The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it is designed to ingest and analyze data from external, untrusted sources such as competitor ad libraries, product reviews, and landing pages.
Ingestion points: Data is pulled from external URLs via browser tools, specifically targeting ad libraries and review platforms mentioned in references/creative-research-automation.md and references/google-ads-audit-checklist.md.
Boundary markers: The instructions in references/audit-guardrails.md provide clear guidance to the agent to distinguish between data and instructions, explicitly telling it to ignore embedded directives like 'ignore previous instructions.'
Capability inventory: The skill utilizes tool integrations for various ad platforms (Google, Meta, LinkedIn, etc.) and browser/Slack connectors for automated research workflows.
Sanitization: Security-focused instructions are present in references/audit-guardrails.md and references/creative-research-automation.md, mandating that external content be treated as data and alerting the agent to common prompt injection patterns.

ads