Agentimus – AI SEO, llms.txt & MCP for AI Agents

Changelog

1.53.4

• Changed: the Report screen is laid out by what it has to show. AI reads takes the full width, with the crawlers that read you in two columns. Visits from AI, What acted here, Citations and In search share one panel, one line each, so a quiet day no longer leaves tall empty cards; a busy line names its top three. The Readiness score card has left the Report: it reads the same whatever days you pick, and the card beside the screen already shows it.
• Changed: “What acted here” names the latest action and what it came in with, not only how many there were.
• Security: a connection key, or an approved assistant’s access token, was honoured on WordPress addresses outside the REST API — admin-ajax.php, admin-post.php, wp-login.php — whenever the address named the MCP endpoint. No REST request was made there, so the check that holds a key to the MCP endpoint never ran, and the key acted with its owner’s full rights, a read-only key included. Found in an internal review. A key now counts only on a request WordPress can route to its REST API, and a path counts only under the REST prefix. Exploiting it needed the MCP server switched on and a valid key or approved grant in hand.
• Fixed: on a site behind a proxy or load balancer other than Cloudflare, real search and AI crawlers could be recorded as impostors — flagged for review, alerted on, and refused if you block impostors. An address that can only belong to the proxy is now “not checked”, never “fake”. An IPv4 address written in IPv6 form is also read as IPv4, so Cloudflare’s and the crawlers’ address lists match it again.
• Fixed: an announcement queued or cancelled while another one was being sent could be lost, or go out after you cancelled it. Alerts raised during a delivery could be dropped the same way.
• Fixed: unpublishing a post, making it private, trashing it or changing its address left the old page and its .md copy in Cloudflare’s cache until they expired. The old address is purged now.
• Fixed: when another plugin’s error broke your site’s replies, screens showed the browser’s own words, such as “Unexpected end of JSON input”. They now say what happened in plain words. A setting whose save could not be confirmed is no longer switched back as if it had failed; the screen asks you to reload instead.
• Fixed: Google Analytics stopped refreshing whenever Search Console could not be read, with no error shown. Each one refreshes on its own now.
• Fixed: the flood limit on the referral beacon never reset under steady traffic, and dropped real visits from AI answers.
• Fixed: share drafts for X, Facebook and LinkedIn could carry HTML codes such as “&” and ” “.
• Fixed: with the Identity name left blank, llms.txt was headed with the site’s address and discovery.json gave no name. Both use the site’s name now.
• Fixed: six PHP 8.4 deprecation notices, and a warning logged on every check of the addresses your site advertises.

1.53.3

• Fixed: a word and its plural are now always read as the same word. “Perfumes” was read as “perfum” while “perfume” stayed “perfume”, so a page that said “perfume” was listed as not answering a search for “perfumes”. The same was true of “theme” and “themes”, “image” and “images”, and “category” and “categories”. “Themes” was also read as “them”, so a page could count as answering a search it does not.
• Fixed: a word ending in “-ing” or “-ed” now matches the same word ending in “e”, in either direction: “changing” and “change”, “updating” and “update”, “writing” and “write”.
• Fixed: a shortcode the page did not render, such as “[woocommerce_my_account]”, was read as words, so a page holding only that tag could count as answering “my account”. Shortcode tags are no longer read as text.
• Changed: the Visitors screen’s “Everyone Who Visited” card says what its numbers mean. It now explains that “people” is who came and “visits” is how many times they came, shows visits per person under the visits, and says that Google Analytics counts each browser as a person. It also says “new” and “had visited before” instead of “first-timers” and “came back”, and “stayed 10+ seconds or opened another page” instead of “stayed to read”, which is what Google actually measures.
• Fixed: when Google Analytics stopped updating, the Visitors cards kept their “Last 30 days” tag and only said the numbers were “over two days old”. They now show the dates the numbers really cover and when they were last updated.
• Fixed: a page waiting on Google showed “since” a date, which read as the day its work was finished. The date is when the page joined the list, and it now says so.
• Note: your site reads its pages again once after updating.

