The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest untrusted data from local context files, including .agents/product-marketing.md, .claude/product-marketing.md, and product-marketing-context.md.
Ingestion points: The agent is explicitly told to read these files before responding to user queries to gain product marketing context.
Boundary markers: There are no specific delimiters or instructions to treat the content of these files as data rather than instructions.
Capability inventory: The skill itself does not define custom tools but relies on the underlying agent's standard file and shell capabilities.
Sanitization: No sanitization or validation of the context file content is performed. While this creates a potential surface for indirect prompt injection, it is a standard pattern for context-aware agents.

paywalls