Review Skill Scan Results
Understand how to read AI Skill Security scan results, including overall verdicts,
rule outcomes, and vulnerability findings.
| Where Can I Use This? | What Do I Need? |
|---|
- Prisma AIRS (AI Skill Security)
|
- No License required for preview
|
A scan produces one of two overall verdicts:
- Allowed—The skill passed all configured blocking security and governance
rules and is recommended for use based on the current policy configuration.
- Blocked—One or more configured blocking rules identified findings that you
should review before trusting or using the skill.
Rule Outcomes
The Rule Outcomes tab provides a detailed breakdown of how
each security and governance rule evaluated the uploaded skill. For each rule, you
can view:
- Rule Status—Indicates whether the rule passed or generated one or more
findings.
- Matching Vulnerabilities—Displays the number of vulnerabilities
associated with the selected rule.
- Rule Details—Provides a description of what the rule detects and the
types of behaviors it is designed to identify.
- Related Vulnerabilities—Lists the findings associated with the selected
rule, including the affected file or component.
- Recommendations—Provides guidance for reviewing and remediating the
identified behavior before trusting or deploying the skill.
Vulnerabilities
The Vulnerabilities tab lists every finding identified during
the scan. Selecting a vulnerability displays detailed information. For each
vulnerability, you can view:
- Finding Summary—A high-level overview describing the detected behavior
and why it was identified.
- Vulnerability Type—The category of behavior detected, such as data
exfiltration or silent operation.
- Description—Additional context explaining the behavior and the
potential security risk.
- Affected Component—The file or resource within the uploaded skill
package where the finding was detected.
- Supporting Details—Additional information to help you understand the
finding and determine whether further investigation or remediation is
needed.
