SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

Selvedge v0.3.11 adds explicit rejection logging and tamper-evident memory for AI agents

0
·1 views

Developer tool Selvedge has released version 0.3.11, introducing formal support for logging rejected coding decisions made by AI agents — not just reverted ones. A new 'reject' change type allows agents to permanently record paths that were considered but never implemented, giving future sessions access to that institutional knowledge. The update also activates the previously dormant 'expires_when' field, letting teams attach invalidation conditions to stored decisions so outdated reasoning can be flagged automatically without network calls or AI inference. To back its append-only memory guarantee, the release introduces a SHA-256 hash chain, making the event log tamper-evident so users can verify no stored entry has been altered. These changes build on v0.3.10's proactive memory delivery, together aiming to prevent AI agents from confidently re-attempting approaches a team has already evaluated and discarded.

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)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Developer Launches Offline PWA Calculator with Multiple Modes and Accessibility Focus

A developer has released v1.0.0 of their first web project, an installable Progressive Web App (PWA) calculator built using HTML, CSS, and Vanilla JavaScript. The app includes multiple calculation modes such as standard, scientific, sales, interest, unit conversions, BMI, and age calculation. It features customizable button sizes, dark and light themes, and full offline functionality powered by a service worker cache. The app is designed with accessibility in mind, targeting older adults, students, shopkeepers, and general users. The developer has made the project publicly available on GitHub and is actively seeking community feedback on layout, design, and future features.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Why AI Agents Fail Repeatedly: The Reliability Crisis Reshaping Developer Tools

AI development has entered the autonomous agent era, where large language models execute multi-step workflows rather than just generating text, but this shift brings compounding failure risks. A 10-step agentic task where each step has a 95% success rate has only about a 60% chance of full completion, creating what developers call the 'Trust Gap.' Agents commonly fail due to reasoning loops that cannot resolve contradictions, context drift that causes them to lose track of original instructions, and tool misuse stemming from hallucinated or malformed API calls. These systemic issues mean the same input can produce vastly different execution paths, making consistent production deployment unreliable. Emerging infrastructure platforms are responding by treating agent reliability as a systems engineering challenge rather than a prompt engineering one.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Diffusion Language Models Challenge Autoregressive AI With Speed and Bidirectional Text

A new wave of diffusion-based language models is emerging as a serious alternative to the sequential token-generation method used by mainstream AI systems like GPT and Claude. Unlike autoregressive models, which generate text one token at a time and cannot revise earlier outputs, diffusion language models use a masking-and-denoising approach that allows bidirectional context and parallel processing. Recent releases highlight the trend's momentum: Inception Labs' Mercury exceeds 1,000 tokens per second, NVIDIA's Nemotron Diffusion models claim 2–8x throughput gains over comparable autoregressive systems, and Google has released Gemma Diffusion as an open-weights model. The technical foundations are being widely discussed this week, with explanatory posts from Cornell's Kuleshov group and researcher Sander Dieleman gaining traction on Hacker News. The core appeal of diffusion models lies in addressing structural weaknesses of autoregressive systems, including irreversible errors, sequential speed limits, and the inability to consider future context during generation.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Next.js instrumentation.ts: How to Fix Request State Leaks in App Router Tracing

Next.js's instrumentation.ts file provides a single initialization point for distributed tracing and error reporting in the App Router, with its register() function running once per server instance rather than per request. The onRequestError hook, stable since Next.js 15, captures errors from Server Components, Route Handlers, Server Actions, and Middleware before Next.js renders its own error response. Vercel's Fluid Compute reuses warm server instances across concurrent requests, meaning any module-level variable used to store request-scoped state becomes shared mutable state that can leak between unrelated users' traces. Only AsyncLocalStorage-based context propagation — the mechanism the OpenTelemetry SDK already relies on — correctly isolates data to a single request. The @vercel/otel package simplifies OpenTelemetry SDK setup by handling exporter selection and span processor configuration behind a single registerOTel() call, reducing boilerplate significantly.

0 comments Read more at DEV Community
Selvedge v0.3.11 adds explicit rejection logging and tamper-evident memory for AI agents · ShortSingh