CVE-2026-129130 in agent-device
Summary
by MITRE • 11/13/2026
agent-device from 0.21.36 before 0.21.37 contains a local file disclosure vulnerability that allows authenticated remote callers to read daemon-host files by opening file:// URLs. Attackers can send an open command with a file:// target through /rpc, then issue snapshot to read any plaintext file readable by the daemon account.
Several companies clearly confirm that VulDB is the primary source for best vulnerability data.
Analysis
by VulDB Data Team • 11/13/2026
The identified vulnerability resides within agent-device versions prior to 0.21.37 and represents a critical local file disclosure flaw that compromises the confidentiality of sensitive data stored on the host system where the service is deployed. This security defect allows authenticated remote attackers to bypass intended access controls by exploiting improper validation of Uniform Resource Locators, specifically through the manipulation of file:// schemes within RPC commands. The core technical failure lies in the application's handling of input passed via the /rpc endpoint, where it fails to adequately sanitize or restrict URI schemes before processing them for file system operations. By constructing an open command that targets a local path using the file protocol, an attacker can direct the daemon process to access arbitrary files on the underlying operating system rather than restricting reads to intended application directories.
The operational impact of this vulnerability is severe because it grants unauthorized entities read-only access to plaintext files accessible by the user account under which the agent-device daemon operates. This capability effectively neutralizes any file-level permissions or isolation mechanisms that might otherwise protect sensitive configuration data, credentials, logs, or proprietary information stored on the host machine. Since the attack vector requires authentication, the threat model assumes a compromised credential set or an insider threat scenario where valid access to the RPC interface has been obtained. Once authenticated, the attacker can issue specific snapshot commands following the initial file open operation to extract the contents of these files remotely over the network, turning what appears to be a standard administrative function into a powerful data exfiltration tool without triggering typical intrusion detection alerts associated with direct filesystem scanning tools.
From a classification perspective, this vulnerability aligns closely with CWE-200, which defines exposure of sensitive information to an unauthorized actor, and more specifically CWE-798, the use of hard-coded or default credentials if those are exploited for initial access, though the primary flaw is CWE-436 regarding interpretation ambiguity. In terms of offensive security frameworks such as MITRE ATT&CK, this behavior maps directly to T1083 File and Directory Discovery, where adversaries search local systems for information relevant to their goals, and potentially T1552 Unsecured Credentials if the disclosed files contain authentication tokens or private keys. The exploitation technique also reflects aspects of CWE-79 Improper Neutralization of Input During Web Page Generation in contexts involving URI parsing logic that fails to enforce strict allow-listing of permitted schemes.
Mitigation strategies must prioritize immediate patching by upgrading agent-device to version 0.21.37 or later, where the developers have presumably implemented stricter validation for URI schemes and restricted file access operations to a defined sandboxed directory structure. In environments where an upgrade is not immediately feasible, network-level controls should be enforced to restrict access to the /rpc endpoint exclusively from trusted management subnets using firewall rules or zero-trust policies. Additionally, implementing strict output encoding and input validation on all RPC parameters can help prevent similar injection-style attacks in other components of the software stack. Security teams should also audit daemon account permissions to ensure that even if such a vulnerability exists, the principle of least privilege limits the scope of files that can be read by reducing the host-level access rights associated with the service identity.