The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The SKILL.md file (Step 5) requires the agent to download and consult instructions from a third-party repository at https://github.com/Leonxlnx/taste-skill. The instructions explicitly forbid maintaining a local mirror and demand checking the remote source directly for updates.
[PROMPT_INJECTION]: The skill's 'Taste Overlay' mechanism relies on external Markdown files to define 'Brief Inference' and 'Anti-slop hard rules'. This creates a significant surface for indirect prompt injection, as an attacker could modify the content of the untrusted remote repository to compromise the agent's logic or bypass established constraints.
[REMOTE_CODE_EXECUTION]: By mandating that the agent apply instructions fetched from an untrusted remote source to its UI implementation workflow, the skill creates a remote dependency. Malicious instructions injected into the remote Markdown could cause the agent to generate and execute harmful scripts within the platform's preview environment.

ui-design