sdk-security-audit
Warn
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: MEDIUMPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [METADATA_POISONING]: The skill metadata contains deceptive authorship information; 'SKILL.md' identifies 'OpenAI' as the author, which conflicts with the identified author context 'ascend'. This misrepresentation could lead to misplaced trust in the skill's safety or performance.
- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted code from external repositories (via 'git clone'). Maliciously crafted files could contain instructions intended to override the agent's behavior, and the skill lacks explicit boundary delimiters to isolate data from instructions.
- [REMOTE_CODE_EXECUTION]: The skill facilitates the cloning of remote repositories from well-known platforms like GitHub and GitLab. While these operations are necessary for the skill's purpose and use reputable sources, they involve the automated downloading of external content and the execution of command-line tools for fetching and cleaning up data.
Audit Metadata