The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides standard deployment instructions using the Salesforce CLI (sf project deploy start). These are legitimate developer workflows for deploying metadata to a Salesforce org.
[PRIVILEGE_ESCALATION]: The skill's Apex templates and guides recommend using the without sharing keyword for certain surfaces like the Service Rep Assistant (SRA). This is documented as a necessary architectural choice to ensure the agent user (EinsteinServiceAgent) can access required records (like Contacts or MessagingSessions) which it may not have sharing access to by default. This is a common and legitimate pattern in Salesforce bot development.
[PROMPT_INJECTION]: The skill includes instructions for the agent to output the full results of its actions using specific platform commands (show_command) without summarizing them. This is a technical requirement to ensure the LWC renderer receives the raw JSON payload necessary to display the UI card correctly.
[INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface by ingesting data from Salesforce objects and user inputs into custom UI components.
Ingestion points: Apex classes (e.g., GetVerificationCardAction.cls) query data from Contact and MessagingSession objects.
Boundary markers: The instructions use explicit directives to use show_command for rendering.
Capability inventory: The skill uses Bash, Read, Write, and Edit tools to generate and deploy code.
Sanitization: The skill uses standard JSON.serialize and JSON.parse methods for data transport between Apex and LWCs.
[EXTERNAL_DOWNLOADS]: The skill references several Salesforce-hosted help and developer documentation sites (help.salesforce.com, developer.salesforce.com). These are well-known, trusted domains and are documented neutrally.

sf-clt-builder