GitHub has made staged publishing generally available for npm and added tighter install-source controls in npm CLI 11.15.0. The change is practical supply-chain plumbing: CI can still build and upload a package, but a maintainer has to approve the final release before users can install it.
Read the piece →
Claude Code v2.1.149 adds usage breakdowns for skills, subagents, plugins and MCP servers, while fixing permission issues that could let commands move or write outside intended workspace boundaries.
Verification scripts are a practical bridge between automated pentest findings and the developer or security engineer who has to prove, prioritise and fix them.
Feature flags do not need to start as a platform migration. For many Laravel teams, a small package with Blade directives and explicit storage choices is enough to make releases safer.
A Texas A&M-led team built a standalone pump that mimics detailed human haemodynamic waveforms and sustained endothelial-cell vessel chips for up to 60 days.
NVIDIA says its Vera Rubin NVL72 rack links 36 Vera CPUs and 72 Rubin GPUs, while Jetson Thor targets physical AI with up to 2,070 FP4 teraflops in a compact module.
NASA says the current Caltech contract runs to 2028 and could be worth up to $30 billion if all options are exercised. The lab’s physical location is not expected to change.
Hubble’s view of MACS J1141.6-1905 is a useful kind of pretty: the central cluster is also a natural telescope. Its mass bends light from more distant galaxies, a gravitational-lensing effect astronomers use to study objects otherwise too faint or remote to see well. The visible spikes belong to nearer foreground stars and come from diffraction around Hubble’s mirror supports; the crowded centre is the cluster, roughly four billion light-years away in Crater. NASA notes that the image combines visible and infrared Hubble observations from programmes studying bright X-ray clusters and their lensed background galaxies. It is also an archive story: Hubble now holds more than 1.7 million observations, and new tools keep making old photons scientifically useful.
On May 25, 2008, NASA’s Phoenix lander touched down in Green Valley, in Vastitas Borealis, after entering the Martian atmosphere at nearly 13,000 mph. It was the first successful stationary soft-lander on Mars since Viking 2, and Mars Reconnaissance Orbiter’s HiRISE camera caught Phoenix descending under parachute — the first time one spacecraft photographed another during a planetary landing. Phoenix was built to study the Martian arctic: water ice, soil chemistry and whether the environment could preserve clues relevant to habitability. Its robotic arm delivered samples to tiny ovens and wet-chemistry instruments. The mission ended later that year as polar winter cut power, but it helped make Mars’s subsurface ice feel less theoretical and more like a material future missions would have to understand.
Direct imaging searches for exoplanets and brown dwarfs produce difficult data: the object of interest is faint, close to a bright star, and easily confused with residual optical artefacts. This paper revisits part of the SPHERE High-contrast Imaging survey for Exoplanets using modern deep-learning methods rather than only classical analysis. The aim is to see whether data-driven reprocessing can flag substellar candidates that were missed or left ambiguous in earlier reductions. The result is not ‘AI finds aliens’; it is a practical example of archival leverage. Expensive telescope campaigns may contain more science than the first pipeline extracted, especially as models for image subtraction, candidate ranking and false-positive control improve.
arXiv:2605.23700v1 →