Safeguard
Vulnerability Analysis•

CVE-2026-26057: Cisco Skill Scanner API Server Exposure

Skill Scanner 1.0.1 and earlier binds its API server to multiple interfaces. Medium severity, 6.5. Upgrade to 1.0.2 or leave the API server off.

Safeguard Security Team
3 min read

CVE-2026-26057 is a medium severity flaw (CVSS 6.5) in Cisco's Skill Scanner, the scanner for AI agent skills. When its API Server is enabled it binds to multiple interfaces, letting callers cause denial of service or upload files. Upgrade to 1.0.2. The record does not list it in CISA's KEV catalog, and the sources we read do not report exploitation.

What's affected

  • Skill Scanner (pip package cisco-ai-skill-scanner) 1.0.1 and earlier, with the API Server enabled
  • Fixed in 1.0.2 and later

The API Server is not enabled by default. The advisory lists no workaround other than upgrading.

How to check if you're exposed

Ask pip:

pip show cisco-ai-skill-scanner

Compare the version with 1.0.2. Then check whether anyone runs the API Server, which is the condition for this issue, and which network interfaces it listens on. A listener on every interface is reachable from other machines on the network.

Safeguard has a free in-browser dependency scan at https://gold.safeguard.sh/scan that reads a manifest or lockfile and flags known vulnerable packages, with no repository access needed.

What to do now

  1. Upgrade cisco-ai-skill-scanner to 1.0.2 or later.
  2. If you run the API Server, bind it to the loopback address or an internal interface only.
  3. Put any shared instance behind authentication at a reverse proxy.
  4. Review the machine for unexpected files if it was reachable from untrusted networks.
  5. Leave the API Server off if you only use the command line.

Why it matters

CWE-668 is exposure of a resource to the wrong sphere. The service was meant for a narrow audience, and binding to multiple interfaces made it reachable by a wider one. The advisory says an attacker could then exhaust memory or upload files to arbitrary folders.

There is a small irony in a security scanner shipping with an exposed listener, and it is a useful reminder that developer tools get the same scrutiny as production code. Anything that listens on a network port needs a deliberate choice about who can reach it.

The medium score reflects low confidentiality and integrity impact. The reach is network wide, but only when the optional server is turned on.

Quick check

Am I affected if I only use the command line?

The advisory ties the issue to the API Server being enabled, which is off by default.

What is the fix?

Upgrade to 1.0.2 or later.

Sources

Last reviewed: 19 October 2026.

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.