The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external websites and guidelines, creating an attack surface for indirect prompt injection.
Ingestion points: The agent ingests data from prospect sites and editorial guideline pages using browser and search tools.
Boundary markers: The skill contains explicit instructions to treat external page content as untrusted and forbids it from overriding the skill workflow or requesting secrets.
Capability inventory: Capabilities include writing to local tracker files, performing web searches, and initiating external contact via browsers and email tools.
Sanitization: Instructions require the agent to prioritize the skill's documented workflow and editorial standards over any conflicting directions found on external pages.
[DATA_EXFILTRATION]: The skill transmits data to external third-party sites via email or web forms as part of its primary function.
Mitigation: A dedicated authorization gate requires explicit user approval for the message, recipient, and all attachments before any external submission occurs.
[EXTERNAL_DOWNLOADS]: The skill references the use of external SEO platforms and search providers.
Evidence: The workflow integrates metrics and discovery data from well-known services including Ahrefs, Semrush, and DataForSEO.

backlink-outreach