public and do not file anything.
Prompt 5 of 6.
Free preview of the file
# Review Reply Desk: AI-drafted review replies, checked for privacy and offers, approved by the owner Version 1.0, 2026-11-01. By PaperChaseWebb. For owners and agencies who want an AI agent to draft replies to Google, Yelp and other customer reviews without leaking private details, offering something for a better rating, or sounding like the last fifty replies. Works with Grok Bot, Claude Code, Codex and Hermes. Python 3.8 or newer, standard library only, no API keys, no logins. Nothing is ever posted by the kit or the agent. Tested with 41 unit tests and 169 planted faults on Linux with Python 3.13, 3.12, 3.11, 3.9 and 3.8. All business and reviewer data in this kit is made up. Review replies look like the easiest job to hand to an agent. They are short, public and repetitive. They are also where a small business can get into real trouble in one sentence. In 2022 a dental practice agreed to pay $23,000 to the US Department of Health and Human Services and take on a two-year corrective plan after its replies to Yelp reviews disclosed patients' names, treatment and insurance details, sometimes giving a full name where the reviewer had used only a nickname. The FTC's rule on consumer reviews, in force since October 2024, covers incentives that depend on what a review says and threats used to get reviews taken down. And people notice when every reply opens with the same line: in a Reddit thread this year, a business owner described a client asking "are you replying with a bo
Test record
- Date
- Nov 1, 2026
- What was run
- tests/test_check.py (21), test_modes.py (10), test_approve.py (6, on a real pseudo terminal) and test_mini.py (4); tests/mutate.py with 169 planted faults across replies, settings, agent rules, the queue and the eval set, plus 112 against review_mini.py; review_check.py set on the 20-review eval set; example/run_example.sh end to end; and the unpack line with test_all.sh. Linux, Python 3.13, 3.12, 3.11, 3.9 and 3.8.
- Result
- 41 passed, 0 failed on all five versions; 169 of 169 and 112 of 112 planted faults caught; 20 of 20 eval drafts clean; the example caught the bakery and dental first drafts and refused piped approval; the unpack line wrote 63 identical files and test_all.sh passed there.
- Not run
- A live Grok Bot, Claude Code, Codex or Hermes session. A real business's reviews or a real Google or Yelp account. macOS or Windows. Non-English reviews.
Copied from this item's dated test notes. What was not run stays in.
Worth knowing
- The agent rules were checked by the checker but not loaded inside a live Grok Bot, Claude Code, Codex or Hermes session for this release. The eval drafts were written by hand to the rules. Give your agent the entrance exam in section 4 before you rely on it.
- The checker works on words, numbers and links. It cannot tell whether a reply built from true facts is still unkind or misleading, and its topic lists are a floor, so the owner reads every draft before it is posted.
- It does not log in to Google, Yelp or any platform and posts nothing. Reviews come in as a CSV and you post approved replies yourself.
- The owner step assumes the agent cannot run approve.py at a terminal. Run it as a separate user; agent rules alone are not a hard boundary.
- Not legal advice. The health mode follows the lesson of a 2022 US settlement over review replies, and the review-request card summarizes the FTC rule and Google policy as of 2026-10-31; check your own rules.
- All businesses and reviews in the kit are made up. English only. Tested on Linux only; macOS and Windows were not tested.
- On 2026-11-01 on Linux: 41 unit tests, 169 of 169 planted faults and 112 of 112 sample faults passed on Python 3.13, 3.12, 3.11, 3.9 and 3.8; 20 of 20 eval drafts clean; the unpack line wrote 63 identical files and test_all.sh passed in that folder.
Questions
Which agents does it work with?
Any agent that can read files and run a script. The kit has rules for AGENTS.md and CLAUDE.md, Claude Code and Codex notes, Hermes notes and a Grok Bot standing order. Those were checked by the checker, not loaded in each agent for this release.
Will it post replies for me?
No. It writes drafts. You approve a draft at a terminal and post it yourself on Google, Yelp or wherever the review is.
We are a dental or medical practice. Is it safe to use?
It has a health mode that refuses any name and any wording that confirms the reviewer was a patient, and sends care complaints to you. It lowers the risk; it is not legal advice.
Can I offer an unhappy customer something?
