SShortSingh.

Programming

0
ProgrammingDEV Community ·

Developer builds tool to stop Claude Code loading all skills at once, saving tokens

A developer noticed that Claude Code sessions were slowing down and consuming thousands of tokens at startup because every installed skill's description was being loaded into the system prompt regardless of relevance. To fix this, they built an open-source tool called 'master-of' that organizes skills behind domain-specific gates — such as design, development, or research — so only the skills relevant to a given task are loaded. A session-start hook automatically flags newly installed skills for sorting, and a check-skills command lets users see which skills are gated and estimate token savings. Plugin changes only take effect on restart, ensuring users remain in control of any modifications. The tool is MIT licensed, requires Bun or Node 22.6+, and is available on GitHub at devrudals/master-of.

0
ProgrammingDEV Community ·

Developer Builds Framework-Free Node.js API to Understand What AI-Generated Code Actually Does

A developer has launched Nexus API, a personal open-source project built using native Node.js capabilities rather than popular frameworks like Express or NestJS. The project was motivated by a growing concern that AI code-generation tools are producing fast results without ensuring developers understand the underlying mechanics. Rather than relying on AI to write and run code directly, the author uses it as a supplementary tool after first attempting problems independently. The project explores foundational concepts such as raw HTTP handling, manual routing, dependency injection, and layered architecture without framework abstractions. The goal is not to replace modern frameworks but to make using them a fully informed, conscious engineering decision.

0
ProgrammingDEV Community ·

Why iptables Shows No Rules After Switching Kubernetes to Cilium

When Kubernetes clusters migrate from kube-proxy to Cilium with kube-proxy replacement mode enabled, the familiar iptables KUBE-SERVICES chain disappears entirely from nodes. This happens because Cilium, a CNI plugin, takes over Service forwarding duties that kube-proxy previously handled by writing rules into the Linux kernel's netfilter/iptables framework. Instead of iptables, Cilium stores and enforces Service forwarding rules using eBPF programs loaded directly into the kernel, which standard iptables commands cannot read. As a result, tools like iptables-save, tcpdump on a ClusterIP, and conntrack return empty or misleading output even when Services are fully functional. Operators troubleshooting connectivity in Cilium-managed clusters must switch to Cilium-specific inspection tools rather than relying on traditional iptables-based diagnostics.

0
ProgrammingDEV Community ·

How to Import Existing AWS Accounts, Including Management, Into AFT Without Starting Fresh

A developer who inherited nine manually created AWS accounts documented how to adopt Account Factory for Terraform (AFT) on top of an already-live Control Tower environment, rather than starting from a clean slate. The process works by submitting an account request file with a matching email address, prompting AFT to recognise the existing account instead of creating a new one, then building out its pipeline downstream. Six standard workload accounts onboarded without issue, but key pitfalls include ensuring the account email matches AWS Organizations rather than Service Catalog, and understanding that only changes to control_tower_parameters trigger re-provisioning. Contrary to advice reportedly given by AWS Support, the Control Tower management account is not unsupported — it is explicitly enumerated as a shared account in AFT's own source code via shared_account.py. The author notes this capability has existed since AFT version 1.3.3, released in February 2022, but remains largely undocumented.

0
ProgrammingDEV Community ·

Critical Vite Flaw CVE-2026-39364 Exploited in Mass Scan for Cloud Credentials

F5 Labs recorded roughly 32,000 scanning events in August 2026 targeting exposed Vite development servers running versions 7.x and 8.x, with attackers probing for cloud credentials stored in .env files. The critical vulnerability CVE-2026-39364 allows unauthenticated attackers to bypass Vite's server.fs.deny restrictions and retrieve sensitive files via specially crafted HTTP requests, requiring no user interaction. Targets include development servers publicly reachable on ports such as 5173/TCP, where stolen AWS or Azure credentials could enable full cloud environment takeover. Vite has released patched versions 7.3.2 and 8.0.5, and administrators are urged to update immediately. As a further precaution, development servers should be restricted to localhost or internal networks, since changing ports alone is not considered a sufficient mitigation.

