Claude Code 'Run Until Done' Mode, AI Concierge, & Mythos Scan for Curl Bugs

Short summary

Claude Code's new 'run until done' mode enables autonomous goal-oriented coding workflows via the /goal command, allowing AI to iterate toward completion conditions. A practical wedding concierge built with Claude revealed both utility and user curiosity about system boundaries. Anthropic's Mythos vulnerability scanner identified one confirmed CVE and 20 additional bugs in the mature Curl codebase, validating AI's role in security analysis.

• •Claude Code /goal command enables agentic execution until completion condition met
• •Wedding concierge project demonstrated real-world prompt-engineering and safety challenges
• •Mythos successfully surfaced vulnerabilities in mature, well-vetted Curl codebase

Generated with AI, which can make mistakes.