agentforce-test

Pass

Audited by Gen Agent Trust Hub on Nov 3, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONOBFUSCATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill makes extensive use of the sf CLI (Salesforce CLI) and curl to perform operations such as starting agent previews, deploying test suites, and executing Salesforce actions via REST APIs. These are standard operations for Salesforce development and testing tools.
  • [PROMPT_INJECTION]: Several files, including assets/payloads/prompt-injection.yaml, assets/guardrail-test-spec.yaml, and references/preview-testing.md, contain explicit prompt injection patterns (e.g., "Ignore all previous instructions", "DAN" personas). These are used as test case data to evaluate an agent's resistance to such attacks and are not intended to override the behavior of the agent skill itself.
  • [OBFUSCATION]: The skill includes Base64-encoded strings (e.g., in assets/payloads/prompt-injection.yaml) and references to hex/ROT13 encoding as part of its adversarial technique catalog. These are presented as payloads to test an agent's ability to decode and resist hidden instructions, representing a core functional feature of the security testing mode.
  • [INDIRECT_PROMPT_INJECTION]: The skill is specifically designed to identify vulnerabilities to indirect prompt injection by analyzing .agent files and testing how agents process external data. It includes remediation guidance for adding boundary markers and sanitizing inputs.
  • [DATA_EXPOSURE]: The skill handles Salesforce session tokens (access tokens) and interacts with org metadata. It follows security best practices by recommending the use of sandboxes for adversarial testing and providing warnings before performing write operations (DML) in production environments.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 3, 2026, 07:56 PM