The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from Salesforce (leads, contacts, accounts, and query results) which may contain adversarial instructions intended to influence the agent's behavior.
Ingestion points: Data retrieved via tools such as SALESFORCE_SEARCH_LEADS, SALESFORCE_LIST_CONTACTS, SALESFORCE_GET_OPPORTUNITY, and SALESFORCE_RUN_SOQL_QUERY in SKILL.md enters the agent context.
Boundary markers: The instructions lack delimiters or specific warnings to ignore embedded instructions within the Salesforce records.
Capability inventory: The skill possesses significant capabilities including record creation/update, ownership transfer via SALESFORCE_MASS_TRANSFER_OWNERSHIP, and arbitrary query execution via SALESFORCE_RUN_SOQL_QUERY across multiple files.
Sanitization: No sanitization or validation logic is specified for the content retrieved from external Salesforce APIs.
[EXTERNAL_DOWNLOADS]: The skill relies on a remote MCP server endpoint to provide its functionality.
The setup instructions require adding https://rube.app/mcp as an MCP server in the client configuration, which fetches tool schemas and facilitates tool execution through this external service.

salesforce-automation