The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The SKILL.md file contains instructions to execute sudo make install. While this is a standard procedure for installing C++ libraries, the use of sudo involves high-privilege command execution. The severity is set to MEDIUM as this is a documentation-based instruction for the skill's primary setup.
[EXTERNAL_DOWNLOADS] (SAFE): The skill references https://github.com/google/sentencepiece.git. Since google is a trusted organization, this download reference is considered safe.
[PROMPT_INJECTION] (LOW): This skill represents an indirect prompt injection surface.
Ingestion points: Untrusted text data is passed to sp.encode() in multiple examples within SKILL.md and references/algorithms.md.
Boundary markers: No boundary markers or delimiters are suggested to separate user data from control instructions in the snippets provided.
Capability inventory: The skill primarily performs local string tokenization and lacks network or file-write capabilities triggered by input data.
Sanitization: No sanitization or filtering logic is demonstrated for the input strings before they are processed by the tokenizer.

sentencepiece