Vibeleaderboard
← All Intel
Intel / repo

gemini-cli v0.65.0 — Release v0.65.0

Source
github.com
Author
gemini-cli-robot
Date

Context

Gemini CLI 0.65.0 is almost entirely fixes, and the one with the clearest consequences concerns folder trust. When you open Gemini CLI in a folder you have not marked as trusted, it ignores that project's own settings and uses an empty set instead. According to the pull request, that empty set still pointed at the real .gemini/settings.json file and was not marked read-only, so running gemini add or gemini mcp remove at project scope wrote the empty settings back to disk and erased every key already in the file.

The fix marks the substituted workspace settings as read-only while the folder is untrusted, makes any attempt to write to them fail before anything reaches disk, and has the mcp add and remove commands refuse with an error for an untrusted project; user-scope changes still work, and trusting the folder restores normal writing. Elsewhere in the list, judging by the pull request titles, the release aligns OAuth callback validation with RFC 9207, stops verification and OAuth retry loops, fixes duplicated tool responses and lost history when resuming sessions, and adds custom OTLP headers to telemetry configuration.

Key takeaways · AI-distilled
  • Gemini CLI v0.65.0 lets telemetry configuration set custom OTLP headers.
  • Two Gemini CLI auth fixes: infinite verification and OAuth retry loops are prevented, and cached credentials are cleared when a user re-selects Google login.
  • Gemini CLI v0.65.0 prevents session poisoning from a rapid Ctrl-C interruption and fixes a hang on pressing Enter in interactive mode.
  • @<directory> references no longer trigger eager recursive file reading, and ignore filtering was optimized with subtree pruning.
  • If one extension fails the check, Gemini CLI now keeps loading the other extensions.
Terms in this piece · Glossary
  • MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
  • tool permissions — The rules governing which tools an agent may call and which need confirmation — the boundary between a mistake and an incident.
Recommended reads
Comments

Checking sign-in…

Loading comments…