Archive
Monday, August 3, 2026
10 Stories

Radar Daily Briefings

A clearer signal for WordPress and engineering

  /\_/\
 (=^.^=)
 (")_(")
				
  /\_/\
 (=^.^=)
 (")_(")
				

Source
Signal

No stories match the selected filters in today's edition.

JFrog reports fabricated SQLite CVE advisories after source audits

JFrog Security Research reports that a newly created GitHub repository published SQLite vulnerability advisories that failed technical verification. The report says six reviewed advisories contained nonexistent functions, incorrect line references, invalid SQL, or fabricated fixes; a broader audit found 54 of 55 advisories from the same account were fabricated.

JFrog compared SQLite source versions, built releases in isolated Docker containers, ran supplied PoCs under AddressSanitizer, and reviewed NVD and GHSA metadata. The tests reportedly found no crashes or claimed vulnerable execution paths.

The findings matter for vulnerability-management and AI-triage systems that ingest CVE data automatically. The supplied evidence is a single research report, and it does not establish the final status of the records.

Karpathy reports Opus 5 generating a procedural Three.js world

Andrej Karpathy reports an informal Opus 5 experiment in which he supplied the first paragraph of The Lord of the Rings, a one-million-token budget, and a request for a Three.js rendering. The model worked for about two hours and produced roughly 5,500 lines of code.

Karpathy says the program procedurally arranged polygon assets in three-dimensional coordinates and animated them to render the story. He reports that the model audited the output by taking screenshots at different points, slowly and with several mistakes.

The demonstration suggests potential for highly customized generated worlds, but remains experimental. Its result was visibly rough, and the evidence is neither controlled nor independently verified.

OpenAI reports ten AI-generated advances in mathematics

Simon Willison reports that OpenAI used an internal version of Astra to pursue ten mathematical problems whose main results had seen no progress for at least a decade. He says the work produced Lean 4 formalizations in the openai/ten-proofs repository, along with a paper and an LLM-generated reconstruction of the proof process.

OpenAI reportedly spent less than $2,000 at GPT-5.6 Sol token prices on each problem. The supplied evidence does not establish how many attempts failed or provide the prompts, system architecture, or evaluation methodology.

The repository and formalizations may support inspection, but the results remain claims reported by a third-party blog in this evidence.

Gutenberg 23.6 details WordPress 7.1 editor changes

The Gutenberg Changelog episode reports the Gutenberg 23.6 release and discusses WordPress 7.1 at Beta 4. Gutenberg 23.6 is described as the final feature release contributing features to WordPress 7.1.

Highlighted changes include configurable tablet and mobile viewports in theme.json, responsive text alignment, grid autofit, SVG icon collections, editable blocks inside Custom HTML blocks, and expanded collaboration notes. The post editor is also moving into an iframe, following the site, template, and pattern editors.

The iframe transition is intended to isolate editor designs from the surrounding admin page, but older blocks and collections may exhibit compatibility quirks. Developers are encouraged to test WordPress 7.1 and review the related guidance.

Modelling Actors with Redex begins with an ISWIM semantics model

The first article in a three-part series builds a runnable model of an ISWIM-style functional programming language using PLT Redex. The source is presented as a literate Racket file that readers can download and run.

The model defines S-expression syntax, values, primitive operations, evaluation contexts, reduction rules, recursive unfolding, and binding declarations for lambda and rec. Evaluation contexts establish left-to-right reduction order, while Redex supplies substitution-related machinery from the declared binding structure.

Examples and unit tests exercise arithmetic, pairs, recursion, and conditionals. Actor-specific imperative actions are deferred to Part II, and the grammar currently permits stuck empty begin expressions.

llm-mcp-client 0.1a0 releases MCP tool access for LLMs

Simon Willison’s Weblog announced llm-mcp-client 0.1a0 on 31 July 2026. The entry labels it a release and describes the package as providing access to tools from MCP servers as LLM tools.

The supplied material does not explain how the client connects MCP servers to an LLM workflow, which protocols or runtimes it supports, or how tools are represented.

This gives engineers a named integration point to investigate for MCP-based LLM applications. However, the release entry alone does not establish compatibility, performance, adoption, or production readiness.

WordPress Playground v3.1.48 releases persistence and PHP-WASM fixes

WordPress Playground v3.1.48 is an official release containing implementation fixes, interface changes, and documentation updates across the browser-based WordPress environment.

The release supports wildcard CORS proxy origins and adds changes for PHP WebAssembly messaging, mounted-directory scans, SQLite stress fixtures, OPFS synchronization, autosaves, browser-stored reopens, and ZIP import and export workflows. It also updates Dock, Blueprint, storage-status, and documentation behavior.

Engineers integrating Playground can install @wp-playground/[email protected] from npm and review the listed changes for affected workflows. The supplied release notes do not include independent compatibility or performance evaluation.

Perl CGI contact form breaks after hosting migration

An article describes a Perl CGI contact form that stopped working after its small-business website was migrated to a new hosting environment. The browser displayed the Perl source instead of executing the script, and the author says the hosting company confirmed that Perl CGI was no longer enabled.

The incident illustrates ecosystem-driven failure: the application had not changed, but its hosting assumptions no longer held. The article also notes that the site had migrated to the NMS project in 2023, despite that software originating around 2000.

For engineers, the case emphasizes checking runtime and hosting dependencies before replacing stable software. The broader discussion is opinion based and supplies no independent measurements of hosting-market trends.

WordPress Training Team recap outlines triage and contributor plans

The WordPress Training Team’s July 28 meeting recap documents ongoing contributor coordination and proposed work. The team reports about 555 open issues, handbook updates, volunteer recruitment, and an upcoming Office Hours session on August 22, 2026.

The proposed workflow divides issues into smaller categories, starting with potential translation-review work, while contributors also help validate feedback, review content, vet topics, and create lesson thumbnails. A regular biweekly GitHub triage meeting remains under discussion and has not been confirmed.

The recap also reports that Learn WordPress is fully translated into Marathi and announces a Plugin Development cohort. These are community and training updates rather than merged engineering changes.

WordPress plugin translation review requests 10 locale approvals

The author of Prevent Browser Caching has submitted translation suggestions for the plugin’s UI and current readme across 10 WordPress locales. The post asks Polyglots teams to review and approve them where appropriate.

The submission covers 86 UI strings and 95 readme strings. The author reports 181 strings waiting in Stable and Stable Readme for each locale and links to the translate.wordpress.org project.

This is a proposed localization review task, not a plugin release or code change. The author notes they do not speak the listed languages, so approval depends on the relevant locale teams’ review.