1.53.2

• Fixed: words written with a hyphen or an underscore, such as “AI-discovery” or “the_excerpt”, now count as the words inside them, and a version such as “5.5.0” or “5.5.x” counts as “5.5”. Pages that answered their search in plain words were listed in Your Content as not answering it. A search still needs its whole word: a page that only says “json” does not answer “json-ld”.
• Changed: “pages are one push from page one” now says what each page still lacks: a paragraph that answers the search, the searched words in its title, or a link from another of your posts. It used to give the same two tips every time, even to pages that had done both.
• Fixed: the same finding counted only the three searches listed for each page, so its searches, impressions and visits were too low. It now counts every search the pages are found for.
• Note: your site reads its pages again once after updating, so pages that were wrongly listed leave Your Content by themselves.

1.53.1

• Fixed: posts written in the classic editor were marked as not answering searches they plainly answer. WordPress adds a classic post’s paragraphs only when it shows the page, and the search check read the post as stored, so it found no text to look in. It now reads the post the way your visitors see it.
• Note: your site reads its pages again once after updating, so classic posts that were wrongly listed in Your Content leave the list by themselves.

1.53.0

• New: traffic for each post and page, in Your Content. Each row shows its bot reads, readers from AI and search clicks, and a new menu orders the list by any of them, so a busy page that needs work comes to the top. Open a row to see who came: a day-by-day chart of bot reads and readers from AI, each bot with its count, when it last read the page and whether it was verified, each AI assistant that sent readers, and the searches that brought people from your search engine. Proven fake bots, refused requests, scripts and command-line tools, and requests from WordPress sites (your own included) are not counted as bot reads; the page says how many were left out. Traffic logged before this update is matched to its pages in the background, and the screen says so until that is done.
• New: the Request Log’s views are now named for what they show — Every request, By visitor and a new By page — with a line under them saying what each one is. By page lists one line per post or page, busiest first, with its bot reads, readers from AI, the bots that read it and its last bot read, over the dates you pick. Click a page to see each of its reads with its time. Bot reads of archives, tag pages and the blog index belong to no single page, so the view says how many there were instead of listing them.
• New: “See every read, with its time” opens the Request Log narrowed to that one page.
• New: WordPress’s own Posts and Pages lists get a sortable “Readers From AI” column, which says under its heading while older traffic is still being matched.
• Changed: people an AI assistant sent to your site are now called “readers from AI” everywhere, instead of “AI visits”. On the Request Log a visit is a bot’s, so the same word used to mean machines on one screen and people on the next. Places that meant the bots now say “requests”.
• New: read-post-traffic gives a connected AI agent the same detail for one post, and read-request-log takes a post to list that page’s reads, and group: "pages" for the By page view.

1.52.1

• Changed: discovery.json now declares wire-format 1.1 of the WP_Discovery Protocol. Version 1.1 adds the optional identity fields Agentimus already sends: what your site is not, its audience, and your projects. Nothing else in the document changes.
• Fixed: discovery.json could fail the protocol’s own schema in two places. A part of your site with no agent card was sent with an empty list where the schema allows only an object, so it is now left out. A tool that takes no input, such as one offered by another plugin, had its empty settings sent as lists too, and they are now sent as empty objects. Tools that check the file against the schema now pass.

1.52.0

• New: a Projects list under Settings → Identity, for software you made, such as a plugin, an app or an open-source library. Each project takes a name, a one-line description, a version, and links to its home page, source code and download. Agentimus publishes it three ways: as Schema.org SoftwareApplication data on your home page, naming you as the author; as a “Projects” section in llms.txt and llms-full.txt; and under identity.projects in discovery.json. AI assistants can then answer “who made this?” and link the right download. Projects are optional: until you add one, nothing changes.
• New: the Request Log can show visits instead of rows. Switch to Visits and each client’s requests are grouped while no more than two minutes pass between them, such as “Orbbot · 5 requests in 40 s” with the pages it read. A burst of pages in seconds is an AI assistant answering a question; one page at a time is a crawl. The log keeps no IP addresses, so two people using the same assistant at the same moment from the same network show as one visit. Your choice of view is remembered.
• Changed: read-request-log takes group: "visits" and then returns visits instead of rows.
• Changed: the Services and Projects cards under Settings → Identity are one line until you add something to them.

1.51.9

• Fixed: the Cloudflare warning “Cloudflare is blocking [company], but your policy says allow” could fire when Cloudflare was only stopping a scanner that used an AI crawler’s name to look for secret files such as “.env”. A blocked request for a file no reader asks for is now counted as a probe and no longer counts toward that warning. The edge numbers still show every blocked request, and the first update re-checks the last three days, so a warning a scanner raised in that time clears the next time the numbers update.

