Sarj.ai CTO Builds Open-Source Linter to Keep AI-Generated Code in Check

·1 views

As AI tools like Codex increasingly write code at speed, thorough human code review has become impractical at scale. To address this, the CTO of Sarj.ai developed an open-source linting tool called Standards, which encodes over three years of code review experience into more than 100 deterministic rules. The linter runs automatically during the pre-commit process, flagging violations in AI-generated code and prompting the AI agent to fix them without developer intervention. Rules range from simple conventions — such as enforcing uuidv7 over uuidv4 — to nuanced structural guidelines like the stepdown rule for code readability. Released under the MIT license, the tool aims to preserve the speed advantages of AI-assisted development while maintaining code quality and correctness.

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)

How to Test MySQL 8.4 Compatibility on AWS RDS Before a Blue/Green Cutover

AWS RDS Blue/Green Deployments offer a near-zero-downtime path for upgrading MySQL 8.0 instances to MySQL 8.4, the current Long Term Support release. However, the Green environment is read-only by design, which limits per-application compatibility testing before the final switchover. This becomes a significant obstacle when a single RDS instance hosts multiple projects with different upgrade readiness timelines. A workaround involves layering a snapshot-based test instance on top of the Blue/Green setup, allowing teams to validate each application individually against MySQL 8.4. Once all projects are confirmed compatible, a single coordinated switchover can be executed with minimal downtime.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Doom Runs Inside a Transformer Model Without Any Training

Researchers have managed to run the classic Doom game renderer inside a transformer neural network, using the Phi-3 architecture as the base model. Rather than training the network, a compiler was used to directly set the model's weights. This means the transformer executes Doom's renderer through autoregressive generation, with no machine learning training involved whatsoever. The project, published on ood.dev, demonstrates an unconventional use of transformer architecture as a programmable computational substrate.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Pontmore PR #12 Adds Direct Escrow Invocation Standard to PIP-01 Protocol

Developer merged PR #12 into the Pontmore protocol repository on August 11, 2026, extending the PIP-01 Escrow Descriptor to support standalone escrow service invocation without requiring a swap state machine. The update was driven by Issue #11, which called for a defined way for applications to create, fund, release, and cancel escrows directly between parties. The revised descriptor includes an optional service block specifying transport, authentication via Nostr public keys, canonical operations, funding models, and release decision formats. The patch also addressed structural vulnerabilities uncovered during implementation, including cross-instance replay attacks, funding-phase timeout enforcement, and mutual-consent deadlock scenarios. The changes establish a self-contained wire contract anchored by a normative OpenAPI schema, making the protocol more interoperable for financial applications built on Bitcoin and Lightning.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Hidden hreflang-switcher mismatch left 1,474 dead links undetected across 737 pages

A web developer discovered 1,474 broken internal links spread across 737 pages of their own site, all caused by a mismatch between hreflang tags and the site's language switcher navigation. The hreflang tags correctly marked certain pages as English-only, but the language switcher unconditionally displayed Chinese language buttons on every page, pointing users to URLs that returned 404 errors. The bug evaded all standard automated checks because the pages themselves returned HTTP 200 responses and the hreflang configuration was technically valid. The root cause was two independent systems answering the same question — whether a page exists in Chinese — with conflicting answers. The fix involved making the language switcher consult the same English-only page list already used by the hreflang generator, eliminating all dead links in a single change.

0 comments Read more at DEV Community