AI-AGENTS
30 days · UTC
Synchronizing with global intelligence nodes...
A Token Saver skill trims Codex context to cut token burn
A custom Token Saver skill prunes reused context in Codex threads to slash token use without tanking results. In a walkthrough, a builder tracked 3.7...
Autonomous agent reportedly breached Hugging Face using OpenAI models — treat agents as adversaries in your stack
A new report details an autonomous agent using OpenAI models to run a multi‑day intrusion against Hugging Face via ordinary web services. The [Radica...
LLM agents: new exploit (“HalluSquatting”) and a simple defensive kill‑switch (“context bombing”)
Researchers showed AI coding agents can be steered to install malware via hallucinated package names—and that planted “context bombs” can make attacki...
GitHub’s PR Inbox is GA — and it now understands agent-authored work
GitHub made its redesigned PR dashboard generally available and added filters that treat agent-authored pull requests like first-class work. The new ...
Claude Sonnet 5 lands in dev workflows: default in Claude Code, cheaper than Opus
Anthropic’s Claude Sonnet 5 just shipped broadly, is default in Claude Code, and undercuts Opus with aggressive pricing. Anthropic launched Sonnet 5 ...
Loop Engineering, Not Prompts: How to Make Coding Agents Ship Safely
AI coding agents are moving from prompt hacks to loop engineering with verifiable checks, tighter scopes, and single‑agent workflows that actually shi...
Edge agents are arriving: AWS Kiro hits iPhone as local-first builds mature
AWS Kiro is landing on iPhone while local-first agents via Ollama and DIY builds like CrankGPT show cloudless AI is getting practical. AWS is taking ...
claude-mem now backfills historical telemetry to PostHog on upgrade
claude-mem added a one-time historical telemetry backfill that sends anonymized past usage counts to PostHog the first time you upgrade. Per the [v13...
Anthropic details agents that write and run code, pushing toward self-improving AI
Anthropic says it now uses autonomous agents to write and run code in its model development loop. In a new write-up, Anthropic outlines measurable pr...
Devin Desktop launches: a hub to run and supervise coding agents with a built-in IDE
Devin launched a macOS desktop app that centralizes local and cloud coding agents behind a built-in IDE. [Devin Desktop](https://devin.ai/desktop/) l...
NeMo Relay adds experimental Cursor hooks for agent observability (manual model routing required)
NVIDIA NeMo Relay can now observe Cursor agent lifecycle events via experimental hooks and a local wrapper, but LLM traffic routing remains manual. N...
Claude Code skills are moving from prompts to repo code — and your skills/ directory is now a signal
AI coding agents are shifting from giant prompts to repo-checked skills, and teams will be judged by their shared skills directories. A developer ana...
Meta patches Meta AI support bot that enabled one-shot account takeovers
Meta fixed a flaw where its Meta AI support bot could bypass 2FA and hand out password reset links, enabling easy account takeovers. TechRadar report...
Datasette adds an in-app Agent for working with datasets
Datasette introduced Datasette Agent, adding an in-app assistant for working with datasets. In his May update, Simon Willison says he launched Datase...
Hermes Agent vs OpenClaw and GoClaw: a practical guide lands on DEV
A new DEV post offers a practical Hermes Agent guide and compares it with OpenClaw and GoClaw. The article promises a hands-on walkthrough and side-b...
Google’s Gemini 3.5 Flash beats its own Pro tier at 4× speed and ~40% lower cost
Google launched Gemini 3.5 Flash, a “budget” model that outperforms Gemini 3.1 Pro on coding/agent benchmarks while running faster and cheaper. Per [...
Microsoft open-sources RAMPART and Clarity to put agent safety into CI/CD
Microsoft open-sourced RAMPART and Clarity to move agent safety testing into your CI/CD pipeline. Microsoft open-sourced [Rampart](https://www.infowo...
Cursor turns its IDE agent into headless infra with a public Agents SDK; Composer 2.5 steadies the hands
Cursor turned its IDE agent into headless infrastructure with a public Agents SDK, while Composer 2.5 made the agent steadier on long tasks. Cursor’s...
SAP shifts Joule Studio and AI Agent Hub to pro-code with gated agent workflows
SAP is reworking Joule Studio and AI Agent Hub to focus on pro-code agent development with gated workflows and GitHub integration. SAP’s first wave s...
How Claude Code Actually Works: a 6‑layer agent runtime
A deep dive maps Claude Code as a six-layer agent runtime with context compression and team orchestration. This visual explainer details Claude Code’...
Shopify makes its AI coding agent public-by-default in Slack
Shopify runs its internal coding agent River only in public Slack channels to turn everyday work into shared learning. Tobias Lütke’s team designed R...
Copilot CLI 1.0.45: OpenTelemetry spans and an /autopilot switch
GitHub Copilot CLI now emits standard OpenTelemetry spans for agent tool calls and adds an /autopilot mode toggle. The [v1.0.45 release](https://gith...
Codeex Browser Agent adds natural-language control of Chrome
Codeex Browser Agent now lets you control Chrome with natural language for end-to-end, multi-tab web automation. The update adds a Chrome plugin and ...