// edition · 2026-05-15

May 15, 2026

27 stories on AI dev tools, agents, and the coding stack — curated from the day's RSS haul by Agentic Dev's pipeline.

Top Signal · CLI Agents

2026 Best Claude Code Skills: Top 8 Skills, Setup Guide & Use Cases

Claude Code Skills are folders containing SKILL.md files that provide Claude Code with reusable workflow instructions, loading approximately 100 tokens per skill at session start and expanding only when triggered. A Dev.to guide outlines eight such skills for 2026, including Frontend Design, Play...

Dev.to - Claude

Tool Updates

Anthropic splits billing again: Agent SDK gets separate credit pools

Anthropic will separate programmatic and interactive usage billing starting June 15, giving Claude subscribers a distinct monthly Agent SDK credit pool. Credit allocations vary by plan: $20 for Pro, $100 for Max 5x, and $200 for Max 20x; unused credits expire at cycle end and do not roll over.

Pricing & Plans The New Stack

How I use Claude for PRs as a frontend engineer

A frontend engineer described using Claude to automate two steps before submitting pull requests: a local diff review via a custom `/local-review` command and a structured PR summary via `/pr-summary`, with Claude restricted from pushing to remote repositories via a CLAUDE.md configuration file.

Workflows & Tips Dev.to - Claude

Anthropic API in production: 5 things the docs don't tell you

A developer documented five production issues with Anthropic's API, including that prompt cache writes cost 1.25× normal input rates and only break even after roughly two reuses, and that 529 overload errors occur on 1-3% of requests during peak hours for Claude Sonnet.

Workflows & Tips Dev.to - Claude

AI Code Review Checklist: Correctness, Security, Performance, Readability

A developer guide outlines a four-stage AI code review process — correctness, security, performance, readability — with separate LLM prompts and checklists for each category. The approach, referencing Google's 2018 code review study, prioritizes logical errors and security issues over style and f...

Workflows & Tips Dev.to - Claude

Work with Codex from anywhere

OpenAI added Codex access to the ChatGPT mobile app, allowing users to monitor, steer, and approve coding tasks remotely across devices and remote environments.

CLI Agents OpenAI Blog

OpenAI brings Codex to the ChatGPT mobile app

OpenAI added Codex to the ChatGPT mobile app on iOS and Android, where it connects to a desktop machine running Codex via a relay layer rather than operating as a standalone product. The feature is rolling out to all Codex users, including free and Go plan subscribers, with macOS support only for...

CLI Agents The New Stack

"Claude 3, Qwen 6: why we set a different fix_verify retry cap per model"

Codens Purple, a code-fixing agent workflow, uses different retry caps per AI model: Claude gets 3 attempts, Qwen gets 6, and other models get 5, based on observed success-rate curves from production data. Claude's higher per-attempt success rate makes additional retries wasteful, while Qwen's se...

Agent Engineering Dev.to - Claude

Build Your Harness While You Wait

A software developer describes workflow strategies for managing AI coding assistants, arguing that idle time during long agent tasks (test runs, builds) should be used to configure prompts and guardrails rather than monitoring outputs manually.

Workflows & Tips Dev.to - Claude

OpenAI’s Codex is now in the ChatGPT mobile app

OpenAI is adding Codex, its AI coding assistant, to the ChatGPT mobile app on iOS and Android. The move follows a recent Codex update that enabled it to operate apps on macOS, as OpenAI competes with Anthropic's Claude Code.

CLI Agents The Verge - AI

OpenAI says Codex is coming to your phone

OpenAI announced that Codex, its AI-powered coding tool, will be made available on mobile devices, giving users the ability to manage coding workflows from their phones.

CLI Agents TechCrunch - AI

Building ResuMatch AI with TDD and AI-Assisted Development (Claude)

A developer built ResuMatch AI, a resume-tailoring app in ASP.NET Core, using Test-Driven Development to validate AI-generated code. The approach involved writing failing tests before prompting Claude to implement features, including a daily limit of three free generations per user.

