new-workspace

Warn

Audited by Gen Agent Trust Hub on Jul 4, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DATA_EXFILTRATION]: The skill reads from ~/.claude/CLAUDE.md (a global configuration file) to extract user identity, locale, and environment facts. This information is then written into a new CLAUDE.md file within the provisioned workspace. Because the skill defaults to creating public GitHub repositories and auto-registering them in a public index, this data may be exposed to the internet without explicit per-run verification of the content being shared.
  • [COMMAND_EXECUTION]: The skill uses extensive shell commands via Bash to manage the file system and interact with external tools. Specifically, it uses mkdir, cp, and rm for directory management, and git and gh for repository operations. The use of rm -rf /tmp/* in the allowed tools list is broader than the specific cleanup of its own temporary directories, representing a potential risk to other temporary files.
  • [EXTERNAL_DOWNLOADS]: The skill clones a repository (danielrosehill/Open-Research-Workspaces-Index) from GitHub to perform the auto-registration step. While this repository is owned by the author of the skill, it involves downloading and interacting with external content at runtime.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 4, 2026, 11:15 PM
Security Audit — agent-trust-hub — new-workspace