The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is instructed to read context from external files such as .agents/product-marketing-context.md or .claude/product-marketing-context.md at the start of an interaction. This creates a surface for indirect prompt injection if those files are populated with untrusted content designed to manipulate the agent's behavior.
Ingestion points: The skill instructions in SKILL.md explicitly direct the agent to read local marketing context files.
Boundary markers: There are no specific instructions or delimiters provided to ensure the agent ignores potential commands embedded within these context files.
Capability inventory: While the provided skill files primarily offer documentation and code snippets, the agent's overall capabilities (including those in the referenced tool registry) could be targeted by instructions found in the context files.
Sanitization: No validation or sanitization process for the external context data is mentioned in the instructions.

analytics-tracking