The Agent Skills Directory

Data Ingestion and Integration: The skill processes external CRM data, including activity histories and conversation transcripts, which is essential for its primary function. While this creates a surface for indirect prompt injection, the skill includes instructions for metadata verification and precise tool usage to manage this context safely.
Ingestion points: summarize_conversation_transcript and get_activity_history in SKILL.md.
Boundary markers: The skill relies on structured tool interactions rather than unconstrained prompt interpolation.
Capability inventory: Write access is scoped to specific CRM tools such as update_record, create_account_plan, and assign_target_to_sdr in SKILL.md.
Sanitization: Instructions require verifying Salesforce object and field metadata before any querying or writing takes place.
Access Control and Policy: The skill's configuration disables implicit invocation, meaning the agent will not perform Salesforce-related tasks unless explicitly directed by the user, providing a layer of oversight for CRM interactions.
Dynamic Resource Generation: Record links are constructed using instance-specific metadata provided at runtime. This allows the agent to provide accurate, clickable links to the user's own Salesforce records without hardcoding environment-specific URLs.

salesforce