Research
Claude Code 2.1.273 Stops Loading a Repo-Chosen Memory Directory Into the Prompt Under Read Blocking
Anthropic shipped claude-code v2.1.273 on 2026-09-15 with a fix that matters for anyone running agents on untrusted repositories: under permissions.blockReadsOutsideWorkingDirectories, a memory directory chosen by a repository's settings is no longer loaded into the prompt, recalled, indexed, or used by memory extraction. The same release fixes a subshell hiding a dangerous rm in bypass mode, fixes sub-agents and background agents being reported as failed when the final streamed reply omitted token usage or a model id, and adds opt-in x-claude-code-* request headers for LLM gateways behind CLAUDE_CODE_GATEWAY_HINT_HEADERS=1. The context meter and auto-compact were also double-counting advisor-tool turns.
↳ Follow the thread