The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources to generate marketing materials, creating a potential attack surface.
Ingestion points: The agent is instructed to "inspect a supplied landing page" and use "available customer material" or "user-reported results" as specified in SKILL.md and references/creative-decisions.md.
Boundary markers: The skill requires the agent to categorize information as EVIDENCE, USER DIRECTION, or HYPOTHESIS, which acts as a logical boundary to separate source data from agent reasoning.
Capability inventory: The agent has access to built-in image generation tools and the ability to write files (images and manifests) to the local campaign workspace.
Sanitization: No explicit instructions are provided for sanitizing or escaping the text content fetched from external landing pages before it is processed.
[DATA_EXFILTRATION]: The skill performs network operations to fetch external content.
Evidence: The instructions in SKILL.md under the section "Preserve the real funnel" direct the agent to "Inspect a supplied landing page if its contents are unknown," which involves making network requests to arbitrary user-provided domains.

cs-marketing-image-ads