The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8). It ingests untrusted data from visitor research, including chat logs and support tickets (Intake section in SKILL.md), which is then processed to build project artifacts. Ingestion points: Intake questions regarding research data. Boundary markers: None present. Capability inventory: Writing to the local docs/ folder and invoking other skills. Sanitization: None present. Additionally, a discrepancy exists between the author identified in the YAML frontmatter ('wondelai') and the uploader context ('emmavelazquez'), which represents potentially deceptive metadata.
[EXTERNAL_DOWNLOADS]: In the Operating Rules section of SKILL.md, the skill instructs the agent to suggest that the user run npx skills add wondelai/skills/<slug> --global. This command initiates a download of code from a third-party source ('wondelai') that is not verified and does not match the provided author context ('emmavelazquez').
[COMMAND_EXECUTION]: The skill recommends the use of the npx command to install additional components. This command can execute arbitrary code from a remote registry. While the agent only offers this as a suggestion to the user, it encourages the adoption of unverified software from an external namespace.

grow-website