How do agents check my live site, and what does that leave on it?

With tools that run on the SyncNow server, never on your PC.

verify_domain proves that the domain is served by the project's hosting folder, by writing a one-time file there, reading it back over the web and removing it; the proof is remembered for 30 days. Since version 13.2.49.84 it writes nothing on the site when a valid proof is already there, and when it does write, its answer says what it wrote and whether it was really removed.

Then fetch_url requests pages of that domain, with no cookie, password or login. When a page answers 401 or 403, it says who refused it, the page or the web server; SyncNow remembers an exact refused address for 7 days, and a whole folder only after two of its addresses were refused by the web server. It follows redirects only inside the proven domain, up to five, waits from 5 to 25 seconds as the agent chooses, and after that reports the outcome as unknown, because the request may have reached the site. It can send the user agent of an AI crawler such as GPTBot or ClaudeBot as a simulation: the site still sees SyncNow's server, not the crawler. Since version 13.2.49.89 it can check several pages of the domain in one request and ask the site to compress. Since version 13.2.49.91 those pages are fetched over one connection to the site, a page is not started when less than 8 seconds of the request are left (it is handed back to be asked again), and each project and person can make at most 40 such requests a minute, so a loop cannot flood the site from SyncNow's single address. When the hosting marks a site as production (a header such as x-aruba-prod), SyncNow remembers it and tells every agent of the project before it acts: agents are told never to send test payloads there, and a request that sends data to that site is refused unless the agent gives a reason, which is recorded.

When a page fails, read_error_log reads the error log inside the hosting folder, first the one the site declares in its .user.ini, php.ini or .htaccess, and filters it by time, words and level. Since version 13.2.49.84 email addresses, IP addresses and phone numbers are masked in the lines it returns, and secret values are hidden as in every read; a log outside the project's folder cannot be reached. How changes reach the site is explained in How do changes reach my live site, and what does a deploy do?.


How do changes reach my live site, and what does a deploy do?
Can my agent clone the project with git and publish changes?
Getting Started
Projects and the Workspace
Agents and Work Sessions
Planning
Rules
Activity, History and Insights
Managers and Teamwork
Connections, Credentials and Security
Source, Safe Editing and Recovery
Updates, Reports and Troubleshooting