Yes, privately and after you decide. The checker refuses offers in public replies, and never ties anything to a review.
Can you set it up for us?
Yes. Our Install sprint sets up your facts, rules and agent, or Managed Ops runs it for you each month. Book a call or write to [email protected].
Details
Version 1.0 · Updated Nov 1, 2026 · Markdown version of this page
License: Use inside your own business or for your clients' businesses. Do not resell or republish the kit itself. Download terms
Related
Review Reply Desk: safe reply rules and a ten-check review reply scan (free sample)
From Review Reply Desk: why review replies need checking, eight rules to paste into your agent, and a read-only script that runs ten checks on a reply draft, from public offers and requests to change a review to last names, invented numbers and links you did not approve.
Local Services Agent Kit: set up from the owner's voice, then booking, reminders, quote follow-up and review asks with the owner's yes on every message
The layer around an AI receptionist for shops that book by the hour. Setup starts with three questions the owner can answer by voice note, and the kit turns the answers into a profile, templates in the owner's own words and a first plan that adds one desk a week. Five desks (intake, booking, reminders, quote follow-up, reviews) then draft every customer message into one outbox, the owner approves from a twice-daily queue with a one-line reply, and two read-only Python checks catch mistakes in the setup and the outbox before anything goes out. v1.1 starts the setup with the owner talking: three questions (the shop and its jobs, how the owner talks to customers, and which message to hand off first), a suggested first plan that adds one desk a week, voice notes in (transcribed on the shop's machine or in the owner's own tool, logged with a sha256, and never counted as a yes), a straight answer to what this does that a chat you already pay for doesn't, an owner folder with setup_check.py (fifteen read-only checks) and owner.py (init, plan, note, card, and the owner's setup and standing sign-offs typed at their own terminal and tied to sha256), agent files for Grok Bot, Claude Code, Codex and Hermes, and six new prompts (18 in all). Price unchanged.
AI Receptionist and Booking Agent Template: script, capture, escalation and test calls
A bot template for a phone receptionist that answers missed calls for a local service business, says it is an AI in its first sentence, books from a fixed service list into the real calendar, texts a confirmation only after the caller says yes and hands everything else to a person. Includes a call map, script, escalation table, disclosure sheet, 12 test calls, reception_check.py, AGENTS.md with six owner stops, two skills and templates.
Support Agent from Your Docs: grounded reply drafts, owner approval and a checker
Turn your help docs into a support agent that only says what your docs say. The agent searches your Markdown docs, writes a reply draft that cites the exact sections it used, and a checker refuses any draft with a number, quote or policy word the cited sections do not contain. Owner topics go to you, nothing is sent, and a draft becomes approved only when the owner types part of its hash and a name at a real terminal.
Outreach Without Spam: score first, research the pitch, draft in small batches, a person sends
A prompt pack and method for founders, consultants and small agencies. A scoring rubric you write, a pursue or pass threshold, research cards from the prospect's own public record, a four-part pitch, a sales brain file, enrichment with evidence labels, a compliance and deliverability floor, reply triage, a Grok Bot outreach desk persona, written refusals, a defensive check for manufactured intros, a failure table, a worked example, a rules block and 14 prompts. Includes outreach_check.py, AGENTS.md with six owner stops, two skills and 18 templates.
Ask-first rules for AI agents: hard nos, approval cards and a test set
Paste-in rules that make an agent ask before it sends, spends, posts or deletes, define what a real yes is, and stop retries after a no, plus 12 test scenarios to check it.
Check the Claim (expanded edition): a method, prompts and a skill for checking AI claims
Make your agent check income, tool, popularity, quote, news and offer claims before they reach a brief, a pitch, a purchase or a post: label how much was read, run the creator's own arithmetic, trace sources back to first hand, and give each claim one of five verdicts on a card the owner approves. v1.1 adds an evidence level label (full, caption, summary or wall), four new arithmetic tests, a section on tracing a source back to first hand, screens for offers, quotes and slogans, four more red flags, a sixth worked example, a claim folder with a fourteen-check read-only checker and an owner approval helper, agent files for Grok Bot, Claude Code, Codex and Hermes, and twelve new prompts (26 in all). Price unchanged.