The Agent Skills Directory

[SAFE]: The skill acts as a developer reference for Salesforce Agentforce, providing documentation on agent design, topic configuration, and Apex integration snippets.
[EXTERNAL_DOWNLOADS]: The skill includes Apex code templates for integrating with OpenAI and Anthropic APIs. These samples use Salesforce's callout: mechanism, which leverages secure Named Credentials to handle authentication rather than hardcoding secrets, representing a standard and safe development practice.
[INDIRECT_PROMPT_INJECTION]: The skill includes patterns for agents that process user-supplied data (e.g., customer messages).
Ingestion points: The userMessage field in the AgentOpenAIIntegration and AgentClaudeIntegration Apex classes in SKILL.md serves as the entry point for untrusted data.
Boundary markers: The provided instruction template in references/agentforce-builder.md includes behavioral guidelines (personality, identity verification) but does not provide specific technical delimiters for input data.
Capability inventory: The agents utilize Apex actions for external API callouts, database queries (SOQL), and logging interactions (DML).
Sanitization: Code snippets provided are minimal templates and do not demonstrate explicit sanitization or filtering of the userMessage content.
Analysis: While this creates a potential attack surface, it is a direct consequence of the skill's primary purpose—building interactive AI agents. The inclusion of identity verification and audit trail instructions mitigates the risk.

agentforce-2025