The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes metadata content from external Salesforce organizations, which constitutes an ingestion point for untrusted data.
Ingestion points: The skill retrieves Apex code bodies, validation rule formulas, flow definitions, and custom labels from external source and target orgs via Tooling and SOQL queries (references/diff-method.md).
Boundary markers: Findings are structured into a predefined "Drift report" format, though no explicit sanitization of the metadata content within the report is mentioned.
Capability inventory: The skill can execute local Python scripts or shell commands via the sf CLI to process inventory data, depending on the detected execution mode (references/execution-modes.md).
Sanitization: The skill lacks explicit instructions to sanitize or escape retrieved metadata content (e.g., comments in Apex classes) before incorporating it into the final report or passing it to subsequent tools.
[SAFE]: The skill implements a robust identity confirmation gate (Phase 0) that verifies the Organization Id on both connectors to prevent "self-diff" errors and performs direction sanity checks to ensure the user does not unintentionally reverse source and target orgs.

sf-orgdiff