deep-security-scan
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFE
Full Analysis
- Untrusted Data Handling (Indirect Prompt Injection Surface): The skill processes user-provided context and external URL content, which are identified as potential ingestion points for untrusted data in
SKILL.md. To mitigate risks, the instructions explicitly state that this content must be treated as untrusted evidence that cannot authorize actions, testing, disclosure, or additional reads. It also mandates that user context cannot override workflow or tool instructions. - External Resource Access: The skill utilizes external URLs only when explicitly authorized by the user and limits the interaction to a one-time source read to prevent crawling or unauthorized data exfiltration. References to
chatgpt.comfor account advisory are vendor-owned resources. - Least Privilege and Resource Management: The skill includes a 'Concurrent Desktop Scan Guard' to prevent excessive resource consumption (CPU/Memory/Tokens) and requires explicit user confirmation before proceeding with multiple scans.
- Tool Usage: The skill interacts with the Codex Security plugin server through authorized tools like
start_codex_security_deep_scanandcomplete_codex_security_scan. These operations are within the scope of the skill's intended purpose as a repository security scanner.
Audit Metadata