CVE-2026-101079: agentverus-scanner Security Decision Flaw

CVE-2026-101079 Overview

CVE-2026-101079 affects agentverus-scanner versions up to 0.8.1, an open-source scanner distributed through the AgentVerus project. The flaw resides in the isSecurityDefenseSkill function within dist/scanner/analyzers/context.js. The function relies on untrusted inputs when making a security decision, which maps to [CWE-20] Improper Input Validation. An attacker with local access and low privileges can manipulate these inputs to influence the scanner's security determinations. The exploit technique has been publicly disclosed. The maintainers were notified through a GitHub issue report but have not responded at the time of publication.

Critical Impact
Local attackers can manipulate inputs consumed by isSecurityDefenseSkill to bias the scanner's security defense decisions, undermining the integrity of scan results.

Affected Products

agentverus agentverus-scanner versions up to and including 0.8.1
The dist/scanner/analyzers/context.js module
The isSecurityDefenseSkill analyzer function

Discovery Timeline

2026-09-28 - CVE-2026-101079 published to the National Vulnerability Database (NVD)
2026-09-28 - Last updated in the NVD database

Technical Details for CVE-2026-101079

Vulnerability Analysis

The vulnerability lives inside isSecurityDefenseSkill, a helper used by the AgentVerus scanner's context analyzer. The function determines whether a given skill or component should be treated as a security defense construct. That determination is used elsewhere in the scanner to shape security-relevant decisions, including whether certain items are flagged or trusted.

Because the function consumes attacker-influenced inputs without sufficient validation, a local actor can craft content that causes isSecurityDefenseSkill to return a value that misrepresents the item's true nature. The result is a reliability gap: scanner output no longer accurately reflects the security posture of what was analyzed. The exploitation path requires local access and some user interaction, limiting opportunistic abuse but enabling targeted manipulation on shared workstations and CI runners.

Root Cause

The root cause is improper input validation [CWE-20]. The isSecurityDefenseSkill logic treats inputs as trustworthy when forming a security decision. No authoritative schema check, allowlist, or integrity signal gates the inputs before they influence the branch taken. Classification logic that should be driven by trusted metadata is instead driven by data an attacker can shape.

Attack Vector

The attack originates from a local position with low privileges and requires user interaction, consistent with the published CVSS 4.0 vector. An attacker places or modifies content that the scanner will subsequently analyze on the same host. When a developer or automation runs agentverus-scanner against that content, the manipulated inputs reach isSecurityDefenseSkill and bias its boolean verdict. The practical impact is limited integrity loss on scanner outputs; confidentiality and availability are not affected per the published metrics.

// No verified exploit code is published. Refer to the maintainer's
// repository and the referenced GitHub issue for technical discussion.
// See: https://github.com/agentverus/agentverus-scanner/issues/28

Detection Methods for CVE-2026-101079

Indicators of Compromise

Unexpected modifications to files residing under the scanner's input or working directories on developer and build hosts.
Scanner reports that classify known risky components as security defense skills, or that silently reclassify previously flagged items.
Presence of agentverus-scanner package versions at or below 0.8.1 in project manifests (package.json, lockfiles) and container images.

Detection Strategies

Compare consecutive scan outputs for the same repository and flag diffs where items transition into a trusted security defense classification without a corresponding code change.
Hash and monitor dist/scanner/analyzers/context.js within installed agentverus-scanner packages to detect tampering with the analyzer logic itself.
Review CI pipelines that consume scanner results and gate deployments; alert when results shift from fail to pass without a reviewer-approved change.

Monitoring Recommendations

Collect scanner run logs, exit codes, and output artifacts centrally so classification changes can be audited over time.
Monitor process and file telemetry on developer workstations and CI runners for writes to directories the scanner analyzes immediately before invocation.
Track the installed version of agentverus-scanner across engineering environments and alert on versions <= 0.8.1.

How to Mitigate CVE-2026-101079

Immediate Actions Required

Inventory all projects, developer machines, and CI images that install agentverus-scanner and identify instances at version 0.8.1 or earlier.
Treat scanner output from affected versions as advisory rather than authoritative until a fixed release is available.
Restrict who can write to directories that feed agentverus-scanner on shared systems and CI runners.

Patch Information

No fixed version has been published at the time of writing. The maintainers were notified through GitHub Issue #28 but have not responded. Track the AgentVerus Scanner Repository and the VulDB CVE-2026-101079 Entry for a patched release.

Workarounds

Pin builds to a verified copy of the scanner and review dist/scanner/analyzers/context.js for unexpected changes before each run.
Run agentverus-scanner inside an isolated, ephemeral environment (container or sandbox) so untrusted inputs cannot persist between invocations.
Require a secondary review step for scanner-driven security decisions until a maintainer-issued fix is confirmed.

bash

# Identify installed agentverus-scanner versions across a workspace
npm ls agentverus-scanner --all 2>/dev/null | grep agentverus-scanner

# Fail CI early if a vulnerable version is present
node -e "const v=require('agentverus-scanner/package.json').version; \
  const s=v.split('.').map(Number); \
  if (s[0]===0 && (s[1]<8 || (s[1]===8 && s[2]<=1))) { \
    console.error('Vulnerable agentverus-scanner version:', v); process.exit(1); \
  }"