The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to suggest running npx skills add wondelai/skills/<slug> --global if a required component is missing. This involves downloading packages from the npm registry and resources from a repository associated with an unverified author ('wondelai').
[REMOTE_CODE_EXECUTION]: The use of npx combined with external repository slugs allows for the execution of code fetched from remote sources at runtime. While this is presented as a user-approved fallback, it enables the introduction of unverified logic into the agent's environment.
[COMMAND_EXECUTION]: The skill explicitly includes shell commands in its operating rules, specifically for package management and installation, which grants the agent the ability to influence the host system's software configuration.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to how it handles session persistence.
Ingestion points: The skill reads docs/GROW-WEBSITE-PLAN.md and other documentation files in the docs/ directory upon initialization and at the start of each phase.
Boundary markers: There are no boundary markers or instructions defined to prevent the agent from following directions that might be maliciously embedded in the project documentation.
Capability inventory: The skill possesses the ability to execute shell commands (npx) and perform arbitrary file writes to the local docs/ directory.
Sanitization: The skill does not perform sanitization, filtering, or validation on the content retrieved from the project files before processing it as state information.

grow-website