init-repo
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches the official installation script for the
uvpackage manager fromastral.shand downloads R packages from the official RStudio CRAN mirror. - [REMOTE_CODE_EXECUTION]: The generated
Dockerfiletemplate includes a command that downloads a script fromastral.shand pipes it to a shell for execution as part of the container build process. - [COMMAND_EXECUTION]: The skill instructs the agent to perform local environment setup tasks such as
git init,uv init,uv sync, and package installation viarenvif the respective tools are available on the host system. - [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided project names and descriptions, which are then interpolated into various generated files including
README.mdandCLAUDE.md. This represents a standard scaffolding surface. - Ingestion points: User input during the
/init-repocommand flow (SKILL.md). - Boundary markers: None identified in the provided templates.
- Capability inventory: The skill uses
git,uv, andrenvcommands to configure the workspace and manage dependencies. - Sanitization: No explicit sanitization or validation of user-provided strings is defined before they are written to the project files.
Audit Metadata