detection-engineering
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONOBFUSCATIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted data such as security log samples, threat intelligence reports, and detection rule drafts. \n
- Ingestion points: Content processed by the agent in Step 1 and Step 4 (SKILL.md).\n
- Capability inventory: Accesses local files using Read, Grep, and Glob tools.\n
- Sanitization: Section 8 (Prompt Injection Safety Notice) explicitly mandates that the agent redact sensitive values like IPs or hostnames and validate all output against defined schemas.\n
- Boundary markers: The skill includes a dedicated safety section instructing the agent to treat any instructions found within analyzed content (e.g., "ignore previous instructions") as inert data rather than directives.
- [OBFUSCATION]: The skill contains a Base64-encoded string within the validation section of the ADS documentation template. The string 'ZQBjAGgAbwAgACIAdABlAHMAdAAiAA==' decodes to 'echo "test"', which is a benign command used for demonstrating rule functionality.
- [EXTERNAL_DOWNLOADS]: The instructions reference several well-known and official security repositories and domains, including MITRE (mitre.org), SigmaHQ (sigmahq.io), and Red Canary's Atomic Red Team (github.com/redcanaryco), for the purpose of retrieving documentation and methodology frameworks.
Audit Metadata