0
ProgrammingDEV Community ·

Reviactyl Launches as Open-Source Alternative to Pterodactyl Game Server Panel

Reviactyl is a newly released open-source game server management panel designed as an alternative to the widely used Pterodactyl platform. The project aims to retain familiar features such as server management, file manager, console access, and Docker-based environments, while offering a more modern user interface. It targets individual server owners, developers, communities, and hosting providers who manage multiple servers from a single dashboard. The source code is publicly available on GitHub, allowing users to fork, modify, and contribute to the project. Reviactyl does not seek to overhaul how game hosting panels work, but instead focuses on improving the overall user experience and customization options.

0
ProgrammingDEV Community ·

How to Spot Real ML Agencies vs. API Wrappers When Hiring in 2026

As of 2026, the gap between agencies claiming to offer custom machine learning and those actually delivering it has grown significantly, with most pitches amounting to API integrations rather than genuine model training. True custom ML involves training models on proprietary data, fine-tuning foundation models with measurable gains, or building specialised architectures — not simply chaining LLM calls or wrapping GPT APIs in a workflow. Operators evaluating agencies should request evidence of shipped production models, a structured data audit process, and a rigorous offline evaluation methodology that includes time-aware cross-validation. A reliable agency will conduct a discovery phase to assess data quality, schema, and leakage risks before quoting any model work. Structuring engagements as two-phase pilots with an exit option after phase one helps ensure financial and technical risk falls on the agency rather than the client.

0
ProgrammingDEV Community ·

Developer Builds Tool to Preview Text Length Limits Across Social and SMS Platforms

A developer has released a free browser-based tool called Multi-Platform Text Length Previewer, designed to help content creators and frontend engineers check how their text fits across different publishing channels. The tool displays character counts and visual overflow for platforms such as Twitter, LinkedIn, Google search snippets, and Instagram, each of which enforces different character limits. A separate SMS panel calculates message encoding and segment counts, using Unicode code-point iteration or GSM septet units rather than JavaScript's standard string length method. The developer notes that the tool's truncation logic is intentionally approximate, as it does not account for word boundaries, grapheme clusters, or carrier-specific SMS concatenation rules. It is intended as an early-stage check to catch obvious issues before final copy is pasted into the actual publishing platform.

0
ProgrammingDEV Community ·

Two Methods to Identify Shopify Stores: Manual Checks and Bulk Detection Tools

Developers and marketers needing to identify Shopify-powered websites can use two main approaches: manual inspection of page source code, response headers, and cookies, or a bulk automated tool for larger lists. Shopify stores leave consistent technical fingerprints, including references to cdn.shopify.com, a powered-by header, and session cookies such as _shopify_y. For bulk detection, a tool called Tech Stack Detector, built on the Apify platform, cross-references sites against over 7,600 open-source technology fingerprints to filter only Shopify stores from a list of domains. The tool also surfaces the broader technology stack of each site, including email, review, and checkout platforms, along with the specific evidence behind each detection. Results can be exported in CSV, Excel, or JSON formats, making it practical for lead generation, agency prospecting, or competitive research.

0
ProgrammingDEV Community ·

Why LLM Red-Team Reports Need Reproducibility, Not Just Screenshots

A software engineer argues that most AI red-teaming reports are little more than screenshots of bad model outputs, which cannot be independently verified or rerun. Credible reports, the author contends, must include a fixed and versioned probe corpus, raw prompts and model replies, and a SHA-256 checksum so compliance reviewers can verify nothing changed. A meaningful 0–100 score derived from how many probes a model failed should replace vague qualitative assessments, enabling CI pipelines to block deployments that exceed a set risk threshold. The author has built a small toolkit covering 35 probes across 17 attack classes, with a free 8-probe scan available without requiring an account. The core argument is that reproducibility transforms red-teaming from an anecdotal claim into a verifiable, auditable record.

