The Agent Skills Directory

[SAFE]: The skill provides a legitimate and well-documented workflow for machine learning tasks without any suspicious behavior.\n- [EXTERNAL_DOWNLOADS]: The skill documentation references a high-energy crystal structure dataset hosted on Figshare, which is a well-known and reputable service for scientific data sharing.\n- [PROMPT_INJECTION]: The instructions in SKILL.md and README.md are purely instructional and do not contain any patterns attempting to override agent behavior, bypass safety filters, or extract system prompts.\n- [DATA_EXFILTRATION]: No patterns for accessing sensitive files (like credentials or SSH keys) or exfiltrating data to untrusted domains were found. Data processing is restricted to user-provided scientific datasets.\n- [REMOTE_CODE_EXECUTION]: The skill utilizes local Python scripts and standard command-line tools (e.g., mace_run_train). There are no instances of downloading and executing remote scripts or performing dynamic code evaluation of untrusted input.\n- [COMMAND_EXECUTION]: The skill executes Python scripts and training utilities within a conda environment as expected for a machine learning development workflow.

ml-mace-finetune