agentforce-architecture-analyze
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes Salesforce CLI (sf) commands using pre-defined local YAML recipes. Input parameters are strictly validated against a restrictive alphanumeric regex (^[A-Za-z0-9_]+$) to prevent command injection and directory traversal.\n- [EXTERNAL_DOWNLOADS]: Fetches metadata from official Salesforce REST API domains. All network communication is restricted to the target organization's Salesforce instance.\n- [DATA_EXFILTRATION]: Retrieves Salesforce metadata and session tokens as part of its core functionality. It implements comprehensive redaction logic to ensure authentication tokens and sensitive identifiers are scrubbed from all logs, error messages, and tracebacks.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests metadata (such as topic descriptions and API names) from the Salesforce environment to generate documentation. While it escapes formatting characters, the inclusion of external metadata in the output constitutes a standard ingestion surface common to documentation tools.
Audit Metadata