1.51.8

• Fixed: on a site with plain permalinks (addresses like “/?p=123”), the Request Log called every tag and category page “Homepage”, because such a page lives at the site root. It now shows the tag or category name, such as “Tag: PHP”. The same fix names these pages correctly in your search results.

1.51.7

• New: the Request Log now names which page a crawler or AI assistant read. A “Page (HTML)” row shows the page’s title, such as a post’s name, “Home” or “Tag: PHP”, as a link that opens the page. Hover it for the full title and address. Rows recorded before this update did not keep the page, so they still read “Page (HTML)”.
• Changed: only the address WordPress resolved is kept, so a link’s query string, which can carry a token or a campaign tag, is never stored.
• Changed: read-request-log returns path and title for each page row.

1.51.6

• Fixed: a tag or category page in your search results was listed by its address, such as “/tag/navigator-locks”, with no button and advice written for a post. It now carries its name, such as “Tag: navigator.locks”, everywhere a page is listed, and its card links to the tag’s edit screen.
• Fixed: when a tag or category page split a search with a post it lists, the split row said the tag page “Earns the click” and suggested pointing the post at it. A tag page only lists posts. Its row now always offers Set this aside, and when one post is up against tag pages, that post is marked as the answer. Setting the tag page aside gives the search back to the post.
• Changed: read-search-opportunities marks each page in a split search with archive, and names tag and category pages the way the screen does.

1.51.5

• Fixed: the notice that Cloudflare is letting training crawlers through counted every request carrying a training crawler’s name that Cloudflare did not block. That included crawlers reading your robots.txt, which is where your ai-train=no line lives, and requests your server turned away with an error. On a site where Cloudflare already stops the training crawlers, it could report hundreds of requests and ask you to block them, when not one page had been served. It now counts only requests that received a page, and dates the notice from the first day one did. Right after updating, the count can read low for a few days while new hours are collected; it never reads high.
• Fixed: the training notice also counted pages served to a training crawler you had allowed in your review queue, so letting one in on purpose kept the notice up. A crawler you allowed no longer counts, unless you also block it as a training crawler: then the block is what you asked for, and it still counts.
• Changed: read-edge-traffic reports a served count for each crawler: the requests that received a page, not counting robots.txt.
• Fixed: asked about a single day, the Cloudflare warnings read “in the last 1 days” and then repeated the day’s count. They now say “in the last day”, once.

1.51.4

• Fixed: the Request Log could warn that Cloudflare was blocking an AI company your policy allows when it was really stopping impostors using that company’s name. The warning checks the blocked requests against the company’s own published addresses and steps down to a note when they all come from somewhere else, but it needed at least ten fakes from the last day, so as an impersonation campaign faded, the warning came back on its last few requests. Once every request blocked that day is proven fake, the warning now steps down however few there were.
• Fixed: a crawler you block on purpose (on your own block list, or a training crawler while your robots.txt says ai-train=no) could still shape a Cloudflare warning about its company. Its blocks set the day the warning said it started, and its requests were counted in the impostor check. Both now leave it out, as the warning itself always did.
• Changed: Cloudflare warnings and the training notice say how many requests fell in the last day, instead of “some”.
• Fixed: the impostor note wore a “Not enforced” badge, although Cloudflare is stopping those requests. It is now badged “Impostors”, on the Request Log and in Settings, and its Findings row counts “requests proven fake” instead of “confirmed impostors”.

1.51.3

• Fixed: a real search crawler could be flagged as a fake on every visit. The identity check asked for an address name that not every operator publishes, and a crawler run from cloud addresses has none, so its operator’s own published address list was never read. An address the operator’s own list names now counts as genuine, even when the name lookup says no, and a crawler whose operator publishes only a list is checked against that list alone. A list that is out of date, missing or silent never clears a failed check. If a crawler is still flagged in your review queue, re-check it to clear the old flag.
• Fixed: in the review queue, a fake crawler using a real crawler’s exact name borrowed the real one’s traffic. Two failed requests beside hundreds of genuine visits were marked “High volume” and sorted above a much larger impersonation, and ignoring such a row did not stick while the real crawler kept visiting. A row caught faking is now weighed on its own failed requests only: for the badge, for its place in the list, and for when an ignored row comes back.

