CVE-2026-83816

CodeWhale: exec_shell parallel-readonly classifier auto-approves fd/rg/git grep exec flags enabling silent RCE

Published Nov 16, 2026 CVSS 3.1 · 8.1 High EPSS n/a KEV · Not listed
Description
CodeWhale is a DeepSeek and MiMo coding agent for the terminal. Prior to 0.9.0, the ExecShellTool::approval_requirement_for path used is_parallel_readonly_command to downgrade commands to automatic approval based on a flag-stripped command allowlist. The allowlist treated fd, rg, and git grep as read-only even when fd -x or fd --exec, rg --pre, or git grep --open-files-in-pager selected an external program. registered_tool_approval_required returned without displaying a prompt for that automatic classification, and the later command-safety check also treated the base utilities as safe. Attacker-controlled content that influenced an agent tool call could therefore cause arbitrary program execution on the developer host without the approval promised by the default interactive profile, resulting in unauthorized access, modification, or disruption. This issue is fixed in version 0.9.0.

Analysis

AI-generated

Technical impact

The vulnerability in CodeWhale allows an attacker to cause arbitrary program execution on a developer’s machine by tricking the agent into auto‑approving commands that contain executable arguments. The flaw arises because the parallel‑readonly command classifier downgrades commands such as fd -x, rg --pre, and git grep --open-files-in-pager to read‑only status, and the later safety check treats these utilities as inherently safe. As a result, agent calls that include attacker‑controlled input can bypass the interactive prompt and run any program on the host, leading to unauthorized access, modification, or system disruption. The entry in the CWE list, CWE‑78, identifies this as an OS command injection in a shell environment, while CWE‑863 indicates a lack of enforcement of a required approval policy.

Affected Systems

The affected software is CodeWhale, a terminal‑based coding agent manufactured by codewhale‑hq, version 0.8.x and earlier. Versions prior to 0.9.0 exhibit the flaw because the ExecShellTool approval requirement uses the is_parallel_readonly_command logic and the tool does not prompt for approval when the classifier flags a command as read‑only. CodeWhale 0.9.0 and later contain the fix that restores the interactive approval requirement and correctly evaluates the safety of base utilities.

Risk and Exploitability

The CVSS score for this vulnerability is 8.1, indicating a high‑severity exposure. EPSS is not available, but the lack of a related KEV listing and the local nature of the attack mean that exploitation would most likely occur through malicious content crafted by a user interacting directly with the tool. An attacker could gain full control of the host by executing arbitrary commands in the context of the user running CodeWhale. The fix in version 0.9.0 eliminates the bypass by enforcing the approval prompt and proper command validation, reducing the risk to zero for compliant deployments.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Upgrade to CodeWhale version 0.9.0 or later to apply the security fix that restores the approval prompt and correct command checks.
  • Validate that the ExecShellTool approval requirements enforce prompting; disable automatic approval for read‑only commands if they are not required.
  • Review any scripts or templates that supply agent tool calls to ensure they do not include untrusted input that could be interpreted as an executable flag.

Generated by OpenCVE AI on November 16, 2026 at 20:22 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories available
History
Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'no', 'Exploitation': 'none', 'Technical Impact': 'total'}, 'version': '2.0.3'}

Type Values Removed Values Added
Description CodeWhale is a DeepSeek and MiMo coding agent for the terminal. Prior to 0.9.0, the ExecShellTool::approval_requirement_for path used is_parallel_readonly_command to downgrade commands to automatic approval based on a flag-stripped command allowlist. The allowlist treated fd, rg, and git grep as read-only even when fd -x or fd --exec, rg --pre, or git grep --open-files-in-pager selected an external program. registered_tool_approval_required returned without displaying a prompt for that automatic classification, and the later command-safety check also treated the base utilities as safe. Attacker-controlled content that influenced an agent tool call could therefore cause arbitrary program execution on the developer host without the approval promised by the default interactive profile, resulting in unauthorized access, modification, or disruption. This issue is fixed in version 0.9.0.
Title CodeWhale: exec_shell parallel-readonly classifier auto-approves fd/rg/git grep exec flags enabling silent RCE
Weaknesses CWE-78
CWE-863
References
Metrics cvssV3_1

{'score': 8.1, 'vector': 'CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H'}

Subscriptions

No data.

Weaknesses
  • CWE-78

    Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection')

  • CWE-863

    Incorrect Authorization

Sources

Status
PUBLISHED
Assigner
GitHub_M
Published
Updated
Nov 16, 2026, 9:21 PM
Reserved
Aug 31, 2026, 11:04 PM
Status
Received
Published
Nov 16, 2026, 7:17 PM
Modified
Nov 16, 2026, 10:18 PM