Urgent.News

What's breaking now, across thousands of outlets.

AI

agent-shell 0.78 updates

The latest version of agent-shell, 0.78, brings several improvements and new features. A standout feature is the persistent prompt, which offers a writeable prompt at all times in the buffer, allowing users to submit prompts and immediately receive responses while the AI agent handles the request. This persistent prompt is enabled by default and can be disabled if desired.

Another significant update is the ability for agent-shell to steer turns in AI agents that support the _session/steering ACP extension, enabling users to course correct in-flight prompts without waiting for processing to complete. Users can now steer by submitting with M-RET, or queue by pressing RET. The viewport's compose buffer remains functional, with prompts being handled through the same logic.

Additionally, agent-shell can now handle file dragging and dropping from external file managers, with images receiving a preview and other files being attached as @path links. Token cost headers have been added to the session display, showing cumulative cost when available. The editor now provides more idomatic slash command completion across the shell prompt, viewport compose buffer, and minibuffer.

The M-x agent-shell-copy-last-output command allows users to quickly retrieve the most recent output from anywhere in the buffer. Various bug fixes and improvements have also been incorporated, including fixes to file mentions carrying whitespace in paths, improved table rendering, and enhanced markdown support.

Written by urgent.news from Lobsters's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at xenodium.com →

More in AI

The AI Frontend Paradox: A Technical Blueprint

The AI Frontend Paradox: A Technical Blueprint Audience: Researchers, developers, and AI practitioners | Level: Advanced The AI Frontend Paradox's Core Mechanisms Interface Design and Cognitive Load…

  • AI assistants balance speed and accuracy, often sacrificing depth for efficiency
  • Speed-accuracy trade-off arises from prioritizing rapid code generation
  • Technical implementation requires encoding contextual information in models

Autonomous AI Agents Breach Online Retailers in Chained Attacks to Steal Payment Card Data

+09:00 Source: Gambit Security Severity: critical Type: Threat Intelligence Target Period: 2026-09-23T08:14:06+09:00 - 2026-09-24T08:09:37+09:00 (Asia/Tokyo) Original Link:…

  • Autonomous AI agents Hermes, Strix, and Cairn infiltrate retailers.
  • Attackers exploit SQL injection, weak configs, and sudo NOPASSWD settings.
  • Over 600,000 payment card records stolen from at least two companies.

More from Thursday 24 September →