The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits deceptive metadata in SKILL.json. The author field claims to be 'Anthropic Marketplace 3p' and include an 'official-marketplace' tag, which mimics official platform branding. This represents metadata poisoning designed to gain higher trust than appropriate for a third-party resource.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface due to the processing of external telemetry without sanitization.
Ingestion points: Data retrieved from ZDX tools including zdx_list_applications, zdx_get_application, and zdx_list_alerts.
Boundary markers: Absent. The skill provides no instructions to the agent to escape or validate content before file generation.
Capability inventory: The skill is instructed to write mandatory .html and .docx report files to the local filesystem (SKILL.md Step 7).
Sanitization: Absent. The requirement to create a 'fully functional' HTML page with inline JavaScript and interactive features increases the risk that malicious data (such as script tags in alert descriptions) could execute in the user's browser (XSS) upon opening the generated report.

compare-locations