The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with external Salesforce data via the sf CLI, creating a surface where malicious data from the org could attempt to influence agent behavior.
Ingestion points: Data and metadata retrieved via sf data query and sf data query --explain commands as described in SKILL.md and references/soql-reference.md.
Boundary markers: Absent. The skill does not specify delimiters to wrap external data during analysis.
Capability inventory: The skill environment in SKILL.md allows file manipulation (Read, Write, Edit) and shell execution restricted to the Salesforce CLI (Bash(sf *)).
Sanitization: Present. SKILL.md contains explicit instructions to prevent SOQL injection in generated Apex code by using bind variables and WITH USER_MODE.

sf-soql