The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The launcher scripts in the scripts/ directory download a platform-specific binary from a personal GitHub repository (pbakaus/impeccable) and execute it with user-supplied arguments. This pattern of downloading and running unvetted code from an untrusted external source is a high-risk security vulnerability.
[EXTERNAL_DOWNLOADS]: The skill relies on external binaries fetched at runtime from github.com/pbakaus/impeccable. Although the script implements a SHA256 integrity check, the source is not a recognized trusted vendor or organization.
[OBFUSCATION]: The file scripts/modern-screenshot.umd.js is provided in a minified format, which serves as a form of obfuscation, making it impossible to verify the logic and security of the code during static analysis.
[PROMPT_INJECTION]: In reference/craft.md, the agent is instructed to hide the use of specific commands from the user to favor natural language interactions. This concealment of the agent's actual tool use can be used to mask suspicious activity.
[INDIRECT_PROMPT_INJECTION]: The skill has an extensive attack surface for indirect prompt injection. It processes untrusted data from user projects (HTML, CSS, JS files) and has high-privilege capabilities, including the ability to write to the filesystem and execute external binaries through its launcher script. There are no explicit instructions or markers to prevent the agent from following instructions embedded within these project files.
[DYNAMIC_EXECUTION]: The skill implements a custom execution environment that dynamically downloads and runs binary code, which bypasses standard auditing and package management controls.

impeccable