The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard system utilities like bsdtar for archiving, cp for staging files while preserving timestamps, and grep to scan for sensitive information.
[SAFE]: A specialized Python script scripts/scrub_metadata.py is included to strip identifying metadata (Author, Producer, OS strings) from PDF, OOXML (DOCX/XLSX/PPTX), and PNG files, significantly reducing the risk of unintentional data leakage.
[SAFE]: The instructions mandate a pre-archive scan of all deliverables for API keys, tokens, .env files, and local home directory paths, ensuring the user is notified of potential credential exposure before packaging.
[INDIRECT_PROMPT_INJECTION]: While the skill processes project code and data which could contain malicious instructions, it implements strong mitigation strategies including mandatory human review of automated scan results ("Report every hit... and let the user decide") and programmatic metadata sanitization.

replication-package