platform-lightning-type-widget-coordinate
Pass
Audited by Gen Agent Trust Hub on Aug 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill acts as an orchestrator that processes user prompts and metadata from external Salesforce environments, which introduces a surface for indirect prompt injection.
- Ingestion points: User prompts are processed to determine orchestration paths, and metadata files (JSON and Apex) are ingested from the local file system and connected Salesforce orgs.
- Boundary markers: The skill enforces a Phase 3 Build Plan STOP gate that requires explicit developer approval before any files are modified or sub-skills are executed.
- Capability inventory: The orchestrator is capable of executing shell commands (sf, jq, grep, git, sed) and invoking sub-skills for code and metadata generation.
- Sanitization: A mandatory 'Hard Rule' explicitly prohibits the use of shell metacharacters such as backticks or $(...) in all tool calls to mitigate command injection risks.
- [COMMAND_EXECUTION]: The skill utilizes standard system CLI tools (sf, jq, grep, sed, git) for Salesforce development workflows. It includes specific defensive instructions to ensure these tools are used safely without risky shell expressions.
- [EXTERNAL_DOWNLOADS]: The skill uses the official Salesforce CLI (sf project retrieve) to fetch metadata from authorized Salesforce orgs. This is standard functionality for the vendor (forcedotcom) and involves trusted, authenticated sources.
Audit Metadata