The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were detected in this skill.
[REMOTE_CODE_EXECUTION]: The skill provides standard installation commands using pip for well-known packages like tensorboard, torch, and tensorflow. No execution from untrusted or unknown remote sources was found.
[DATA_EXFILTRATION]: All network operations are limited to local TensorBoard server hosting (defaulting to localhost) or interactions with established machine learning frameworks. No sensitive data access or external exfiltration patterns were identified.
[INDIRECT_PROMPT_INJECTION]: The skill documents logging of text and metadata (e.g., model predictions logged via writer.add_text). While this represents a data ingestion surface where untrusted model outputs could be stored, it is a standard functionality of the visualization toolkit and does not pose a direct threat in this context.
[OBFUSCATION]: The skill content and instructions are provided in clear, plain-text markdown and standard Python/Bash code. No obfuscation techniques such as Base64 encoding, zero-width characters, or homoglyph attacks were detected.
[DYNAMIC_CONTEXT_INJECTION]: The skill does not utilize dynamic context injection patterns (e.g., the !command syntax) for executing shell commands at load time.

mlops-tensorboard