Fix repository config pre-approves a broad tool permission in Gemini CLI

A repository-local config grants an unbounded tool permission - `Bash`, `WebFetch`, `Write(*)`, `mcp__*` - or switches a permission gate off entirely. The approval prompt exists to bound the blast radius of a model doing something unexpected; a repository should not be the thing that answers it.

medium likely Gemini CLI CWE-732 / ASI ASI03

The vulnerable pattern in Gemini CLI

This finding comes from the Gemini CLI fixture in the owlwarden test suite. This entry pre-approves the tool for anything, so the approval prompt never appears for it again in this project. This file also declares a command that runs when the workspace opens, so the two together are an execution with the approval already granted.

The corrected configuration

Remove the auto-approve entry from `.gemini/settings.json`. If a command is run often enough to be tedious, add that exact command rather than the tool that runs it.

// .gemini/settings.json
"autoApprove": ["pnpm test"]

If you are not using Gemini CLI

Replace the wildcard with the specific commands the project actually needs, and leave everything else to the approval prompt. A list of five exact commands is more useful to the team than one wildcard, and it is reviewable.

Check your own repository

npx owlwarden scan
npx owlwarden explain agent-permission-wildcard

Runs on your machine. No account, no telemetry, no network unless you ask. In CI, SARIF uploads to code scanning and the exit code is the gate.

Other Gemini CLI checks

Rules with a tested Gemini CLI example.

• agent-config-env-redirect high Repository config redirects the agent's API traffic
• agent-hook-autoexec high Repository config executes a command when the workspace is opened
• agent-hook-untrusted-command high Hook command reaches outside the project

agent-permission-wildcard for every agent host / All rules / owlwarden