StackChief - Learn how to code and share ideas.

Researchers handed six Codex CLI and Claude Code setups ongoing projects, their message history and the next request. The agents kept the code building and rarely did the requested work, because they lost track of what the project currently required. What that means for handovers, and for your job.

A new benchmark built from real CVEs ran Claude Code and SWE-agent on security-sensitive tasks. Most code that passed the tests could still be exploited: agents rarely found the validators and policies that kept it safe. What to change in prompts, instruction files and review.

Canva's CEO told staff that 1,271 roles, about 22% of the company, will be cut, changed or moved as Product, Growth and Platform merge and Sydney becomes home base for product teams. How it compares with four other tech cuts that week, and what the latest BLS layoffs data shows.

A new study read 3,387 AGENTS.md, CLAUDE.md, Copilot and Cursor instruction files from 1,000 GitHub repositories. Popular repos keep several formats that don't always agree, most files skip security and ownership, and many are written once. Here is which agent reads which file, and what to fix.

15 scenario and debugging questions for application engineers, each with what it tests, an answer outline and a follow-up.

Researchers analyzed 1.53 million public agent skills and found about 88,000 that tell coding agents to act without asking, skip safeguards or keep going. Most of those lines are unscoped, and skill scanners miss most of them. Here are the numbers and a skill audit checklist.