/

CVE-2026-112015

Severity:

HIGH RISK

(7.5)

Path Traversal (CWE-22) in Omnigent Agent Bundles

CVE-2026-112015

Deferred

Vulnerability type

Category:

Path Traversal

CWE:

CWE-22

Impact:

Unauthorized Host File Read and Write

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

Omnigent contains a path traversal vulnerability in nested sub-agent working directory handling. An authenticated user may upload a bundle that causes a sub-agent to access files outside the intended workspace, potentially reading or modifying host files as the runner user.

Omnigent contains a path traversal vulnerability in nested sub-agent working directory handling. An authenticated user may upload a bundle that causes a sub-agent to access files outside the intended workspace, potentially reading or modifying host files as the runner user.

Affected context

The issue affects Omnigent deployments that execute uploaded bundles containing nested sub-agents and accept os_env.cwd values for those agents. Risk arises when validation restricts only the top-level working directory while nested sub-agent paths can use absolute locations or parent-directory traversal. The practical impact depends on the runner user's filesystem permissions and the files exposed to the execution environment.

The issue affects Omnigent deployments that execute uploaded bundles containing nested sub-agents and accept os_env.cwd values for those agents. Risk arises when validation restricts only the top-level working directory while nested sub-agent paths can use absolute locations or parent-directory traversal. The practical impact depends on the runner user's filesystem permissions and the files exposed to the execution environment.

Exploit preconditions

  • Attack surface: Bundle upload and nested sub-agent execution
  • Required access: An authenticated account permitted to upload or run bundles
  • Trigger point: Processing the os_env.cwd value for a nested sub-agent
  • Why exploitation is possible: Validation is applied to the root bundle context but does not consistently constrain nested working-directory values, allowing paths to escape the intended workspace
  • Attack surface: Bundle upload and nested sub-agent execution
  • Required access: An authenticated account permitted to upload or run bundles
  • Trigger point: Processing the os_env.cwd value for a nested sub-agent
  • Why exploitation is possible: Validation is applied to the root bundle context but does not consistently constrain nested working-directory values, allowing paths to escape the intended workspace

Why this happens

The vulnerability results from incomplete path validation across nested execution contexts. A working-directory value supplied by a bundle is trusted after validation has been applied only to the root context, allowing absolute paths or parent-directory traversal to bypass the workspace boundary. This maps to CWE-22, Improper Limitation of a Pathname to a Restricted Directory, because attacker-influenced paths are not consistently canonicalized and checked against an approved base directory.

The vulnerability results from incomplete path validation across nested execution contexts. A working-directory value supplied by a bundle is trusted after validation has been applied only to the root context, allowing absolute paths or parent-directory traversal to bypass the workspace boundary. This maps to CWE-22, Improper Limitation of a Pathname to a Restricted Directory, because attacker-influenced paths are not consistently canonicalized and checked against an approved base directory.

Potential impact

  • An attacker may read sensitive host or application files
  • An attacker may modify or overwrite files accessible to the runner user
  • Workspace isolation may be bypassed
  • Credentials, configuration data, or source code could be exposed
  • The runner environment could be altered to support further compromise
  • An attacker may read sensitive host or application files
  • An attacker may modify or overwrite files accessible to the runner user
  • Workspace isolation may be bypassed
  • Credentials, configuration data, or source code could be exposed
  • The runner environment could be altered to support further compromise

General mitigation guidance

Mitigation generally focuses on enforcing workspace boundaries for every

execution context, including nested sub-agents.

Common defensive patterns include:

  • Canonicalizing candidate paths before authorization checks
  • Rejecting absolute paths and parent-directory traversal where they are not required
  • Resolving paths against a fixed, trusted workspace and verifying the result remains beneath it
  • Applying identical validation rules recursively to nested agents
  • Running bundles with least-privilege identities and isolated filesystem access
  • Restricting writable locations and avoiding access to sensitive host paths
  • Treating uploaded bundle metadata as untrusted input by default

Mitigation generally focuses on enforcing workspace boundaries for every

execution context, including nested sub-agents.

Common defensive patterns include:

  • Canonicalizing candidate paths before authorization checks
  • Rejecting absolute paths and parent-directory traversal where they are not required
  • Resolving paths against a fixed, trusted workspace and verifying the result remains beneath it
  • Applying identical validation rules recursively to nested agents
  • Running bundles with least-privilege identities and isolated filesystem access
  • Restricting writable locations and avoiding access to sensitive host paths
  • Treating uploaded bundle metadata as untrusted input by default

How teams detect issues like this

Organizations can look for:

  • Static analysis rules for unchecked joins, normalization gaps, and direct use of user-controlled working directories
  • Code review signals showing validation only at the root bundle or parent-agent level
  • Tests covering absolute paths, parent-directory traversal, symlink behavior, and nested execution contexts
  • Runtime policy alerts for bundle processes accessing paths outside approved workspaces
  • Filesystem monitoring for unexpected reads or writes by runner identities
  • Automated security review tools that identify path confinement and authorization inconsistencies

Organizations can look for:

  • Static analysis rules for unchecked joins, normalization gaps, and direct use of user-controlled working directories
  • Code review signals showing validation only at the root bundle or parent-agent level
  • Tests covering absolute paths, parent-directory traversal, symlink behavior, and nested execution contexts
  • Runtime policy alerts for bundle processes accessing paths outside approved workspaces
  • Filesystem monitoring for unexpected reads or writes by runner identities
  • Automated security review tools that identify path confinement and authorization inconsistencies

Sources

SourceLink
github.comgithub.com/omnigent-ai/omnigent/blob/v0.13.0/omnigent/server/bundles.py#L138-L155
github.comgithub.com/omnigent-ai/omnigent/commit/4a406d5b3b3f3a7c3401eab659ab04e728923b94
github.comgithub.com/omnigent-ai/omnigent/releases/tag/v0.14.0
github.comgithub.com/omnigent-ai/omnigent/security/advisories/GHSA-p5x3-3gqh-x44g
vulncheck.comvulncheck.com/advisories/omnigent-0.3.0-before-0.14.0-path-traversal-via-sub-agent-os-env-cwd

Frequently asked

questions

It can be exploitable through the application's bundle-upload or bundle-execution

interface when that interface is reachable by an authenticated user. Exploitability

depends on the user's permissions and the runner's filesystem access.

The attacker generally needs authenticated access that permits uploading or

executing bundles containing nested sub-agents. The resulting impact is limited

by the runner user's operating-system permissions.

Resolve and canonicalize every requested working directory, then verify that it

remains within a trusted workspace before execution. Apply the same checks to

nested agents and treat uploaded metadata as untrusted.

Use centralized path-confinement logic, test nested and adversarial path cases,

enforce least privilege, and monitor runtime filesystem access for violations.

It can be exploitable through the application's bundle-upload or bundle-execution

interface when that interface is reachable by an authenticated user. Exploitability

depends on the user's permissions and the runner's filesystem access.

The attacker generally needs authenticated access that permits uploading or

executing bundles containing nested sub-agents. The resulting impact is limited

by the runner user's operating-system permissions.

Resolve and canonicalize every requested working directory, then verify that it

remains within a trusted workspace before execution. Apply the same checks to

nested agents and treat uploaded metadata as untrusted.

Use centralized path-confinement logic, test nested and adversarial path cases,

enforce least privilege, and monitor runtime filesystem access for violations.

Find out what's already
exploitable in your codebase.