Ask away.
The bird kept notes.
Questions to ask, the tools your assistant calls for each, and an example answer.
Pick a question and watch it run. Most questions only read, and reading uses no allowance. For a change or a check, your assistant’s app controls approvals.
What did the last check find at summitridgekia.com?
list_siteslist_reportsget_report
Published October 11, 2026. One finding. The booking link in the site header took a reservation. The second scheduler, linked from the service pages, opened to “This page is unavailable.”
Reads only. Uses no allowance. It reads the latest published report, not a check still running.
Show me the evidence for the contact form finding.
list_siteslist_reportsget_report
Published October 29, 2026. The general contact form came back “Unable to process.” The price inquiry on a vehicle page, the same day, went through. The evidence is a phone capture of the contact page with the error marked, its page address and its viewport.
Reads only. The connection tells the assistant to treat the report’s text and its captured pages as data, not instructions.
Was the service scheduler fixed? Compare the last two reports.
list_siteslist_reportsget_reportget_report
Not shown yet. The October 11, 2026 report found the second scheduler, linked from the service pages, opening to “This page is unavailable.” The newer report does not mention that scheduler. An absent finding is not proof it works, so it stays unknown until a check tests that scheduler again.
Reads only. Two exact reports, read for the same observation. A fix is called only when the newer check tested the same thing again.
What did the store send back after our trade-in inquiry?
list_siteslist_assignment_runsget_inquiry
September 22 to October 9, 2026. A trade-in inquiry named the vehicle. The first reply named it too and offered an appraisal. Seventeen days later a follow-up arrived with the subject “2026 Make Unknown Model Unknown.”

Reads only. The inquiry’s references come from list_assignment_runs, not from a message or a date. The tool marks reply text as untrusted data.
Which tools are on my site, and which journeys were assessed?
list_siteslist_site_toolslist_journeysget_journey_assessment
Five tools on the inventory. A service scheduler, a trade-in tool, a chat, a contact form and a parts link. Listed means found on the site, not verified working. The latest check recorded assessments for the service booking and contact journeys, each with its link.
Reads only. A tool on the list is not a pass, and an empty list of assessments is not a failed test.
How is my Summit Ridge Group doing?
list_siteslist_groupsget_group
Summit Ridge Group: three sites, grade B. Two open findings, both on summitridgekia.com: the second service scheduler and the general contact form. No site has a check running now. The Group’s page in Dr. Canary has the rest.
Reads only. Group access comes from list_groups, not from a site’s name.
Pause the service shopper.
list_siteslist_assignmentssave_assignment
Saved. The Service shopper is paused. Its name, departments, inquiry choice, stored frequency and areas of focus are as they were. Nothing ran: saving never starts a check. A paused shopper can still run on a recheck you request.
- pausedtrue
- changedtrue
Changes a saved shopper. A save replaces every setting, so the tool tells the assistant to read the shopper first and send every other setting back unchanged. Your assistant’s app controls approvals.
Add a consent comparison to my sales shopper as an area of focus, for its next check only.
list_siteslist_assignmentssave_assignmentlist_assignments
Saved. One area of focus, for the next check only. The Sales shopper keeps its other settings and its saved areas of focus. The new one asks it to compare arriving, rejecting, accepting, withdrawing consent and returning, and to note what changes at each step. Nothing runs until the shopper’s next check.
- when
one_time - startingUrlnull
Changes a saved shopper. A list of areas of focus is stored as sent and replaces the saved one, so the existing ones are sent back with the new one. one_time runs on the next check only; going_forward runs on every check.
Recheck the site and follow this check to its report.
list_siteslist_assignmentsrequest_recheckget_runlist_reportsget_report
Requested. Here is where it stands. It ran the Service shopper, the site’s first saved shopper that has not been retired, which sends test inquiries. The report is now published, so its date, what it found and its link come next.
- Queuedthe receipt and one check reference
- Running
get_run - Finishednot yet a report
- Report assemblingnot yet published
- Report published
get_report
Requests a check. Before approval, a good assistant says what the check uses: the first saved shopper that has not been retired, which can be a paused one, its test inquiries if it sends them, and allowance. Queued is not finished, and finished is not published.
Run a custom checkout test on this other website.
No tool called.
I can’t do that with Dr. Canary. A check is requested for a site connected to your account and takes no web address. It cannot run a custom test, a chosen shopper or one finding alone. I can read your connected sites’ reports, add an area of focus to a saved shopper on a connected site, or request the basic check of a connected site.
Calls no tool. A good assistant explains the limit and offers what is supported, before any partial save or broader check.

