The Agent Skills Directory

[DATA_EXFILTRATION]: Hardcoded local file paths in SKILL.md (e.g., C:\Users\coper\Documents\GitHub\agent-skills\skills) expose the host's directory structure and local username to the agent context.
[REMOTE_CODE_EXECUTION]: The scripts/unpack_skill.py utility uses zipfile.extractall() without path validation. This makes the system vulnerable to 'Zip Slip' attacks, where a maliciously crafted .skill archive could use relative paths (e.g., ../../) to overwrite arbitrary files on the host system during the unpacking process.
[COMMAND_EXECUTION]: The skill performs destructive file system operations. The scripts/unpack_skill.py script uses shutil.rmtree() to delete existing directories and Path.unlink() to delete the source archive after extraction. If the target paths were manipulated, this could lead to unintended data loss.
[COMMAND_EXECUTION]: The bootstrap process is designed to automatically create multiple AI agents with comprehensive system prompts without requiring user confirmation for the prompt content, which reduces human oversight over the agent's behavioral instructions.

project-bootstrap