Workflows & Tips Dev.to - Claude

The Rust sidecar pattern that fixes Python AI’s biggest weakness

A software architecture pattern pairs Python for AI/ML logic with a Rust sidecar that handles WebSocket connections and Kafka message fan-out, using a single Kafka consumer to distribute messages to thousands of concurrent clients via an internal broadcast channel.

Agent Engineering The New Stack

Agent Poke: Scheduled Check-ins for Codex and Claude Code

A developer released "agent-poke," an open-source Docker-based tool that sends scheduled "Hey!" messages to OpenAI Codex CLI and Claude Code at four fixed times daily to trigger subscription usage windows. The tool uses cron jobs and requires manual login via each CLI's official auth flow.

CLI Agents Dev.to - Claude

Ecosystem

I Made My Go Linter Talk to Claude ? Here's What I Learned About MCP

A developer built "godepvis," a Go static analyzer that flags code issues such as functions over 50 lines, methods with 6 or more parameters, and misuse of context.Background(). The tool was integrated with Claude Code via the Model Context Protocol (MCP) to enable automated code analysis.

MCP & Integrations Dev.to - Claude

Waymark v4.7.0 is Live — The Ultimate MCP Security Layer

Waymark released version 4.7.0 of its MCP security layer, which intercepts and enforces access policies between AI agents and filesystem or database tools. The update includes 30% faster policy evaluation, 50% lighter dashboard rendering, a symlink bypass fix, and transactional approval workflows.

MCP & Integrations Dev.to - Claude

I Tried TencentDB Agent Memory — Here's What the Token Reduction Looks Like

Tencent Cloud released TencentDB Agent Memory under MIT license in May 2026, a four-tier memory system for AI agents that offloads verbose tool output to local files while maintaining a compressed graph in context. Self-reported benchmarks show token reductions of 33–61% and task success improvem...

Open Source Tools Dev.to - Claude

Microsoft starts canceling Claude Code licenses

Microsoft is canceling most of its internal Claude Code licenses, roughly six months after opening access to thousands of employees in December. The company plans to redirect affected developers to its own Copilot CLI tool instead.

Industry & Funding The Verge - AI

Not so locked in any more

A medium-sized technology company used AI coding agents to rewrite its native iPhone and Android apps in React Native, citing improved framework capabilities and the reduced cost of future migrations. The anecdote illustrates a broader trend: AI-assisted programming is reducing the long-term risk...

Opinion & Analysis Simon Willison

Cloud code: Conductor joins the rush toward remote coding agents

Conductor, an AI coding startup founded in 2024, raised a $22 million Series A and launched Conductor Cloud, moving its coding agents from a local Mac app to hosted environments. Anthropic, Mistral, and Roo Code have made similar shifts toward cloud-based coding agents in recent months.

Industry & Funding The New Stack

Laravel MCP Implementation Cost: What Companies Should Budget in 2026

Laravel's official MCP server package entered public beta in September 2025, with production builds now deployed across fintech, healthcare, and SaaS. A practitioner's cost breakdown estimates architecture and tool design alone at $6,000–$15,000, with OAuth 2.1 authentication setup adding $2,500–...

MCP & Integrations Dev.to - AI

datasette-ip-rate-limit 0.1a0

Simon Willison released datasette-ip-rate-limit 0.1a0, a Datasette plugin that blocks IPs exceeding configurable request thresholds, built using OpenAI's Codex to address aggressive crawler traffic on datasette.io. The production configuration limits demo database paths to 60 requests per 60 seco...

Open Source Tools Simon Willison

Quoting Mitchell Hashimoto

HashiCorp co-founder Mitchell Hashimoto commented that programming languages have become fungible rather than lock-in, citing Bun's port from Zig to Rust — a transition he estimated took roughly one to two weeks — as evidence that language choice is increasingly expendable.

Opinion & Analysis Simon Willison