data360-code-extension-generate

Pass

Audited by Gen Agent Trust Hub on Nov 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill frequently executes shell commands to interact with the Salesforce CLI (sf), Node.js (npm), and Python (pip). It also utilizes local utility scripts, check_env.sh and venv_hook.sh, to validate and configure the developer environment, including venv creation and path configuration.
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to download several critical components from the official Salesforce and project ecosystem, including the @salesforce/cli, the @salesforce/plugin-data-code-extension plugin, and the salesforce-data-customcode Python SDK. These resources belong to the vendor infrastructure and are used as intended for the development lifecycle.
  • [DYNAMIC_EXECUTION]: A core feature of the skill is the generation of PySpark transformation logic in entrypoint.py. This generated code is then executed locally using the sf data-code-extension script run command to test functionality against real org data before deployment.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests untrusted data from the Salesforce org (schema fields and sample data records) via sf api request or MCP tools during Phase 2.
  • Ingestion points: Salesforce Data Lake Object (DLO) schema fields and query results fetched in references/schema-discovery.md and references/mcp-integration.md.
  • Boundary markers: Absent; the data is directly reviewed by the agent for mapping fields.
  • Capability inventory: The skill can execute shell commands, write to the filesystem, and perform network requests to the Salesforce API across all referenced scripts.
  • Sanitization: The skill mitigates risks through strict mapping rules (e.g., lowercase only) and template-driven code generation, though it does not implement specific sanitization for adversarial instructions embedded in data metadata.
  • [SAFE]: Security is prioritized through several agent rules: Phase 0b requires a read-only environment check; Rule 6 explicitly prohibits printing secrets like access tokens; Rule 4 requires user confirmation for all org-modifying operations (deploy, delete); and privileged operations like sudo are handed off as manual user actions.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 13, 2026, 11:17 PM