AI · Article
Ten Coding Agent Deletion Incidents Traced to Shell Commands, With Safer Forms
GeekNews ·
At a glance
A review of 10 cases where coding agents deleted files, from September to November, traced each deletion to the shell command that actually ran. Most were not a wide delete by design. A command meant for one small folder was reinterpreted by the shell and pointed at something else, such as a drive root or the home folder. The author reproduced several of these on macOS and Windows 11 using test folders.
Key points
- Most cases involved commands run through a shell twice, so the outer shell expanded a variable to an empty value and turned a delete into a root-level one.
- Some path tools returned a root path for an input like C:\, which wiped about 600GB and could not be blocked by an empty-value check.
- Some commands were broader than expected, such as a path split on spaces that deleted macOS ~/Library, or a git clean that removed a whole parent folder.
- Seven of the ten cases happened on Windows, and a few deletions kept running for minutes after the agent stopped.
Why it matters
Coding agents can run real delete commands on your computer, so knowing how small mistakes in shell commands cause large losses helps you judge how much to trust them.
Original source
코딩 에이전트 삭제 사고 10건, 실행된 셸 명령으로 본 원인과 안전한 형태
GeekNews · Nov 17, 22:50 KST
Read on GeekNews ↗AI summarized the original article. Check the original for accuracy. Copyright belongs to GeekNews.
Claude Code · Ranking and more news
More from this field
- Google Brings Duplex Calling to Gemini, Letting AI Phone Businesses for You
Google has moved its Duplex phone-calling technology into the Gemini app, so Gemini can now call businesses for you. The system has changed a lot since 2018, shifting from narrow custom models to a single stack built on Gemini.
Google Research blog · Nov 21, 08:30 KST - Google Releases Project Genie, an AI That Generates Explorable Worlds
Google has released Project Genie, an AI tool that creates interactive worlds you can explore. The video shows how to access it, walks through a generated world and a T-Rex demo, and discusses why this early technology matters.
The AI Advantage · Nov 21, 08:21 KST - Anthropic Employee Says Claude Helped Prove Riemann Hypothesis
An Anthropic employee and mathematician, Levent Alpöge, says he used the company's Claude chatbot to create a detailed proof of the Riemann hypothesis over about six weeks. The proof has not yet been checked, and it is over 250,000 lines long in Lean, a computer language used to confirm maths proofs.
BBC News · Nov 21, 08:08 KST - Cole Medin Live-Streams His AI Software Factory Turning Ideas Into Working Software
Cole Medin runs his AI software factory live, taking rough requests from viewers through planning, building, and checking until working software runs on screen. He explains how it works and what he learned building it. The stream also shows a game the system built.
Cole Medin · Nov 21, 08:08 KST