Skip to content
Agentsmedium signalverified

Gemini CLI stops loading repository MCP servers in an untrusted workspace

A fix merged into Gemini CLI on August 28, 2026 changes the order in which workspace trust is decided, and it shipped the next day in the nightly build v0.59.0-nightly.20260829. Environment signals such as GEMINI_RESTRICTED_MODE are now read before local options, and an untrusted signal ends the check immediately. In the A2A server, MCP server definitions that come from the repository are dropped when the workspace is not trusted. The stated purpose is to keep the server from starting processes it should not start.

By Redakcija WebAiRadarPublished 2 min readwritten by a model

Trust in an agent tool is not a label on a folder. It decides which files are read, which tools are exposed, and which processes get spawned. This fix moves that decision earlier and makes its failure mode a refusal rather than a fallback.

What changed in the trust check

In the core trust utility, the check now evaluates GEMINI_RESTRICTED_MODE, GEMINI_FOLDER_TRUST and GEMINI_CLI_TRUST_WORKSPACE before it looks at the local option that enables folder trust. When one of those signals says the environment is untrusted, the function returns an untrusted result at once, with the source recorded as the environment.

The configuration layer follows the same order. Untrusted environment variables are evaluated early, so the folder check returns false before any stored context or fallback is consulted. Components that ask whether the folder is trusted now get the answer the environment intended, not the one a fallback produced.

Why dropping the server definitions matters

A repository can carry its own MCP server definitions in configuration a person never opens. Those definitions describe processes the agent launches, so a definition that survives an untrusted-workspace check is a way to run code by being cloned and opened.

In the A2A server the configuration loader now checks the trusted argument, and when the workspace is not trusted it strips the repository-scoped MCP server definitions before they reach policy settings or configuration parameters. The pull request adds a test suite for the trust resolution, the fail-closed behavior and the filtering.

What this is, and what it is not

This is a hardening change, not a response to a published vulnerability. No advisory accompanies it, and the pull request describes the behavior as unintended rather than exploited.

It is also, for now, only in a nightly build. Anyone tracking the stable channel gets it when the 0.59 line is released, and until then the change exists for people who follow nightly builds on purpose.

„This prevents unintended process execution during server startup“
luisfelipe-alt, in the pull request summary

Sources

BrandsGemini

Related

OPENCLAW 2026.10.232,768characters, and past that a key could slip thr
Agentsmedium signal

OpenClaw 2026.10.2 stops long tool output from slipping credentials past redaction

A bug in OpenClaw's secret masking let a credential through whenever it straddled the boundary between two slices of a long text. The fix ships in 2026.10.2 along with repaired updates from 2026.10.1 and a new compaction default for OpenAI and Anthropic API keys.

OpenClawverified

The OpenClaw web interface with a conversation on the left and the Bits & Bolts reference MCP app open beside it, showing a 3D keycap model and an Attach view + note button.
Agentsmedium signal

OpenClaw 2026.10.1 lets agents search installed skills and changes how sessions message each other

OpenClaw released version 2026.10.1 on October 10, 2026, its first stable release since 2026.9.7. Agents can now search the skills installed on a machine and read only the one a task needs, instead of carrying the whole library in the prompt. Opt-in MCP Apps run beside chat, and full backups can go to an external drive or Cloudflare R2. Three changes can break existing setups: session-to-session messaging, the old required argument, and Workboard data stored before July 2026.

OpenClawverified

CLAUDE CODE2.1.294hooks written as instructions now block what t
Agentsmedium signal

Claude Code 2.1.294 fixes prompt hooks that let through the actions they were written to block

Anthropic released Claude Code 2.1.294 on October 8, 2026. It fixes prompt and agent hooks phrased as instructions, which could allow exactly the actions they were meant to stop. Version 2.1.295 followed the same day with an onFailure setting that makes a crashed or timed-out command or HTTP hook block the action. Releases 2.1.292 and 2.1.293 from earlier in the week closed several permission and sandbox gaps.

Claude Code Releases (GitHub)verified