The Agent Skills Directory

[COMMAND_EXECUTION]: The skill explicitly directs the agent to use shell commands such as grep to perform self-audits on its generated code for forbidden colors, fonts, and icons. It also recommends using curl, wget, or Invoke-WebRequest to interact with external resources.\n- [EXTERNAL_DOWNLOADS]: The instructions encourage fetching remote assets (images, icons, and fonts) from arbitrary external URLs using command-line tools. The skill documentation specifically states that this approach is intended to bypass the SSRF and content-type filtering of previously removed tools, effectively instructing the agent to circumvent security constraints.\n- [INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by instructing the agent to download and process external assets from untrusted sources. \n
Ingestion points: Shell-based downloads (curl, wget, Invoke-WebRequest) defined in SKILL.md.\n
Boundary markers: None provided for remote asset downloads.\n
Capability inventory: Subprocess calls for curl, wget, grep, and Invoke-WebRequest as described in SKILL.md.\n
Sanitization: No validation or sanitization is performed on the downloaded external content.

ui-design