The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting data from external and potentially untrusted sources while maintaining access to sensitive capabilities.\n
Ingestion points: The skill retrieves content from the internet via the WebFetch and WebSearch tools and reads local project files including .nutmeg.user.md, docs/entity-resolution-routing.md, and docs/accuracy-guardrail.md.\n
Boundary markers: The instructions lack explicit delimiters or specific warnings for the agent to treat data from these external or file-based sources as untrusted or to ignore embedded instructions.\n
Capability inventory: The skill is authorized to use powerful tools including Bash for command execution, Write for file modification, and Agent for delegating tasks to other specialized assistants.\n
Sanitization: There are no defined procedures for sanitizing, validating, or escaping content retrieved from web sources or project files before it influences agent routing logic or tool parameters.

nutmeg