Postmortem: Truncated AI Review Stream Passed Patch Gate as Valid JSON

·1 views

A postmortem describes a patch gate that accepted an AI reviewer's verdict after a streamed response was cut off mid-object. The gate's repair step appended a closing brace to the broken JSON, producing a parseable result that read only the verdict field and approved the change. One AI-authored patch reached the protected branch without a full review as a result, and the faulty parser ran for 19 days. A later replay on a healthy connection returned a failing verdict for the same diff, exposing the instability. The team replaced the parser with one that requires a declared byte length, a SHA-256 digest, and an end sentinel, and it now fails closed on incomplete responses.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Why Input Lag Can Mislead Developers: The Real Causes Behind Sluggish UIs

A developer describes users reporting sluggish input in a web app, where profiling pointed at JavaScript but the actual bottleneck was more complicated. The article explains that browsers pass input events through several stages, including OS processing, event coalescing, dispatch, and handler execution, any of which can introduce delay. It then outlines the rendering pipeline that follows JavaScript work, covering style recalculation, layout, paint, and compositing, which can postpone visible updates. The author notes that common DevTools signals such as Event Timing, Frames, and Long Tasks are easy to misread, since a handler may run late because of earlier work. Input latency therefore includes both event delay and render delay, and diagnosing it requires looking at the full pipeline rather than only the longest JavaScript task.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Meta open-sources code for building Muse AI gadgets

Meta has released open-source code that lets developers build gadgets powered by its Muse AI model. The article describes a builder who used the Amblyotube repository and the Muse model on a Raspberry Pi to create a gaze-tracking therapy setup for amblyopia. Muse's compact, quantized design allowed on-device inference at around 60 frames per second with a model under 4 MB. Fine-tuning on a few labeled sessions reportedly reduced calibration error from 2.4 degrees to 0.7 degrees. The author encourages others to assemble similar projects from the publicly available code on GitHub.

0 comments Read more at DEV Community

ProgrammingHacker News ·

Essay Argues Writing Is Closer to Thought Than Code

A blog post by Daniel Miessler contends that writing is more closely related to the process of thinking than to the act of coding. The piece appeared on Hacker News, where it received 3 points and no comments at the time of capture. The source material provided is limited to the headline, URL, and engagement metrics, so the full argument is not summarized here. Readers would need to consult the original article for the author's specific reasoning and examples.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

Study finds AI-assisted developers slower on some tasks, prompting a book on agentic software development

The author argues that a single measurement showing a productivity gain or loss should not be taken as a verdict on AI coding tools, since the full body of results forms a trend that slopes in a particular direction. The piece distinguishes between evidence grades, noting that weakly supported claims can motivate practices but cannot establish them, and that small studies generalize only to similar groups. It draws on the Lindy effect to argue that Parnas's 1972 information-hiding principles remain relevant because they limit the errors an AI agent can make. The author also notes that AGENTS.md, a configuration convention roughly eighteen months old, is promising but not yet proven to be lasting infrastructure. The post introduces a book on the Agentic Development Lifecycle and invites readers to consider how they would ship AI-generated code safely.

0 comments Read more at DEV Community