/

CVE-2026-55540

Severity:

HIGH RISK

(7.1)

Path Traversal (CWE-22) in PraisonAI Workspace Tools

CVE-2026-55540

Deferred

Vulnerability type

Category:

Path Traversal

CWE:

CWE-22

Impact:

Unauthorized File Access

Severity:

HIGH RISK

(7.1)

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

PraisonAI may incorrectly enforce workspace boundaries because it resolves absolute paths without resolving symbolic links. If a symbolic link inside the configured workspace points elsewhere, an attacker may use file and code tools to access files outside the intended directory.

Affected context

The issue occurs in PraisonAI deployments that use workspace-restricted file or code tools and rely on is_path_within_directory() to enforce those restrictions. The risk is present when a symbolic link within the configured workspace can be created or controlled and points to a location outside that workspace.

Exploit preconditions

  • Attack surface: Workspace file and code tools that accept filesystem paths
  • Required access: The ability to create, modify, or otherwise control a symbolic link within the configured workspace, along with access to an affected tool
  • Trigger point: A tool operation checks whether a requested path remains inside the workspace
  • Why exploitation is possible: The boundary check resolves lexical or absolute paths but does not fully resolve symbolic links before comparing locations

Why this happens

The underlying bug is an incomplete filesystem boundary check. Using os.path.abspath() normalizes a path without resolving symbolic links, so a path that appears to remain under the workspace can ultimately refer to a different directory. This behavior maps to CWE-22, Path Traversal, because attacker-influenced path resolution can bypass an intended directory restriction.

Potential impact

  • An attacker may read files outside the configured workspace.
  • Sensitive credentials, configuration data, or source code could be exposed.
  • File and code tools could potentially access unintended resources.
  • Compromise of accessible data could support further attacks against connected services.

General mitigation guidance

Mitigation generally focuses on enforcing filesystem boundaries after complete

path resolution rather than relying on lexical normalization alone.

Common defensive patterns include:

  • Resolve symbolic links and canonicalize both the requested path and the permitted workspace before comparison.
  • Reject paths whose canonical targets fall outside the configured directory.
  • Avoid following symbolic links in workspace-controlled content where the workflow permits.
  • Use allowlisted directories and narrowly scoped tool capabilities.
  • Run file and code tools with least-privilege identities and filesystem access.
  • Treat workspace contents and path metadata as potentially attacker-controlled.

How teams detect issues like this

Organizations can identify similar weaknesses through:

  • Static analysis checks for filesystem boundary checks based only on absolute or normalized paths.
  • Code review searches for path validation that occurs before symbolic-link resolution.
  • Tests that exercise links, nested links, relative paths, alternate separators, and boundary-prefix cases.
  • Runtime monitoring for tool operations accessing canonical paths outside approved workspace roots.
  • Policy enforcement requiring canonical-path comparisons and least-privilege filesystem permissions.

Automated code review and security scanning tools may help flag incomplete path validation patterns.

Sources

SourceLink
github.comgithub.com/MervinPraison/PraisonAI/commit/2f9677abb2ea68eab864ee8b6a828fd0141612e1
github.comgithub.com/MervinPraison/PraisonAI/releases/tag/v4.6.58
github.comgithub.com/MervinPraison/PraisonAI/security/advisories/GHSA-ch89-h4r2-c8f8

Frequently asked

questions

Exploitability depends on how PraisonAI is exposed. It may be remotely reachable when an attacker can invoke the affected tools and can influence workspace contents or symbolic links.

The attacker generally needs tool access plus the ability to create or control a symbolic link inside the configured workspace, or another way to influence the checked path.

Resolve symbolic links and canonicalize paths before comparing them with a canonical, explicitly permitted workspace root. Reject any resolved path outside that root.

Combine canonical-path validation with controlled workspace permissions, restricted tool capabilities, symlink-aware tests, and least-privilege runtime identities.

Find out what's already
exploitable in your codebase.