0
ProgrammingDEV Community ·

Study finds AI-generated code flags zero true error-swallowing cases across 120 samples

A developer ran a controlled experiment generating 120 code samples using Qwen2.5-Coder 1.5B across 12 tasks in Python and TypeScript to test whether small AI models routinely hide or suppress errors. A Semgrep-based static detector flagged four candidates as potential error-swallowers, but manual review confirmed all four were either false positives or legitimate, documented fallbacks. The experiment found that Python generations mostly let exceptions propagate or raised them explicitly, while TypeScript samples favored try/catch blocks with logging or re-throwing. The researcher's key conclusion is that whether code truly swallows an error cannot be determined from syntax alone, as the same pattern can represent either a sound design contract or a genuine hazard depending on context. The study acknowledges significant limitations, including a single small model family, only 12 tasks, and sole-rater adjudication, and makes no broader claims about AI code versus human-written code.

0
ProgrammingHacker News ·

DIY E-Ink Frame Identifies Birdsong and Displays Vintage-Style Illustrations

A developer has built an e-ink picture frame that listens for birdsong and automatically displays matching illustrations styled after 1800s natural history prints. The open-source project, shared on GitHub under the name 'fugleramme', combines audio bird identification with an e-ink display. When a bird call is detected nearby, the device renders a vintage-style drawing of the identified species. The project was shared on Hacker News, where it attracted community attention as a creative blend of modern machine listening and retro aesthetic display technology.

0
ProgrammingDEV Community ·

How to Build a WebMCP Scoring Tool in Plain HTML With No Dependencies

A developer tutorial published on DEV Community walks through building a WebMCP tool from scratch using only a plain HTML file served via Python's built-in HTTP server. The guide demonstrates how to register a tool called 'score_facts' that evaluates a YAML document against five required fields — name, goal, who, what, and why — and returns a 0–100 score. Chrome supports the document.modelContext API natively behind an experimental flag, while other browsers can use a polyfill as a fallback. The tutorial emphasises defensive coding practices, including input validation and try/catch error handling, so that tool failures return structured error objects rather than raw stack traces. A security section also introduces URL allowlisting to prevent remote-fetch tools from being exploited as open proxies.

0
ProgrammingDEV Community ·

Mockoon's reimport feature leaves 559 Stripe API routes outdated after spec update

When Stripe released its 'basil' API version in March 2025, developers using Mockoon to mock Stripe's API found that reimporting the updated OpenAPI spec does not replace or update existing routes — it only adds new ones. This left 559 routes tied to the old API, meaning fields moved or removed in basil, such as current_period_start and current_period_end, continued returning stale data instead of reflecting the new spec. Additionally, Mockoon's static mock responses leave over 223,000 fields empty across the full Stripe spec, causing test flows to break when critical values like customer ID return blank. Developers must manually delete and re-add changed routes, risking loss of any custom edits made to those routes. The article contrasts this behavior with spec-driven mock servers that re-read the OpenAPI spec on every start, which automatically reflect the latest API changes without manual intervention.

0
ProgrammingDEV Community ·

Six AI Threats Every Organisation Should Know: A Practical Taxonomy

A structured AI threat awareness framework identifies six key risks facing organisations that use artificial intelligence tools, divided into four attack vectors and two impact categories. The vectors include Shadow AI, where employees use unsanctioned tools outside IT governance; Rogue Agents, where AI systems are granted excessive autonomy; Prompt Injection, where hidden instructions hijack model behaviour; and supply chain risks from unvetted models or datasets. When these vectors succeed, they typically result in either sensitive data leaving organisational control or AI-generated hallucinations being acted upon without verification. Each threat carries a defined exploit chain, warning indicators, and recommended remediation steps such as model inventories, DLP controls, and human approval gates. The framework is intended to help security and IT teams build consistent, structured defences against emerging AI-specific risks.

