Chat Assistant Disclosure Desk: agent fixes for your website chat, AI said up front, a person on request, owner approved
by PaperChaseWebb
Let an agent tidy your website chat assistant's greeting, instructions, handoff text and AI use page, and review a week of saved chats, without the assistant passing as a person, keeping visitors from a person, or making up prices and promises. It writes fixed copies and never touches the live chat. A fix check refuses a removed AI line, a deleted emergency line, a new discount or a new phone number. A chat check flags every reply that claimed to be human, ignored a request for a person or quoted a price you never wrote down. You approve at your terminal and paste it in yourself.
- Price
- $29
- Difficulty
- Intermediate
- Setup
- 45 min
- Format
- .md file
chat-assistant-disclosure-desk-ai-said-up-front-person-on-request-owner-approved.md · 186 KB · instant download after checkout
Want this built for you instead? We can build it for you.
What you end up with
Your chat assistant says it is AI in its first message, answers honestly when asked, hands off to a person with your real contact details, and quotes only prices and policies you wrote down, one approved change at a time, with nothing deleted and no claim that anything is compliant.
What you get
- 21 sections and an appendix in one .md file, with a one-line unpack for all 80 files
- chat_check.py config: 19 checks on the greeting, instructions and handoff text, from no AI line and "never reveal that you are an AI" to discounts, card number requests and a handoff with no contact
- chat_check.py page: 12 checks on your AI use page, including what a person reviews, how long chats are kept and how to reach someone
- chat_check.py chats: 16 checks on every saved chat, such as no disclosure, "yes, I'm a real person", a person request ignored, a price or warranty facts.md does not make, a booking that never happened and a card number asked for
- chat_check.py fix: 12 checks that compare the agent's copy with the original, so a deleted sentence, a removed AI line, a new discount, phone or promise fails
- A claims check that stops reports saying the chat is compliant or that the agent updated the live chat
- The owner step: approve.py needs a real terminal and part of the file hash, refuses agent sessions and piped input, and checks the original has not changed before you mark it published
- A 10-chat eval set for a made-up furniture store, rules and two skills for Grok Bot, Claude Code, Codex and Hermes, six prompts and four templates
- The test harness: 52 unit tests and 447 planted faults you can repeat
Who it is for
Small businesses and agencies with an AI chat assistant on a website, who want an agent to tidy it and review its chats, with a person approving every change.
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, logs in to nothing and changes nothing in your chat tool.
Time to result: The worked example in 5 minutes; your own assistant and a week of chats checked in about 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: 80 files written; 52 tests OK; 447 of 447 planted faults caught.
Watch the flow
Run sh example/run_example.sh and read it next to section 17.
Expected: A plumber's assistant files with 19 findings, a bad saved chat with 10, the agent's first-try fixes caught with 23, the second tries clean, a first-try report with 5 findings and a piped approval refused.
Put in your assistant
Fill in site/settings.json and site/facts.md from the templates, paste the greeting, instructions and handoff text into site/assistant/, save a week of chats into site/chats/, then run python3 desk/chat_check.py site and python3 desk/chat_check.py config site/assistant.
Expected: A short list of settings and assistant lines to fix first.
Fix and review
Have your agent follow skills/assistant-fix and skills/chat-review, then run chat_check.py fix queue/fixes and approve each file at your terminal.
Expected: Fixed files in queue/fixes with 0 high findings, and a list of visitors to call back.
Sample prompts, free to use
Follow skills/assistant-fix for every file in site/assistant and site/pages/ai-use.md. Do not change the live chat. List the codes fixed and the codes left for me, with the reason for each.
Prompt 1 of 6.
Follow skills/chat-review for site/chats. Do not edit any chat. List the visitors I should call back and why, with no personal details.
Prompt 2 of 6.
Run chat_check.py chats site/chats. For each T06, T07 and T08, tell me the price or policy the visitor asked about so I can add it to facts.md or tell the assistant to hand off.
Prompt 4 of 6.
