CVE-2026-85623: a shared Goose recipe could start commands before the user approved it
Summary
CVE-2026-85623 covers Goose recipes, shareable agent configs that can declare stdio extensions and retry.checks shell commands. Goose ran those commands when a recipe loaded, and its recipe security scan did not look at those fields. The CVE record lists goose up to 1.49.0 and a CVSS 4.0 score of 8.7. Goose's maintainers said the command-line behavior is by design, like running a shell script, but fixed the Desktop app so nothing starts until the user approves the recipe. That fix shipped in Goose 1.52.0 on Sep 23, 2026.
Severity
Timeline
Researcher George Chen opens GitHub issue #10325 after an earlier private report got no response.
A maintainer says the CLI case is not a security boundary violation but there may be a Desktop consent-ordering bug; the reporter reproduces pre-approval execution on Desktop 1.45.0.
CVE-2026-85623 published by VulnCheck.
Fix merged (PR #12094: require recipe consent before extensions start) and the issue is closed.
Goose 1.52.0 released with the fix.
How it happened
When a recipe ran, Goose started its declared extensions, which spawns whatever command the recipe names, before any consent screen. The built-in scan only checked the natural-language fields for hidden Unicode characters, so a recipe with a visible but malicious command passed with no warning. On Desktop, a goose:// deeplink could trigger this before the user clicked anything. The CVE record classifies it as CWE-94. The fix makes Desktop wait for approval before starting extensions. Maintainers consider goose run --recipe on the CLI an explicit choice to run the recipe.
How to check your app
- Update Goose Desktop to 1.52.0 or later.
- Before running a shared recipe, read its extensions (cmd and args) and retry.checks sections.
- Treat goose run --recipe with an untrusted file like running an untrusted shell script.
Sources
- GitHub issue #10325 (aaif-goose/goose)
- Goose 1.52.0 release notes
- Fix PR #12094
- CVE record CVE-2026-85623
- VulnCheck advisory