1.51.2

• Fixed: a page your own site redirects — a retired page folded into the homepage, say — was filed by the Google Index card under “Google chose a different address”, the harshest heading, with a line that blamed Google for an address your site picked. Google’s own wording for it is “Page with redirect”, and it now has its own group, “Sent elsewhere by this site”, whose row names where the page goes. It still counts, the way a page you block on purpose does; nothing is hidden.
• Fixed: asking the Google Index tool about a page the daily check had not reached yet answered with an error instead of “not checked yet” — the tool’s own output rules did not allow an empty row. An assistant asking about a brand-new page now gets the honest answer.
• New: the share card names your X account. X already builds a link preview from the Open Graph tags, image included, whether or not you have connected X to Agentimus. When you have, the card also carries a twitter:site tag with the connected account, so the preview is attributed to you. Nothing is guessed when X is not connected.

1.51.1

• Security: a connection key, or an approved assistant’s access token, was accepted on two neighbouring admin routes and not only at the MCP endpoint it was issued for — the check matched the route by prefix, and the route that creates keys starts with the same letters. In WordPress’s plain ?rest_route= form that seated the key as the administrator who created it, and one request could mint a write key from a read-only one whenever agent writes were on. Reported by Ananda Dhakal of Patchstack. The route is now matched exactly in every form WordPress accepts, and a second check holds every plugin credential to the route WordPress actually dispatched, refusing it anywhere else. If you are not sure who holds a read-only key, create a new one; that retires the old one.

1.51.0

• New: every long list turns the same way now — the Request Log, Agent Access, the content worklist and the Google Index card all carry one numbered pager, with a constant seven slots so the strip never changes width and a button never shifts under your pointer between clicks. The Request Log’s Network column joins the sortable ones, and its refresh button no longer quietly resets you to page one.
• Improved: one language for links across the plugin. A link inside a sentence wears a quiet underline, because colour cannot be the only thing that says “clickable” when a status word can paint the identical colour beside it. A jump to another screen ends in a small drawn chevron; a button that acts in place — Refresh, Purge edge cache, Re-check — answers your pointer with a soft highlight instead of an underline promising a destination it does not have; and every link that leaves your site keeps its small outward arrow. The Search Console and Bing Webmaster links on the Visibility screens also stop wearing a blue no other part of the plugin uses.
• Fixed: a description derived from a post’s opening paragraph was cut wherever the word budget ran out — mid-clause, and at the length cap sometimes mid-word — and that line is served: the meta description, the structured data and the page’s markdown twin all carry it. It now ends at a sentence boundary when whole sentences fit, and marks a genuine mid-cut honestly with an ellipsis when they cannot.
• Fixed: editing a category or tag description never cleared that archive page at your CDN. A publish purges the pages it changes, but a term edit saves no post, so nothing fired and the edge kept serving the old archive for the rest of its cache life. The term’s archive page is purged now, on edit and on delete.
• Fixed: acting on a review-bell card — Ignore, Allow, Block or Re-check — made the Dashboard’s “Who Reached Your Site” and “What Your Site Runs” cards flash back to their first-load loading state. The refreshed numbers an action returns now carry everything those cards read, the same payload the Dashboard itself is served.
• Fixed: one refusal path recorded nothing — a proven impostor turned away at a .well-known document vanished from the Request Log while the same refusal at llms.txt was written down. Every refusal leaves its line now.

1.50.1

• Fixed: the “Put this away” link is gone from the Findings screen. Putting a suggestion away only moved it into a fold at the foot of the same list — a collapsible in disguise — and only one kind of row ever carried the control, so part of the list arrived offering to be dismissed while the rest did not. Every finding now simply stays in its list, anything you had put away comes back on its own, and a connected assistant no longer sees a put-away ledger either.
• Improved: the Dashboard opens with its four small tiles, and the Today card now sits directly above the full-width cards it belongs with. The tile row used to sit between Today and everything else, cutting the run of cards in two.

