AnQiCMS Help Document and User Guide
The loading time of cross-domain images is read as zero in the front-end monitoring, the missing is the timing pass response header. The official document defines: this header specifies which sources are allowed to see the attribute values obtained through the resource timing interface, otherwise these values will be reported as zero due to cross-domain restrictions. The value can be a wildcard to release all sources, or a comma-separated list of sources. It only affects timing readings and does not change the access permissions of the resources, so adding this header solves the attribution problem, not the access problem.
The 103 intermediate response used for pre-loading resources on the first screen is a prompt issued by the server before it organizes the final page, allowing the client to pre-connect or pre-load resources. The official documentation suggests that for compatibility and security reasons, it should only be sent in HTTP protocol version 2 or higher, as most browsers only support it under these conditions. The reason is that under old protocols, the boundary handling between intermediate responses and final responses is inconsistent, which is easy to confuse with the content of waiting responses.
The security response header is added but not visible in some responses, usually stuck at two places: first, the header addition instruction only takes effect for specific response codes, the effective codes listed in the mainstream Web server documentation include two hundred successful segments and three hundred redirects, non-successful responses need to be added unconditionally with the always parameter; second, the hierarchical inheritance rule - once the header addition instruction is defined at the same level, the entire set of definitions at the upper level no longer inherits. The troubleshooting sequence is to first reproduce the response code missing the header, and then locate which level of configuration generates it.
On September 15, 2026, an edge platform announcement divided search, AI training, and AI agents into three independent controls, so the site does not have to choose between retaining search and blocking training. The announcement also provided two ratios: the mixed-purpose crawler accounted for 36.6% of the verified crawling traffic, less than 1% of the site side would block search crawlers, and about 17% restricted AI training. After the switches are opened, the access rules should be judged according to the purpose of access, not according to the source domain.
A measurement covering thirty-six institutional sites, forty days of traffic, and about one hundred and thirty self-reported identity scraping scripts shows that the compliance rate varies greatly by instruction type: about 60.9% for scraping frequency instructions, about 31% for precise path prohibition, and about 30.7% for whole section prohibition; and less than four in ten scripts will re-read the exclusion file within one hundred and sixty-eight hours. These numbers indicate that relying solely on the exclusion file is not sufficient, and there must also be anti-crawling and permission layers within the site.
The original agreement divides the failure to read into two opposite results: when the response status code falls within the five hundred to five hundred and ninety-nine segment of server errors, the crawling party must consider it as completely prohibited; when it falls within the four hundred to four hundred and ninety-nine segment of client errors, the crawling party can access any resource on the server. The agreement also states that these rules are not access authorization and cannot replace content security measures; the background of the station and the user group permissions are the real door.
This site description file, according to the current proposal, is to help agents use the website, while the early expression is to help the language model understand the web page, and the service object has expanded from the model to the agent. The file is a plain text list, and only one item is required in the specifications: a first-level title with a project or site name, the rest of the abstract sections, descriptive sections, and file lists grouped by secondary titles are optional. AnQiCMS supports automatic generation of site llms.txt and is different from the site map.
A generative engine optimization study decomposes the content structure into three layers: document architecture, information blocks, and visual emphasis, and conducts experiments on six mainstream generative engines. The report shows a 17.3% increase in citation rate and a 18.5% increase in subjective quality, without changing the main theme content. This article explains what these two percentages refer to, why structural improvement does not equal content quality improvement, and its division of labor with automatic site generation such as llms.txt.
Most web servers default to writing the program version into the Server response header and error pages, the Nginx server_tokens directive is set to on by default, after it is turned off, the two places no longer contain version information. Removing this fingerprint can reduce the convenience of version collision, but it only reduces information exposure and does not constitute a defense. This article discusses where the version number leaks out and what remains after it is turned off, as well as where to make changes at each layer for different deployment methods such as Baota and Docker.
When your own image is directly referenced by other websites, it is the resource party that decides whether to allow it, not the referencing party. The Cross-Origin-Resource-Policy response header is set by the referenced resource, indicating which sources can load it, with values divided into three levels: same-origin, same-site, and cross-origin. The default is to block cross-site requests without cors. This article discusses which level each tier allows, whether it is on the same end as the initiating party, and whether images and fonts should be released separately.
