The Agent Skills Directory

Human-Verified Write Operations: The skill implements a strong security posture for data integrity by requiring explicit user approval before performing any write or update operations in Salesforce. It follows a multi-step process—clarify, draft, and then act—ensuring that the user maintains full control over CRM changes.\n- Database Query Guardrails: To protect performance and prevent unauthorized data access, the skill includes specific instructions for constructing SOQL queries. It prohibits aggregate functions and limits queries to essential fields and filterable attributes, reducing the risk of broad data exposure or resource exhaustion.\n- Controlled Access to Salesforce Tools: The skill explicitly defines a restricted set of 'Agentforce Sales' surfaces and tools it is authorized to use. It intentionally omits high-privilege operations like bulk deletions or arbitrary automation execution, adhering to the principle of least privilege.\n- External Data Ingestion Considerations: The skill is designed to process external information such as CRM records and conversation transcripts. While this introduces a potential surface for indirect prompt injection, the risk is effectively mitigated by the skill's mandatory user-review phases and its reliance on verified record identifiers rather than autonomous decision-making.\n
Ingestion points: Data enters the agent context through tools like soql_query, get_record_details, and summarize_conversation_transcript as defined in SKILL.md.\n
Boundary markers: The skill uses clear, structured markdown headers and tables for presentation, though it does not explicitly define 'ignore' delimiters for external text content.\n
Capability inventory: Sensitive actions include create_record, update_record, and assign_target_to_sdr.\n
Sanitization: Security is maintained through a 'First draft' workflow where the agent must present proposed changes for user verification before execution.

salesforce