Visualithic Phyllix

Changelog

2.0.24

• Added: a note on the Licence panel when a site has not yet connected — activating a key was silently refused with no visible message in that state (the request never left the site), which looked exactly like a rejected or broken key. It now says plainly that connecting comes first.
• Housekeeping: cleared the last few Plugin Check findings ahead of resubmission — two admin-log columns (who fixed a call to action / a Boost rewrite) were plain text with no possible dynamic content but went through the table cell unescaped; both now run through esc_html() at output like everywhere else. One AJAX handler (the automation “run now” poll) was missing the same “nonce already verified above” annotation every other handler in this file already carries, so the scanner flagged its one $_POST read even though check_ajax_referer() already covers it.

2.0.23

• Fixed: the “Write the approved ones” step on Home could show “Running” / “Writing N articles” forever, with no way to clear it — even a hard refresh or an incognito window made no difference. The cause: once an article’s own record is updated to reflect that its page actually went live, that status was never added to the list of “this one is finished, stop counting it” statuses used in three places, so a genuinely finished, published article kept being counted as still being written, indefinitely. Fixed at the source with one shared list instead of three separate copies, so it can’t happen again.
• Fixed a related, more serious risk this same gap caused: the “Stop everything currently being written” button did not recognise an already-published article as something it must never touch, unlike an already-drafted one. It’s now refused the same way, so a stuck-looking step can never be “stopped” into silently marking a live, published article as failed and sending its idea back to be written all over again.

2.0.22

• Changed: Quick fixes and Win enquiries now also get applied on the regular 5-minute check, not only on the weekly run — so a fix that recovers (for example, a call to action that failed only because no destination was configured yet) doesn’t have to wait up to a week for its next chance. This never applies more than your weekly settings already allow; it only means less waiting when something becomes fixable between scheduled runs.

2.0.21

• Fixed: the floating “automation is running” progress banner could end up unreadable — white text with no visible background, only its border showing — when something else on the page (a WordPress core notice, the active theme, another plugin) won a styling conflict against it. Its colour and stacking are now locked down against that regardless of what else is on the page.

2.0.20

• Fixed: a call to action that failed to auto-apply because no destination was configured at the time stayed stuck in that state forever — every automated run kept retrying the exact same broken, empty-button proposal from when it was first composed, failing the same way every time, even long after real destinations were added in Settings. It’s now reset for a fresh attempt when this specific failure happens, so it genuinely recovers on the next run instead of being permanently stuck.

2.0.19

• Changed: each group under Published’s “Other automated actions” (Competitor gaps, AI Visibility, Win enquiries, and the rest) is now its own collapsible section, closed by default — a much shorter page at a glance, with the full detail still one click away.

2.0.18

• Added: Published’s “Pages Boost has improved” list now shows where each improvement was originally found — Results, Improve pages, or Keywords — next to who fixed it and when. Covers both automation’s own picks and your own manual clicks from those three screens.

2.0.17

• Fixed: Quick Fixes’ “link an orphaned page in” almost always failed with “Nothing could be linked safely,” even across pages that clearly had a natural place to link from. The step choosing which phrase to turn into a link was only ever shown a short, paraphrased summary of each candidate page — never its real text — so it was effectively guessing at wording it had never actually read, then presenting that guess with the confidence of an exact quote. It now reads the real text of a handful of the most topically relevant candidates (narrowed first, for free, using topic data already captured when each page was crawled) before choosing a phrase, so what it picks is a genuine quote rather than a coincidence.
• Fixed: a phrase could pass verification and still fail to apply, because the two steps checked different things — verification used a loose substring search that said “found” for a phrase sitting inside a heading, while applying it used a stricter rule that correctly refuses to turn a heading into a random inline link. Both now share the exact same rule, so a fix is never marked as findable and then silently fail later.

2.0.16

• Fixed: automated page improvements (Boost, drawing on Results, Improve pages and Keywords) always stopped after diagnosing a plan and sat waiting for approval forever, even with automation turned on and even on a page with nothing stopping it — the setting’s own description already promised full automatic prepare-and-apply, but nothing ever actually carried an automated diagnosis through to the rewrite. It now does, under the exact same conditions the setting already describes (never on a page in your navigation menu).
• Fixed: a call to action’s automatic destination was picked by whichever contact link happened to be listed first, regardless of type — a page with several WhatsApp numbers and one quote page could end up offering two WhatsApp numbers instead of a quote page and a WhatsApp number. Automation now picks by destination type (a page on your site, then email, then WhatsApp, then phone, then text message), one of each of the top two types present, however many links you’ve added or however they’re ordered.
• Added: Published now shows every call to action and every page improvement that has actually gone live, whether you added it yourself or automation did — collapsed by default, with who did it, when, a link to view it live, and an “Edit here” that lets you change the wording or destination and update the live page, or undo it back to how it was before.
• Fixed: undoing an applied page improvement didn’t update its own record, so it could still show as “applied” after being reverted. Undoing a call to action or a page improvement now correctly puts it back to waiting for review either way.