1.50.0

• New: you can write your own robots.txt rules from the Crawler Policy card. A plain text box is appended to the end of the generated file, exactly as written — appended, never replacing, so an edit cannot break the Content-Signal line or the crawler blocks the plugin writes, and clearing the box is the whole reset. The live robots.txt is shown right above the box, fetched from your site exactly as a crawler receives it — which also honestly shows when a robots.txt file on disk or an SEO plugin owns the job instead (then the extra rules do not apply, and the box says so), and warns when your server wraps the file in an error code, which a strict crawler may ignore. Write your own Sitemap line and the plugin stops adding its own rather than printing two.
• New: the review bell’s cards now say why a visitor is unchecked, instead of one blank “not verified” covering every case. Each card names its own cause — identity checking is turned off, this bot cannot be checked because its operator publishes nothing to check against, its visits gave no clear answer yet, or it carries a real cryptographic signature from an operator this site does not know. Each cause reads differently on the card, so you can tell at a glance which unchecked rows are waiting for an answer and which will never have one — and the small info icon carries the full reason.
• New: each recommendation on a review card is now a small labelled chip that names the kind of action it asks for — nothing left to do, one plugin setting, or a rule at your host or CDN — with the full advice shown on hover. Three cards used to read as three similar sentences; now you can see before reading a word that one asks nothing of you, one is a click in Settings, and one needs your hosting panel. The cards are several lines shorter, and on a proven impostor the genuine engine’s verified traffic is counted right in the stats line so the fake never wears the real crawler’s numbers.
• New: the Client Decisions dialog opens each list with a plain notice saying what your decision actually reaches. A block is enforced at this plugin’s machine files — llms.txt, the markdown pages, discovery — not across your whole site, and it is separate from robots.txt, which is only a request a bot may ignore. Trusting a client sends nothing to the bot and changes nothing in robots.txt. When blocking is switched off in Settings, the Blocked list says so instead of quietly presenting rules that are not being enforced. And a trusted client whose name is also on the robots.txt block list now says so on its row — trusted at the door but publicly asked to stay away can be exactly what you meant, but the two may only disagree out loud.
• Fixed: the review panel could advise turning on “Verify bot identities” when it was already on — the nudge showed for any unchecked row that claimed a checkable bot, whether or not the setting was enabled. It now knows the setting’s live state and only suggests what would actually change something. The same fault was in the forged-signature card, which could recommend “turn on blocking” while blocking was already on; it now says the next attempt will be refused automatically.
• Fixed: a valid signature from an operator this plugin does not recognise displayed as “Passed” in a card’s details, indistinguishable from a recognised operator’s. Proof of identity is not a reason to trust someone: it now reads “Not confirmed”, names the signer, and says the site cannot vouch for them.
• Fixed: an entire explanation paragraph on review cards showed the “help” cursor as if it were hoverable — a style collision, and the question-mark cursor now appears only on the icons that actually carry a bubble.
• Fixed: the readiness card said “Everything looks good” when its advisor simply had no setup step left to suggest — a verdict it could not make, since the Cited figure above it can sit at 38% with every setup box ticked. It now says what it measured: “Nothing left to set up.”
• Fixed: the Report’s Citations card said “checks run on a schedule” to owners whose schedule is switched off. It now reads the schedule’s real state: off means “a check runs only when you start one.” And its “Open Citations” link landed on the Search tab; it now opens the Citations results it names.
• Improved: the Client Decisions rows read cleanly — a known crawler’s name is no longer printed twice, the decision verb no longer repeats the tab’s own name on every row, columns line up, long names shorten instead of pushing the date out of line, and on a phone each client folds to two tidy lines with a proper thumb-sized button.
• Improved: plain words on the undo buttons — “Stop trusting” and “Stop ignoring” replace the made-up “Un-trust” and “Un-ignore” — and the buttons look like buttons before you hover them, without turning red for what is a routine, reversible act.

1.49.0

• New: a warning that your CDN is blocking an AI company is now verified before it may keep saying so. Cloudflare counts crawlers by their User-Agent string, and anyone can wear one — on the site this was built for, a vulnerability scanner wore OpenAI’s crawler name through a thousand blocked requests a day, and the warning read “Cloudflare is blocking OpenAI” about a company that never visited. While such a warning is live, the blocked traffic’s source addresses are now checked hourly against what the company publishes. When everything checked proves to be an impostor, the warning stands down to a note that your edge is doing its job — nothing needs allowing. One verified request keeps the warning, and so does an inconclusive check: “could not say” is never read as “not them”.
• Improved: the demoted notice keeps everything the warning had — the date the blocking started, and your fold if you had made one — and it becomes a warning again by itself the moment verified traffic starts being refused.
• Improved: a connected assistant reading the edge summary sees the same verdict: how many of the blocked requests proved to be the company, how many proved to be impostors, and how many could not be determined.

The most recent releases are listed here, back to 1.49.0. Every earlier one, back to 1.22.0, is kept in the full history in the same detail: https://github.com/heera/agentimus/blob/main/CHANGELOG.md