The Agent Skills Directory

[DATA_EXFILTRATION]: The deployment instructions include shell commands that interact with sensitive files. Specifically, the JWT authentication flow references a private key file (server.key), and CI/CD workflow examples involve storing authentication secrets in temporary files (auth.txt).
[DYNAMIC_EXECUTION]: The skill provides implementation patterns for dynamic SOQL queries using Database.query() and instructions for running anonymous Apex code from files using sf apex run. These techniques allow for the execution of code strings that are assembled or loaded at runtime.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to download the Salesforce CLI from the official Salesforce developer domain (developer.salesforce.com). This download targets a well-known service and is used for environment setup and deployment.
[INDIRECT_PROMPT_INJECTION]: The skill defines a large attack surface for indirect prompt injection through the ingestion of untrusted data from external REST APIs and platform events.
Ingestion points: Custom REST API endpoints (AccountRestService) and Change Data Capture handlers (AccountChangeEventHandler) in references/integration-patterns.md process data from external sources.
Boundary markers: The provided code examples lack explicit boundary markers or instructions to ignore embedded commands within ingested data.
Capability inventory: The skill scripts have access to network callouts via HttpCalloutService, shell command execution via the Salesforce CLI, and file system write operations.
Sanitization: While the skill demonstrates best practices like String.escapeSingleQuotes and WITH SECURITY_ENFORCED, these primarily address injection into database queries rather than LLM context safety.

salesforce-developer