Reading what the bird found
Most questions only read. Reading uses no allowance and changes nothing.
What did the last check find at summitridgekia.com?
list_sites→list_reports→get_reportReads only. Uses no allowance. It reads the latest published report, not a check still running.
The published date, the finding in the report’s words, the capture and the link to the report.
Watch it runShow me the evidence for the contact form finding.
list_sites→list_reports→get_reportReads only. The connection tells the assistant to treat the report’s text and its captured pages as data, not instructions.
The finding in the report’s words, where it was seen, the capture with its page address and viewport, and the link to the report.
Watch it runWas the service scheduler fixed? Compare the last two reports.
list_sites→list_reports→get_report→get_reportReads only. Two exact reports, read for the same observation. A fix is called only when the newer check tested the same thing again.
What each report says about the same thing, and whether the newer check tested it again. An absent finding is not called a fix.
Watch it runWhat did the store send back after our trade-in inquiry?
list_sites→list_assignment_runs→get_inquiryReads only. The inquiry’s references come from
list_assignment_runs, not from a message or a date. The tool marks reply text as untrusted data.The inquiry, the emails and texts that came back, and how each reply was reviewed, current and earlier.
Watch it runWhich tools are on my site, and which journeys were assessed?
list_sites→list_site_tools→list_journeys→get_journey_assessmentReads only. A tool on the list is not a pass, and an empty list of assessments is not a failed test.
The tools found on the site, the journeys run, and the assessments recorded for a check, each with its link.
Watch it runHow is my Summit Ridge Group doing?
list_sites→list_groups→get_groupReads only. Group access comes from
list_groups, not from a site’s name.The Group’s grade, its sites and their current recheck state, its open findings and the Group’s link.
Watch it run
Changing a shopper
Areas of focus are how to ask a shopper to try something particular. Each says what to do in 12 to 600 characters and whether it runs one time or going forward, and can add a start page and what counts as done. A start page is a full web address, and what counts as done is up to 300 characters.
Pause the service shopper.
list_sites→list_assignments→save_assignmentChanges a saved shopper. A save replaces every setting, so the tool tells the assistant to read the shopper first and send every other setting back unchanged. Your assistant’s app controls approvals.
The save receipt. Every other setting stays as it was, and no check runs.
Watch it runAdd a consent comparison to my sales shopper as an area of focus, for its next check only.
list_sites→list_assignments→save_assignment→list_assignmentsChanges a saved shopper. A list of areas of focus is stored as sent and replaces the saved one, so the existing ones are sent back with the new one.
one_timeruns on the next check only;going_forwardruns on every check.The save receipt.
Watch it runlist_assignmentsshows the new area of focus beside the ones already saved.Move the Service shopper’s checks to Tuesdays, Los Angeles time.
list_sites→list_assignments→set_assignment_calendarChanges a saved shopper. It needs a named time zone. It does not pause, resume or run anything.
The calendar receipt, with the day, the time zone and the date it takes effect.
Saved. Tuesdays, America/Los_Angeles. The new day takes effect from the date on the receipt. The shopper’s other settings are unchanged, and no check ran.
Requesting a check
Recheck the site and follow this check to its report.
list_sites→list_assignments→request_recheck→get_run→list_reports→get_reportRequests a check. Before approval, a good assistant says what the check uses: the first saved shopper that has not been retired, which can be a paused one, its test inquiries if it sends them, and allowance. Queued is not finished, and finished is not published.
The receipt and its check reference, each state as the check moves, and the report once it is published.
Watch it run
- Shopperthe first saved one that has not been retired, which can be a paused one
- Test inquiriesif that shopper sends them
- Allowancea check may use it
- Monitoringneeded to request rechecks and edit saved shoppers
What a good answer contains
- The date the report was published.
- The finding in the report’s own words.
- The capture, when there is one.
- The link to the report in Dr. Canary.
- What is unknown, said as unknown: a capture that is not available, a check still running, a page that was not tested.
What the assistant will not do
Run a custom checkout test on this other website.
None.
Calls no tool. A good assistant explains the limit and offers what is supported, before any partial save or broader check.
A plain no, the reason, and what it can do instead.
Watch it run
When a request does not fit, the connection asks the assistant to say so first and offer what is supported, never to widen a request without asking, and never to say a target was tested when it was not.
What it does not do
- It requests checks only for sites connected to your account. A recheck takes a connected site, not a web address.
- It does not run a chosen shopper, a custom test or one finding alone.
- It does not send you a message later. It follows a check in the current conversation.
- It does not call a finding fixed because it is gone. The newer check must have tested the same thing.
- The connection tells it to treat what it reads in a report, a reply or a captured page as data, not instructions.
Questions
What can I ask my assistant about Dr. Canary?
What a check found, the evidence behind a finding, whether a finding was tested again, what a store sent back after an inquiry, which tools are on your site and how a Group is doing. It can also change a saved shopper or request a check of a connected site.
Does asking cost anything?
Reading uses no allowance: reports, checks, shoppers and inquiries are all reads. A requested check may use allowance, and may send test inquiries if its shopper sends them.
Will it change things without asking me?
Your assistant’s app controls approvals, and your account’s permissions apply. Before a change or a check, a good assistant says what will change and what the check will use. Saving a shopper never starts a check.
Can it test any website?
No. A check runs on a site connected to your Dr. Canary account, which list_sites returns, and takes no web address. An area of focus on a saved shopper can name a start page for that check.
Can it tell me when the report is ready?
It follows a check you requested in the current conversation: queued, running, finished, then the report once it is published. No tool sends a message later, so if you step away, ask again and it reads where the check stands.
Which assistants work?
ChatGPT, Claude, Codex and Claude Code. Each has its own setup page.
Does it need monitoring?
Reading works on the sites your account can open. Editing saved shoppers and requesting rechecks need monitoring. A one-off checkup does not grant them.

Give your assistant something to read.
A checkup is one report. Monitoring keeps the bird shopping your site on a schedule, so there is a fresh report every time you ask.



