The Agent Skills Directory

[DATA_EXFILTRATION]: The skill extracts personal identity facts, locale, and timezone from the local file ~/.claude/CLAUDE.md. It then pushes this data to a GitHub repository that is configured to be public by default in the gh repo create command. This creates a risk of exposing sensitive career-related information, such as salary data and job search details, to the public internet.- [COMMAND_EXECUTION]: The skill uses several shell tools for file and repository management. These include mkdir and cp for workspace scaffolding, git for local version control, and gh (GitHub CLI) for remote repository creation and status checks.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through data ingestion. • Ingestion points: Content is read from the local ~/.claude/CLAUDE.md file. • Boundary markers: Absent. The extracted facts are directly interpolated into new workspace files. • Capability inventory: The skill has shell execution capabilities via mkdir, cp, git, and gh. • Sanitization: Absent. There is no validation or escaping of the content read from the source file before it is used to personalize the new workspace.

new-workspace