simplify-and-refactor-code-isomorphically
Fail
Audited by Gen Agent Trust Hub on Jun 25, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The script
scripts/install_jsm.shdownloads and executes remote code via piped bash/PowerShell commands from an external domain (https://jeffreys-skills.md/install.shandinstall.ps1). This is a high-risk pattern allowing unverified code execution from a non-whitelisted source. - [EXTERNAL_DOWNLOADS]: The skill utilizes a custom package manager (
jsm) to install additional components at runtime. This introduces a supply chain risk by downloading and executing unverified code from a third-party source not in standard package registries. - [COMMAND_EXECUTION]: During the baseline and verification phases, the skill automatically executes shell commands (e.g., test and lint scripts) extracted directly from the target project's manifest files, such as
package.jsonandCargo.toml. This allows for arbitrary code execution if the repository being analyzed is malicious. - [PROMPT_INJECTION]: The skill processes untrusted source code and configuration files from project repositories and uses this data to drive its workflow, including command execution. The lack of sanitization or robust boundary markers creates an extensive surface for indirect prompt injection.
- [DATA_EXFILTRATION]: The
scripts/check_skills.shscript records the output ofjsm whoami, which often contains the user's email address, and stores it in plaintext in the project's artifact directory. This poses a risk of exposing sensitive user information if the artifacts are shared, committed to version control, or accessed by other agents.
Recommendations
- HIGH: Downloads and executes remote code from: https://jeffreys-skills.md/install.sh - DO NOT USE without thorough review
- AI detected serious security threats
Audit Metadata