Galleries crash on PHP 8.3 after the 4.2.0 update
- Recurring bug
- High
- 14 conversations
- Getting worse
- Pro
Help desk intelligence, beyond AI replies
Overdesk reads the recent conversations in your queue, ranks what's going wrong, from bugs to customers ready to leave, and files the ones you approve as issues in GitHub, Linear or Jira.
Draft replies are included.
14 days with every Pro feature. No card.
You’ll need admin access to your help desk, an Anthropic API key with credit, and your docs site’s address.
Works with Help Scout FreeScout Zendesk Intercom
Monday morning. 212 new conversations since last week. Some of them are the same problem.
Overdesk groups them into one finding, ranked by severity, with how many conversations mention it and whether it is getting worse.
TrackerGitHub
Repositoryframewise/gallery-pro
Written from 10 of 14 conversations.
Email addresses and card numbers are taken out first.
Issue #482 opened.
Gallery pages show a critical error after updating to 4.2.0. Reported on PHP 8.3 sites.
Update to 4.2.0 on PHP 8.3. Open any page with a gallery.
PHP Fatal error: Uncaught TypeError: count(): Argument #1 ($value) must be of type Countable|array, null given in .../class-lightbox.php:214
WordPress 6.6 · PHP 8.3 · Framewise Gallery Pro 4.2.0
Rolling back to 4.1.3 works. Clearing the cache did not.
Does it happen on PHP 8.2?
You review the brief, then file the issue in GitHub, Linear or Jira. Email addresses and card numbers are taken out first.
What the AI reads
Hi, since updating to 4.2.0 our gallery page shows a critical error. My licence key is FWG-7K2P-9QX4-M3ZLLicence key · hidden[[license_key_ljpllngh]]. Login is admin, password: Sunset!Harbor42.Password · hidden[[code_hjkeelmn]] You can also reach me at dana@northfield.exampleEmail address · hidden[[email_kdebjbpg]].
Hi Dana, thanks for the details. This error appears after updating to 4.2.0 on PHP 8.3. Rolling back to 4.1.3 from your account page brings your galleries back while we look into it. You also shared your admin password by email, so please change it.
Drafted from your past answers and your code. Overdesk replaces the licence key’s body and the password with stand-ins before the AI reads the ticket. The draft is an example, not live output.
Hi, since updating to 4.2.0 our gallery page shows a critical error.
Draft by Overdesk · not sentSent by Sam
Hi Dana, thanks for the details. This error appears after updating to 4.2.0 on PHP 8.3. Rolling back to 4.1.3 from your account page brings your galleries back while we look into it. You also shared your admin password by email, so please change it.
That was a script. Run it on your own queue.
Drafts can wait in Help Scout as an unsent reply, and a person sends it, here Sam. Overdesk will not send a reply on its own if it repeats a key or password it hid.
A made-up queue for a made-up plugin. Bug and churn findings and customer context are Pro features.
“Overdesk’s weekly report picked up a group of tickets about a performance problem. I logged an issue from the report, and my coding agent read the tickets, investigated, and found the cause. I wouldn’t have even noticed it.”
An example, start to finish
Framewise Gallery Pro is a made-up plugin. This is how the chain runs on a queue like yours.
When something needs a decision, the weekly summary opens with it.
A recurring problem becomes one finding, with how many conversations mention it and whether it is getting worse.
You review the brief and file it, with email addresses and card numbers taken out.
A developer, or your own tools, picks up the issue and the brief.
Overdesk stops at the issue. Your developers, or your own tools, take it from there.
In a note on the ticket, in Slack, and on a test site of its own. The examples are made up.
Turn it on, then write @Overdesk in a note on a Help Scout, Zendesk, Intercom or FreeScout conversation. It searches your docs, past tickets and code the way a teammate would, and answers in a new note that cites what it read. A question leaves the draft as it was.
Turn it on in Settings, then mention the bot and ask in plain English: what bugs came in this week, what needs a decision, what your latest report found. It answers from your own workspace data, and it never takes an action or messages a customer.
