speculative-decoding
Fail
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to download source code from external, non-trusted GitHub repositories. Evidence: git clone https://github.com/FasterDecoding/Medusa and git clone https://github.com/hao-ai-lab/LookaheadDecoding in SKILL.md.
- [REMOTE_CODE_EXECUTION]: The skill provides commands to install and execute code from third-party repositories. Evidence: pip install -e . after cloning external repositories in SKILL.md.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user prompts which are interpolated into model generation calls.
- Ingestion points: Prompt variables in SKILL.md.
- Boundary markers: Absent.
- Capability inventory: Package installation (pip), file system access (git clone), and network operations for model loading.
- Sanitization: No input sanitization is present.
Recommendations
- AI detected serious security threats
Audit Metadata