isolated-setup-install
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- External Resource Acquisition: The skill downloads the latest version of the Claude Code tool via the official NPM registry and clones the Magpie framework from its repository. These actions use standard package management and version control systems to obtain necessary components from established organizations.
- Command Execution Patterns: The setup process involves running various system utilities such as
git,npm, andchmodto configure the environment and set file permissions. These commands are integral to the skill's purpose of automating a complex software installation. - Sandbox Bypass Mechanism: The skill utilizes a platform-specific parameter to perform certain file writes outside the standard agent sandbox. This is a potential consideration as it elevates the agent's permissions, but it is applied specifically to machine-local configuration files and is paired with a requirement for user acknowledgment.
- Persistence and System Integration: The installation process configures git hooks and system scripts, and proposes changes to shell initialization files. While these actions establish persistent behavior on the system, they are the primary intended function of the skill and are implemented through user-approved merges rather than silent modifications.
- Indirect Prompt Injection Surface: By reading directory structures, environment variables, and command outputs from the local host, the skill exposes an attack surface where untrusted data could influence agent behavior. The skill manages this by adhering to a strictly defined, step-by-step procedural document.
Audit Metadata