Free preview of the file
# Chat Assistant Disclosure Desk: agent fixes for your website chat, AI said up front, a person on request, owner approved Version 1.0, 2026-11-01. By PaperChaseWebb. For small businesses and the agencies that run their websites, who use an AI chat assistant and want an AI agent to tidy its greeting, instructions, handoff text and AI use page, and to review saved chats, without the assistant passing as a person, keeping visitors from a person, or making up prices and promises. Works with Grok Bot, Claude Code, Codex and Hermes. Python 3.8 or newer, standard library only, no API keys, no logins. The kit and the agent never change your live chat or website. Tested with 52 unit tests and 447 planted faults on Linux with Python 3.13, 3.12, 3.11, 3.9 and 3.8. All business, contact and chat data in this kit is made up. Not legal advice. The rules are getting plain. Maine has required since September 2025 that a business chatbot tell consumers clearly they are not talking with a human. California has required bots that sell to say so since 2019, and from January 2027 its largest businesses must get a person to a customer who asks within 15 minutes. Utah requires an honest answer when someone asks whether they are talking to AI. In the EU, people must be told they are talking to an AI system, since August 2026. And in Moffatt v. Air Canada a tribunal held the airline to a refund policy its chatbot made up. Most small business chat assistants are close. The widget says "Hi, I'm Jes
Test record
- Date
- Nov 1, 2026
- What was run
- tests/test_lib.py (8), test_config.py (9), test_chats.py (11), test_fix.py (6), test_modes.py (7), test_mini.py (3) and test_approve.py (8, on a real pseudo terminal); tests/mutate.py with 426 planted faults across assistant files, saved chats, bad fixes, reports, settings, agent rules and the eval set, plus 21 against chat_mini.py; chat_check.py set on the 10-chat 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
- 52 passed, 0 failed on all five versions; 447 of 447 planted faults caught and all clean baselines clean; the eval set matched its 38 expected codes and every fixed file had 0 high findings; the example caught the first-try fixes with 23 findings and the first-try report with 5, and refused piped approval; the unpack line wrote every file and test_all.sh passed there.
- Not run
- A live Grok Bot, Claude Code, Codex or Hermes session. Chat tool settings and live widgets. Voice and phone bots. Languages other than English. macOS or Windows. Legal review.
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 fixes were written by hand to the rules. Give your agent the entrance exam in section 15 before you rely on it.
- Not legal advice. Nothing in the kit says a chat, a page or a business meets any law. Section 20 lists the public sources it was built from.
- The checker reads the text you copy out of your chat tool. It does not read the tool's settings or see what the live chat says today, so ask the live chat yourself after each change. Voice and phone bots are not covered. English only.
- 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.
- All businesses, chats, names and numbers in the kit are made up. Tested on Linux only; macOS and Windows were not tested.
- On 2026-11-01 on Linux: 52 unit tests and 447 of 447 planted faults passed on Python 3.13, 3.12, 3.11, 3.9 and 3.8; the 10-chat eval set matched its 38 expected codes; the unpack line wrote every file 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 change my live chat or website?
No. The agent writes fixed copies into a queue folder. You approve each one at a terminal and paste it into your chat tool yourself.
Which chat tools does it work with?
Any tool where you can copy the greeting, instructions and handoff message and export chats as text. The kit never logs in to the tool.
Does my assistant have to stop using a name?
No. It can keep a name as long as it says it is an AI assistant, for example "Jess, Copperline's AI assistant".
Can you set it up for us?
Yes. Our Install sprint sets up your settings, facts file and eval set, or Managed Ops runs the fixes and chat reviews for you each week. 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. Do not resell or republish the kit itself. Download terms
Related
Chat Assistant Disclosure Desk: safe chat assistant rules and a ten-check chat scan (free sample)
From Chat Assistant Disclosure Desk: the ways website chat assistants slip, ten rules to paste into your agent, and a read-only script that runs ten checks on one saved chat, from no AI line in the first message and "yes, I'm a real person" to a request for a person with no contact given, made-up waivers, card number requests and bookings that never happened.
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.
Text Message Consent Desk: agent fixes for texts and send lists, opt-ins checked, every STOP honored, owner approved
Let an agent tidy your marketing and reminder texts and build each send list, without texting anyone who said stop, anyone with no record of signing up, or anyone outside quiet hours in their own time zone. It writes fixed copies and lists and never sends or edits your records. A fix check refuses deleted prices, moved send times and invented offers. A list check tests every number against your consent records, replies and send history. You approve at your terminal and send it yourself.
Review Reply Desk: AI-drafted review replies, checked for privacy and offers, approved by the owner
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.
Consent and Privacy Fix Desk: agent fixes for cookie banners, trackers and privacy policies, checked against the original, owner approved
Let an agent make your analytics and ad pixels wait for a real choice, fix the cookie banner and bring the privacy policy in line with what your pages load. It writes fixed copies, never the live page. A fix check refuses deleted trackers, edited tracker code, new third-party scripts, removed words and a Reject button smaller than Accept. A policy check catches tools the policy leaves out and retention periods the agent made up. A tested consent.js honors Global Privacy Control. You approve at your terminal and put it live yourself.
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.
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.