Claude Code gets plumbing fixes
Anthropic's latest Claude Code GitHub release is mostly infrastructure work: config commands, sandbox knobs, Bun 1.4, MCP OAuth polish, subagent UI fixes, and a long tail of reliability patches. It fits the day’s theme: less frontier-model spectacle, more maintenance that matters to developers using agents daily.
This is not flashy, but it is the kind of release that decides whether agentic coding feels dependable or brittle in real projects.
- –`/config key=value` makes scripted and remote Claude Code sessions easier to tune without leaving the prompt
- –Fixes for prompt caching, network-drive writes, long-running transcript cleanup, and nested subagents target real workflow failure modes
- –MCP OAuth and connector polish matters because Claude Code is increasingly an agent runtime, not just a terminal chatbot
- –The release reinforces a broader trend: AI tooling competition is shifting toward reliability, permissions, context, and integration plumbing
DISCOVERED
2h ago
2026-06-18
PUBLISHED
3h ago
2026-06-18
RELEVANCE
AUTHOR
AiChinaNews