The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains no instructions that attempt to override agent behavior, bypass safety guidelines, or extract system prompts. The language is purely instructional and focused on model inference procedures.
[DATA_EXFILTRATION]: No sensitive file paths, hardcoded credentials, or unauthorized network operations were detected. The skill references local paths relative to the package root and documentation files within the repository.
[REMOTE_CODE_EXECUTION]: No remote script downloads or piped shell executions from external sources were found. The skill describes running a local Python module (cosmos_framework.scripts.inference) using the standard uv package manager.
[COMMAND_EXECUTION]: Shell commands are restricted to local environment tools for model inference and help documentation. The instructions emphasize safety features like guardrails and reproducible seeds.
[OBFUSCATION]: No encoded content, hidden characters, or homoglyph substitutions were detected in the instructions or metadata.
[SAFE]: The skill operates within the expected scope of a documentation guide for the Cosmos3 framework, providing clear mappings from user queries to relevant project files.

cosmos3-inference