CVE-2026-55437: Coder Agent Log XSS Vulnerability – Patch Guide & Detection

Technical summary

The vulnerability exists in the `AgentLogLine` dashboard component, which processes ANSI escape sequences using the `ansi-to-html` library without setting the `escapeXML: true` parameter. The resulting HTML is inserted into the DOM via React's `dangerouslySetInnerHTML` property. Because server-side sanitization does not neutralize HTML metacharacters, any HTML tags or JavaScript in log output bypass client-side protections. The attack vector is stored XSS: malicious content persists in agent logs until viewed. Versions 2.29.17, 2.32.7, 2.33.8, and 2.34.2 resolve this by enabling XML escaping in the `ansi-to-html` call, ensuring HTML metacharacters are escaped before DOM insertion.

Business impact

An authenticated attacker with the ability to manipulate workspace agent output can execute arbitrary JavaScript in the browsers of administrators or developers who view those logs. This may result in session hijacking, credential theft, lateral movement to other workspaces, or defacement of the dashboard interface. The attack surface is limited to users with visibility into agent logs, but in typical Coder deployments, multiple team members may monitor logs for troubleshooting. The impact is heightened if logs contain sensitive information that attackers can exfiltrate via injected scripts.

Affected systems

Coder versions prior to 2.29.17, 2.32.7, 2.33.8, and 2.34.2 are affected. Organizations must identify which version is deployed and verify the patch level. Consult the Coder release notes or your instance's admin panel to confirm the running version. The vulnerability only manifests if the dashboard is actively used to view agent logs; headless or CLI-only deployments have reduced exposure but should still patch.

Exploitability

Exploitation is feasible but requires preconditions. The attacker must either control a workspace agent (through a compromised agent token or access to the agent runtime) or have the ability to inject content into agent log streams. Once payload is in place, exploitation is trivial—simply waiting for a user to view the logs in the dashboard is sufficient. No complex user interaction beyond normal log viewing is required. The CVSS score of 5.4 (MEDIUM) reflects these requirements: network-accessible but requires prior authentication and user interaction to trigger the XSS payload.

Remediation

Upgrade to version 2.29.17, 2.32.7, 2.33.8, or 2.34.2 or later, depending on your current release line. No workarounds are documented; patching is the only mitigation. Organizations unable to patch immediately should restrict access to the agent logs dashboard to trusted administrators and monitor for suspicious log content or unusual script execution in dashboards.

Patch guidance

Coder publishes updates in parallel release lines. Identify your current version (Settings > About or `coder version` command). If running 2.29.x, upgrade to 2.29.17 or later. If running 2.32.x, upgrade to 2.32.7 or later. If running 2.33.x, upgrade to 2.33.8 or later. If running 2.34.x, upgrade to 2.34.2 or later. For versions newer than 2.34.2, verify you are on the latest patch release. Test in a non-production environment first. Plan for brief downtime during the upgrade; Coder deployments typically restart cleanly.

Detection guidance

Monitor dashboard access logs for unusual patterns of log viewing, especially during off-hours. Audit agent log content for HTML tags, script elements, or ANSI sequences that appear anomalous. Review workspace agent token usage and revoke any tokens not associated with known deployments. In the dashboard, look for log lines containing characters like `<`, `>`, or `&` in unexpected contexts—these may indicate escape failures. Correlate log viewing events with subsequent suspicious network requests from user browsers (e.g., beacon calls to attacker infrastructure).

Why prioritize this

This vulnerability merits prompt patching because it affects a core dashboard feature and requires no special exploit code. However, the MEDIUM severity and the requirement for prior agent control or log injection limit the immediate risk in most environments. Organizations with strict access controls on agent configuration and log sources can deprioritize slightly, but public-facing Coder instances or those in multi-tenant environments should patch within 2 weeks. The lack of a KEV listing indicates this is not yet widely exploited in the wild.

Risk score, explained

The CVSS 3.1 score of 5.4 (MEDIUM, AV:N/AC:L/PR:L/UI:R/S:C/C:L/I:L/A:N) reflects: network accessibility (AV:N), low attack complexity (AC:L), requirement for prior authentication (PR:L), user interaction needed to view logs (UI:R), changed scope due to cross-context JavaScript execution (S:C), and limited confidentiality and integrity impact (C:L/I:L). Availability is not impacted. The score correctly places this below CRITICAL/HIGH but above LOW—a real risk that should not be ignored.

Frequently asked questions