The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Node.js scripts to execute the Salesforce CLI (sf) for interacting with org APIs. It employs child_process.spawnSync with argument arrays, avoiding shell interpretation and mitigating potential command injection. Additionally, input parameters such as org aliases and API paths are validated against strict regular expressions and character checks.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external Salesforce environments (worksheets, prompt templates, and agent outputs). It proactively mitigates the risk of indirect prompt injection by wrapping all retrieved data in a trustBoundary object that explicitly instructs the AI agent to treat the content as data rather than instructions. SKILL.md further reinforces these guardrails by requiring the agent to disregard instructions embedded in retrieved data unless explicitly restated by the user.
[SAFE]: The skill follows security best practices by delegating authentication to the local Salesforce CLI configuration rather than handling sensitive credentials or tokens directly. It also implements data sanitization (truncation and control character removal) before presenting results to the AI agent.

sf-ai-agentforce-grid