0
ProgrammingDEV Community ·

Why Multi-Agent AI Systems Fail in Production and How to Fix Them

Multi-agent AI systems that perform well in development often break down under real production traffic, not due to model errors but because of poorly designed orchestration layers. Key failure modes include silent errors from malformed inter-agent responses, context drift across long conversations, and conflicting outputs between agents with overlapping responsibilities. Developers are advised to treat every inter-agent handoff like a validated API contract, enforcing schema checks before downstream agents act on received data. State management across agents should evolve from simple shared scratchpads toward externalized state stores — databases or graphs — especially for long-running or auditable workflows. Resolving agent disagreements also requires a deliberate strategy, such as arbitration passes or deterministic tie-break rules, rather than defaulting to whichever agent responds last.

0
ProgrammingDEV Community ·

CauterRule v0.3.1 Fixes Silent Semantic Matching Bug, Doubles Agent Recall

Open-source tool CauterRule released version 0.3.1, addressing a critical flaw in its replay matcher that had rendered the semantic matching channel effectively inactive since launch. The bug stemmed from two issues: a semantic similarity floor set too high at 0.80, and trajectory embeddings polluted by class labels that artificially suppressed cosine scores. Developers lowered the floor to 0.62 based on measured paraphrase distributions and stripped class labels from signature embeddings to fix both problems. The fixes required no model or prompt changes, yet golden recall nearly doubled — rising from 0.17 to 0.38 on GPT and from 0.23 to 0.43 on LLaMA across 4,742 trajectory runs. The v0.3.1 release is available on GitHub and via pip, with full CLI support, framework adapters, and official rule packs included.

0
ProgrammingDEV Community ·

Concurrency Bugs Explained: Why Shared State Needs Synchronization

Concurrency within a single process involves multiple execution units, such as threads or goroutines, running simultaneously and potentially accessing the same data. A classic example shows how two threads each incrementing a shared counter can produce an incorrect result of 1 instead of 2, because both threads read the same stale value before either writes back. This race condition arises from the three-step nature of a simple increment operation — read, compute, and write — which can interleave unpredictably. Languages like Java, Go, and Python address this through synchronization primitives such as locks and mutexes, which enforce exclusive access to shared state during critical operations. An alternative approach, message passing, avoids shared state altogether by having execution units coordinate through exchanged messages rather than direct memory access.

0
ProgrammingDEV Community ·

Red Hat's AI-Driven Kubernetes Tool Raises Serious Cluster Security Concerns

Red Hat is developing an open-source MCP server that allows AI assistants such as GitHub Copilot and Cursor to directly read and perform operations on Kubernetes and OpenShift clusters. While the tool promises productivity gains by letting engineers describe tasks in plain language instead of running multiple kubectl commands, it also gives AI models a path to a cluster's control plane. Security experts warn that an AI assistant with write access carries the same power as a human admin, but is guided by fuzzy language interpretation and can be manipulated through crafted log lines or annotations — a form of prompt injection. The RBAC identity assigned to the assistant determines its actual permissions, meaning overly broad roles like cluster-admin could expose entire production environments to unintended changes. Recommended safeguards include starting with read-only access, scoping RBAC tightly per namespace, and requiring human approval before any write operations are executed.

0
ProgrammingDEV Community ·

WebRTC Series Part 2: How Browsers Find a Working Peer-to-Peer Path

A new hands-on tutorial in a WebRTC blog series explains how browsers move beyond discovering network addresses to actually establishing a usable peer-to-peer connection. The lab guides readers through building a local WebRTC connectivity playground using just two browser tabs, an HTML file, and a Python HTTP server — no backend required. Using the browser's BroadcastChannel API as a stand-in signaling mechanism, two tabs exchange ICE candidates, form candidate pairs, and run connectivity checks. The experiment demonstrates key ICE states — from 'checking' to 'connected' — and explains the difference between host candidates from local interfaces and srflx candidates discovered via STUN. It also touches on how modern browsers may use mDNS to avoid exposing raw local IP addresses directly to web pages.

← NewerPage 1293 of 3288Older →