/

CVE-2026-91989

Severity:

HIGH RISK

(7.5)

Path Traversal (CWE-22) in Atomic Agents Dashboard

CVE-2026-91989

Deferred

Vulnerability type

Category:

Path Traversal

CWE:

CWE-22

Impact:

Arbitrary File Read

Severity:

HIGH RISK

(7.5)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

A path traversal vulnerability in the atomic-agents-stack dashboard HTTP server allows remote attackers to request files outside the intended agents_root directory. The issue becomes risky when the dashboard is network-accessible and request paths are not canonicalized and constrained before file access.

Affected context

The vulnerability occurs in the dashboard HTTP server's DashboardHandler.do_GET request-processing path in atomic-agents-stack versions before 1.1.0. It affects deployments where remote requests reach the dashboard and path containment checks can be bypassed with directory traversal segments, allowing access beyond the configured agents_root directory.

Exploit preconditions

  • Attack surface: The dashboard HTTP server and its file-serving request handler - Required access: Network access to the dashboard endpoint; authentication

requirements depend on the deployment configuration

  • Trigger point: A GET request whose path contains traversal sequences - Why exploitation is possible: The server's path containment logic does not

reliably canonicalize and constrain the requested path before opening a file

Why this happens

The underlying bug is improper limitation of a pathname to an intended directory, which maps to CWE-22: Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal'). The insecure pattern occurs when user-controlled URL paths are combined with a trusted base directory without robust normalization, resolution, and boundary verification. As a result, path components can cause file access to escape the designated document root.

Potential impact

  • An attacker may read sensitive application files outside the intended dashboard directory.
  • Exposed files could contain credentials, tokens, configuration data, or source code.
  • The vulnerability could support follow-on compromise if obtained secrets enable access to other systems.
  • Confidentiality of the host or service environment could be reduced.

General mitigation guidance

Mitigation generally focuses on enforcing a strict document-root boundary for every requested resource. Common defensive patterns include canonicalizing and resolving paths before authorization, rejecting requests whose resolved paths are outside the intended root, avoiding direct file serving when an allowlisted route model is sufficient, and reducing trust in client-supplied path components. Dashboard services can also be isolated from sensitive host files, run with least-privilege filesystem permissions, and be protected by authentication and network access controls.

How teams detect issues like this

Detection approaches include static analysis for filesystem APIs that combine request-derived paths with configured directories, and code review of URL canonicalization and containment checks. Runtime monitoring can identify traversal markers, repeated access-denied responses, and requests for unexpected paths. Policy enforcement can require allowlisted resource mappings, normalized-path boundary tests, and security-focused regression tests in CI. Automated code review tools may help flag unsafe path construction patterns.

Sources

SourceLink
github.comgithub.com/dep0we/atomic-agents-stack/security/advisories/GHSA-rm43-82j9-r4mj
vulncheck.comvulncheck.com/advisories/atomic-agents-stack-before-1.1.0-path-traversal-via-dashboard-serve-py

Frequently asked

questions

It can be remotely exploitable when an attacker can reach the dashboard HTTP server and the affected request handler is exposed.

Not necessarily. Authentication requirements depend on how the dashboard is deployed and protected, but network reachability and access to the relevant endpoint are required.

Resolve and canonicalize each requested path, verify that it remains within the intended root directory, and prefer an allowlist of explicitly served resources.

Treat request paths as untrusted input, apply centralized path-validation helpers, run the service with least-privilege filesystem access, and include traversal cases in code review and automated security testing.

Find out what's already
exploitable in your codebase.