When a customer reports a bug in a WordPress plugin or theme you list in Settings, Overdesk can start a private test copy of WordPress with it and try what they describe. The reply says what it checked, or that it was not tested.
Connect your help desk, then add the rest of your stack.
Overdesk reads your queue here, and drafts arrive where your team already works.
File a finding as an issue after you review it. GitHub also lets a draft read your code at the customer’s version, when the ticket or their account says which one.
Drafts learn from your past answers and published help articles. Add the Notion pages your team writes, and only the Google Docs you pick.
Drafts can cite a customer’s licence, plan and orders. Read-only: Overdesk never changes an order, a subscription or a licence.
On Pro, alerts for serious new problems and a daily morning briefing go to Slack or ClickUp. In Slack you can also ask Overdesk about your queue (Pro).
Before the AI reads a ticket, Overdesk swaps out the email addresses, card numbers, licence and API keys, and labelled passwords customers paste in.
It does not catch everything. How redaction works →
Held for a person: this reply repeats a licence key’s stand-in.
Example draft
Hi Dana, thanks for the details. Your licence key is FWG-Licence key · hidden[[license_key_ljpllngh]]. Rolling back to 4.1.3 from your account page brings your galleries back.
Overdesk will not send it on its own. Instead of an unsent draft reply, it goes to your team as an internal note, stand-in included.
Pasted
define( 'DB_PASSWORD', 'Harbor!Lamp77' );The AI reads
define( 'Password · hidden[[code_algmcpeb]]' );Hidden before the AI reads itPasted
Stripe secret key: sk_test_FICTIONAL42The AI reads
Stripe secret key: Key · hidden[[license_key_oeeenjlc]]Hidden before the AI reads itPasted
Login: admin / password: Quarry#Lantern58The AI reads
Login: admin / Password · hidden[[code_akclckmf]]Hidden before the AI reads itPasted
My card number is 4242 4242 4242 4242The AI reads
My card number is Card number · hidden[[credit_card_jfdiegmi]]Hidden before the AI reads itBuilt for people who can’t afford an AI mistake in front of a customer.
The success tax
Most support bots bill per resolution, outcome or task, so the better they get, the bigger the invoice. An Overdesk seat costs the same at any volume.
At 500 automated resolutions a month and 5 seats:
List prices read on each vendor’s own pricing page on 9 October 2026. Zendesk’s lower rate is for resolutions you commit to in advance, its higher one is pay as you go. Each vendor also bills its own seats (Help Scout $25 to $75 per user a month) and counts a resolution or an outcome its own way, so check its page before you budget.
Overdesk’s row is the Pro plan at the seats you pick. It runs alongside your help desk, whose seats you keep, and its AI usage is billed to you by Anthropic, outside this row.
14 days with every Pro feature. No card.
No. Findings come to you in the weekly summary, Slack or email, and your team keeps replying in its help desk. You open Overdesk when you want to file one as an issue. How findings work.
No, not unless it is switched on. By default a draft waits as an internal note, and on Help Scout you can choose an unsent draft reply instead. Sending on its own needs two switches: Overdesk turns it on for your workspace, then your admin turns it on. With both switches on, Overdesk sends the replies your rules approve on Help Scout, Zendesk, Intercom and FreeScout. It sends only on these channels: email on Help Scout, email and API tickets on Zendesk, Messenger and email on Intercom, and email on FreeScout. A reply on any other channel waits for your team. Overdesk will not send a reply on its own if it repeats a key or password it hid. How replies go out.
It can read your code from GitHub to write better drafts, at the customer’s version when the ticket or their account says which one. Overdesk never writes code or opens pull requests: it stops at the issue. What it reads from GitHub.
Yours for drafts, triage and reports: they run on the Anthropic account you connect, and Anthropic bills you for them. Search is the exception. Indexing and searching your help content and past conversations runs on an Overdesk-owned Voyage AI account, using pseudonymised text. How we handle your data.
Something we didn’t cover? Email us or book a demo.