The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the sf (Salesforce CLI) via the Bash tool to perform operations such as querying data (sf data query), creating records, and deploying/retrieving metadata (sf project deploy/retrieve). These capabilities are scoped to the sf command and are consistent with the skill's stated purpose of Salesforce development.
[INDIRECT_PROMPT_INJECTION]: The skill assists in building components (OmniScripts, FlexCards, Integration Procedures) that ingest and process external data, creating a surface for indirect injection.
- Ingestion points: OmniScript input elements (Text, Telephone, etc.), Integration Procedure HTTP Actions fetching data from external REST APIs, and Data Mapper Extract actions querying Salesforce records.
- Boundary markers: The skill does not explicitly define or instruct the agent to use specific delimiters (like XML tags or triple backticks) when processing untrusted data retrieved from the Salesforce org or external APIs.
- Capability inventory: The agent has access to sf data and sf api commands which allow for reading and writing data, as well as sf project deploy for modifying org metadata.
- Sanitization: The documentation encourages best practices such as adding validation to input elements and configuring error handling on action elements, which provides some mitigation at the application level.

sf-omnistudio