investigating-agentforce-d360
Pass
Audited by Gen Agent Trust Hub on Jul 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes the Salesforce CLI tool (sf) to manage authentication and retrieve access tokens from the local environment. It uses the subprocess.run method with structured argument lists to ensure command integrity.
- [EXTERNAL_DOWNLOADS]: The skill retrieves session logs and audit metadata from the Salesforce Data Cloud Query REST API. It uses the standard Python urllib.request library to perform authorized POST requests to official Salesforce service endpoints.
- [PROMPT_INJECTION]: The skill handles LLM prompts and responses from previous sessions which could contain malicious data. It mitigates indirect injection risks by using markdown code fences for encapsulation and applying strict character decoding and truncation to all rendered content.
- [SAFE]: The skill incorporates extensive security checks, including organization ID and UUID validation via regex in scripts/_shared/fs_guard.py. It also employs centralized SQL literal escaping to prevent injection when constructing queries for the Data Cloud API.
Audit Metadata