The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external lead generation and professional profile sources (LinkedIn, Apollo, etc.), which creates a potential surface for indirect prompt injection if those sources contain adversarial content.
Ingestion points: Data enters the context through sandbase_call_tool calls to services like agentbody_linkedin_profile and apollo_company_enrich as described in SKILL.md and references/sandbase-api-map.md.
Boundary markers: The skill lacks explicit prompt delimiters to separate untrusted tool output from internal instructions, though it does instruct the agent to separate factual observations from recommendations.
Capability inventory: The skill is restricted to read-only research tools and does not have file system write access or arbitrary shell execution capabilities.
Sanitization: No specific data sanitization or filtering instructions are provided for handling external content.

lead-research