The Agent Skills Directory

PROMPT_INJECTION (SAFE): The content is purely educational and technical documentation. There are no instructions that attempt to override system behavior or bypass safety filters.\n- DATA_EXFILTRATION (SAFE): No access to sensitive files, hardcoded credentials, or unauthorized network communication patterns were identified. The documentation refers to well-known academic and open-source domains.\n- EXTERNAL_DOWNLOADS (SAFE): While the documentation references GitHub repositories (e.g., hao-ai-lab/LookaheadDecoding, FasterDecoding/Medusa), these are cited for informational purposes. The skill does not perform automated downloads or executions.\n- REMOTE_CODE_EXECUTION (SAFE): The provided Python and Bash snippets are illustrative examples for users to understand the implementation of inference algorithms and do not facilitate remote code execution by the agent.\n- NO_CODE (SAFE): The skill consists of markdown files for reference and does not include any scripts or automation that would be executed by an AI agent in a functional capacity.

speculative-decoding