The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to automatically ingest and process data from external files, creating a vulnerability surface where malicious instructions could be embedded in data.
Ingestion points: The skill instructs the agent to read from .agents/product-marketing.md, .claude/product-marketing.md, or product-marketing-context.md at the start of the interaction.
Boundary markers: The instructions do not define any delimiters or specific boundary markers to distinguish between legitimate data and potential instructions within the external context files.
Capability inventory: The agent leverages its core reasoning and tool-invocation capabilities (such as generating marketing strategies and channel architectures) based on the data ingested from these files.
Sanitization: No sanitization, validation, or filtering of the external file content is described, allowing potentially malicious instructions to enter the agent's context unchecked.

community-marketing