The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The source catalog file 'references/source-catalog.json' includes a reference to 'https://brandarchive.xyz/', which has been flagged as a malicious URL by automated security scanners. The instructions in 'SKILL.md' direct the agent to visit these external sources using its browsing or computer-use capabilities.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it analyzes external websites and user-supplied screenshots or videos to extract design principles. 1. Ingestion points: Untrusted content from external websites listed in the catalog and user-provided visual media. 2. Boundary markers: Absent; there are no instructions to filter or ignore potentially malicious text or hidden instructions embedded in the analyzed references. 3. Capability inventory: The skill executes local Python scripts for catalog management and utilizes the agent's browsing or computer-use tools to interact with external sites. 4. Sanitization: Absent; the content retrieved from external sources is processed and interpreted without specific validation or sanitization steps.
[COMMAND_EXECUTION]: The skill utilizes local Python scripts ('scripts/search_catalog.py' and 'scripts/validate_catalog.py') to manage and validate its internal source catalog. While these are local files provided as part of the skill, they represent an execution surface within the agent's environment.

balise-visual-references