The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill directs the agent to ingest and process data from external or user-provided files, which creates a potential surface for indirect prompt injection should those files contain adversarial instructions.\n
Ingestion points: In SKILL.md, the agent is instructed to read .agents/product-marketing-context.md or .claude/product-marketing-context.md for context. Additionally, references/reverse-engineering.md provides a framework for analyzing large datasets of scraped social media content.\n
Boundary markers: The skill does not provide instructions to use delimiters or ignore potential commands embedded within the marketing context or social media data.\n
Capability inventory: The skill configuration contains no allowed-tools and the files consist entirely of markdown instructions with no executable code, shell commands, or network operations.\n
Sanitization: No sanitization or validation logic is specified for the external data being processed by the agent.

social-content