2.0.15

• The real fix for the daily automated cycle stalling after reading the site and pulling Search Console data, then going quiet on some hosting — every AI call outside the article-writing pipeline (reading a page, building the knowledge base, mapping content, finding SEO/keyword gaps, checking AI Visibility, checking competitors, proposing ideas, scoring findings, composing a Win-enquiries call to action, writing the service-areas page, learning from a published edit) was still holding its own WordPress request open for the length of the call — exactly the exposure the article pipeline’s own async rework already closed. On a host whose CDN/edge layer or PHP-FPM kills a long-held connection, that call just vanished mid-run with no error, which is what read as automation reliably completing the early, cheap steps and then stopping. All of it now starts the call and returns immediately, resuming once the result actually lands, the same pattern the writer already used.
• Found and fixed two real, already-live bugs while doing this: competitor discovery’s own docblock said it fired asynchronously, but the actual call passed extra arguments to the synchronous method instead — PHP silently drops arguments a function doesn’t declare, so this ran fully blocking for months and its result was thrown away every time, meaning discovered competitors were never actually added. Separately, the three automated “is this worth an idea” sweeps (competitor gaps, keyword gaps, AI Visibility losses) scored every finding with a blocking AI call inside a loop, one call per finding, back to back, in a single request — now each finding is queued as its own async job instead.
• A synchronous helper only ever used inside an already-async job (competitor-name extraction inside the AI Visibility check) also moved onto the same async pattern, plus a couple of confirmed-dead synchronous functions removed where an async twin had already fully replaced them.

2.0.14

• Fixed: the real bug behind two remaining translator-comment warnings — the comment sat above the variable assignment, not directly above the translated string two lines below it (a multi-line ternary). Moved to the correct line, above each branch.
• Fixed every remaining “output not escaped” warning on the admin screens, root and branch. A phpcs:ignore/phpcs:disable comment does not reliably suppress this check, however correct the reasoning underneath it, so every instance of two patterns was rebuilt to need no suppression at all: (1) a card-building method that returned a pre-built HTML string for another method to echo — render_pipeline(), render_idea(), render_knowledge_card(), rules_list_html(), render_prep_state(), render_conversion(), render_seo_check() and step_bar() now print their own output directly, and the few places that still need the HTML as a string (an AJAX response) capture it locally instead; (2) a value already cast to a number earlier in the method, then printed several lines later — every such value is now cast again at the exact point it is printed, which is what the check is actually able to verify.

2.0.11

• Fixed: three untranslated placeholders in automated keyword/competitor-gap proposals were missing their “translators:” comments — added.
• Fixed: the review panel’s status line had its translator comments sitting above the wrong line for the sniff that checks for them — moved directly above each translated string.
• Fixed: Tested up to used a patch version (7.1.1); wp.org only accepts major.minor — trimmed to 7.1.
• Fixed: the short description was cut off at wp.org’s 150-character limit — shortened to fit.

2.0.10

• Major release: optional automation. Off by default — Phyllix keeps working exactly as before (drafts only, every page edit shown as a proposal first) until you deliberately turn something on in Settings → Automation.
• Granular, opt-in levels: automatically send findings to your Ideas queue, approve ideas, write drafts, publish articles, and apply quick fixes/calls to action/page improvements — each its own toggle, so you can automate one stage and keep the rest manual. A configurable pace and a budget split between new articles and improving existing pages.
• Full audit trail: every idea, draft and published article records when it was proposed, approved, written and published, and by whom — visible everywhere that matters, distinguishing your own actions from automation’s.
• Navigation-menu pages are never touched by any automated action, at any setting, with no override — including automatic internal linking, a gap closed in this release.
• Automatic publishing now checks for finished drafts every few minutes rather than once a week, and works through a backlog oldest-first rather than newest-first, so nothing genuinely finished is ever left stuck waiting behind newer work.
• Published pages: “Edit here” now opens everything in one place — the SEO fields, the article body, then Update/Unpublish/Trash/View/Cancel, with the full review (score, what was fixed automatically, what is still open) underneath. Unpublish and Trash reach a published article for the first time; “Mark ready to publish” does the same for your own drafts. “Reconnect Phyllix” in Settings covers a stuck connection, and “Clear error & try again” fixes a failed step showing a stale error.
• AI Visibility question generation rebuilt on genuine AI reasoning instead of a template. Win-enquiries calls to action can compose and apply themselves automatically when you turn that on. A “possible duplicate” flag on drafts that weakly overlap an existing page.
• A number of correctness and billing-accuracy fixes found in an internal audit ahead of this release: several real-money AI calls that were running untagged (never double-charged to you, but invisible to our own cost tracking), a queue retry-counting bug, a budget-reservation race between overlapping automation runs, and one AI-research destination missing from External Services — added.
• Every automation feature above is entirely optional. If you do nothing in Settings → Automation, Phyllix behaves exactly as it always has.

Versions before 2.0.0

Full version-by-version history available in the plugin’s own repository. Every version before 2.0.0 predates the automation feature entirely — each one describes Phyllix’s manual-only, draft-and-propose behaviour, which remains the default today.