The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to suggest a command (npx skills add wondelai/skills/<slug> --global) to the user. This involves the installation and execution of code from a remote source that is not recognized as a trusted vendor, potentially leading to the execution of unverified scripts.
[COMMAND_EXECUTION]: Operating Rule 4 directs the AI to provide specific shell commands for the user to run. This encourages the execution of terminal commands outside of the controlled environment, which could be used to modify system settings or install untrusted software.
[EXTERNAL_DOWNLOADS]: The skill references an external repository structure (wondelai/skills/) for additional components. The reliance on third-party resources from an unverified author introduces potential supply chain risks.
[PROMPT_INJECTION]: The skill reads from local documentation files (e.g., docs/CREATE-WEBSITE-PLAN.md) to resume its state, creating a surface for indirect prompt injection.
Ingestion points: Files in the docs/ folder are read during the initial session resume phase (Operating Rule 1).
Boundary markers: No delimiters or warnings are used when reading these files to prevent the execution of embedded instructions.
Capability inventory: The skill has the ability to read and write files, call other skills, and generate terminal commands.
Sanitization: There is no documented logic for sanitizing the content of the project files before the agent processes them.

create-website