skill-creator
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSMETADATA_POISONING
Full Analysis
- [COMMAND_EXECUTION]: The skill contains several Python scripts (
run_eval.py,improve_description.py,run_loop.py) that usesubprocess.run()andsubprocess.Popen()to execute theclaudeCLI andlsof. This is the intended primary purpose of the skill to facilitate testing and optimization within the development environment. - [EXTERNAL_DOWNLOADS]: The skill's
eval-viewer/viewer.htmlreferences a well-known service (SheetJS viacdn.sheetjs.com) to enable spreadsheet rendering in the generated HTML reports. This is documented neutrally as a functional dependency for data visualization. - [METADATA_POISONING]: The skill uses a 'pushy' description pattern to ensure the agent invokes the skill reliably. While this is an intentional design pattern for skill discovery, it is documented as a behavioral characteristic.
Audit Metadata