sf-ai-agentforce-observability
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/auth.pyutilizessubprocess.runto invoke the Salesforce CLI (sf) for organization discovery. This execution is scoped to a specific local binary and does not utilize a shell environment for the call. - [CREDENTIALS_UNSAFE]: The authentication module in
scripts/auth.pyis designed to read RSA private keys from the standard Salesforce CLI directory (~/.sf/jwt/). This is a standard security practice for establishing JWT Bearer authentication with Salesforce services. - [EXTERNAL_DOWNLOADS]: The skill specifies dependencies on well-known Python packages (such as
polars,pyarrow, andhttpx) and communicates with official Salesforce domains for authentication and data retrieval from Data Cloud. - [INDIRECT_PROMPT_INJECTION]: The skill processes session telemetry data that could contain untrusted user-supplied content. However, the ingestion and analysis are handled via structured data libraries (Polars), and the skill provides visibility into agent behavior without executing untrusted instructions.
Audit Metadata