Goose CVE-2026-111079

Three concrete conditions must hold: (1) Goose's terminal integration feature is enabled; (2) the feature captures shell command history and forwards it, unsanitized, to the configured LLM provider endpoint during queries; and (3) a user actually enters sensitive data - API tokens, passwords, or secret-bearing environment variables - into the terminal while that history is being captured. Limiting factors that reduce realistic risk: the feature must be turned on (default state not confirmed in the sources), a human must input the secret (UI:R), and the recipient is the configured LLM provider and its logs rather than an arbitrary remote attacker, so impact is bounded to confidentiality exposure toward that provider and any downstream log storage.

This is a confidentiality-only information-disclosure design flaw (CWE-497) in Goose's terminal integration, not an attacker-driven exploit. The supplied CVSS:3.1 vector (AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:N/A:N = 7.4) reflects that unredacted shell history crosses a trust boundary to an external LLM provider, with scope change (S:C) and high confidentiality impact (C:H) but no integrity or availability effect. There is no inbound 'attacker' in the classic sense - the 'network' element is data egress to the configured LLM endpoint and its logs, triggered when a user enters secrets (UI:R). EPSS is not provided in the input, the CVE is not on CISA KEV, and no proof-of-concept or exploit reference is present (only the Red Hat advisory and Bugzilla tracker), so there is no public exploit identified at time of analysis. The vuln.today decision label is 'this_week', indicating moderate prioritization rather than emergency action; this aligns with a real-but-not-actively-exploited data-hygiene issue whose realization depends on the feature being enabled and a user actually typing secrets. Conflicting signal to note: the description attributes the flaw to the product 'Goose' while the CPE and tags list Red Hat Enterprise Linux 9 and 10 - verify the affected component and packaging with Red Hat, as this affects both scope and footprint.

With Goose's terminal integration enabled, a developer pastes or types an API token, password, or secret-bearing environment variable into the shell. Goose captures that command history verbatim and includes it in a query to the configured LLM provider, so the credential leaves the organization's trust boundary and is retained in provider-side logs. No public exploit code is required or identified - the disclosure occurs through normal, intended feature use rather than a crafted attack.