// edition · 2026-06-15
15 stories on AI dev tools, agents, and the coding stack, curated from the day's RSS haul by Agentic Dev's pipeline.
A guide published on Dev.to outlines techniques for standardizing Claude Code behavior across development teams, including committing a shared CLAUDE.md file to the project repository and using PostToolUse hooks to enforce TypeScript checks after each file edit.
Anthropic began metering Claude Agent SDK and `claude -p` usage separately from subscription plans on June 15, 2026. Automated and programmatic Claude usage now draws from a monthly credit, with overages billed at standard API rates, while interactive app and terminal use remains on existing flat...
A developer published a method to add native macOS notifications to the Claude Code VS Code extension using `terminal-notifier` and Claude Code's hooks system configured in `~/.claude/settings.json`. The setup triggers notifications for three events: permission requests, plan reviews, and task co...
A developer guide on Dev.to outlines steps for migrating CLAUDE.md files and system prompts from Claude Opus 4.8 to Claude Fable 5, using a Goal-Reason-Boundary-Verification prompt structure and a bundled `/claude-api` migration skill that handles model ID swaps and breaking parameter changes.
A developer built five automation tools using Anthropic's Claude API to handle tasks including email-to-task extraction, meeting transcript summarization, bug diagnosis, code review pre-checks, and clipboard-based knowledge logging. Claimed time savings range from 20 minutes per day on repeated l...
Xiaomi open-sourced MiMo Code, a terminal-native coding agent harness it claims outperforms Anthropic's Claude Code on agentic tasks exceeding 200 steps. The benchmark is self-reported, based on Xiaomi's own beta testing and a survey of 576 developers.
As AI agents increasingly take autonomous actions in production environments — provisioning resources, modifying settings, deleting data — traditional logging practices are proving inadequate for post-incident investigation. Regulatory frameworks including the SEC's disclosure rules and the EU's ...
Claude Code includes a `/config` menu with an "output style" setting offering modes such as "Learning," which provides explanations alongside generated code, and "Proactive," which anticipates next steps in automation tasks. The setting is documented but not prominently surfaced, according to one...
A developer post describes using hash time-lock contracts (HTLCs) to execute forward trades between two anonymous software agents without custodians or clearinghouses, by locking funds at agreement time and restricting claims to a future settlement window. The mechanism eliminates counterparty ri...
PagerDuty's Chief AI Officer argued that most AI incident management tools lack a necessary "agent harness" layer providing memory, contextual data, and system topology awareness, with roughly 70% of incidents attributed to live system modifications and updates.
A developer released speccraft, an open-source MIT-licensed Claude Code plugin that enforces spec-first, test-driven development by blocking code edits if no corresponding test file was updated first. The tool also auto-injects repository context at session start and routes specs to auxiliary AI ...
A developer benchmark called Ship-Bench tested a model referred to as "Fable 5" (an Anthropic model) against a composite of top-performing models, finding it scored an average of 96.49 with a perfect 100 in architecture, surpassing DeepSeek v4 Pro's prior high of 94.18, at an API cost of approxim...
Anthropic released Claude Fable 5, a new AI model, which was subsequently banned by the U.S. government within days of launch following a reported jailbreak, which Anthropic characterized as not significant. OpenRouter also released a "Fusion" feature combining multiple LLMs, accessible via API.
Claude's safety classifiers, designed to block bioweapon synthesis and cyberattack content, have reportedly flagged benign requests including pulled pork recipes, RNA sequencing data, a Snake game, and personal medical questions, silently routing them to an older model instead.
Researchers Arvind Narayanan and Sayash Kapoor argue AI has not caused mass software engineering layoffs, citing New York WARN Act data in which none of 160+ companies filing notices in the first year checked the AI-related layoff box. They identify the main bottlenecks to automation as requireme...