The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to translate natural language into Salesforce SOQL queries, which constitutes a surface for indirect injection where malicious input could attempt to generate resource-heavy or unsafe queries.
Ingestion points: Natural language user requests for query generation and optimization as defined in SKILL.md.
Boundary markers: The skill does not define specific delimiters for user input but provides strict structural templates for the generated SOQL output.
Capability inventory: The skill generates SOQL/SOSL code and documents Salesforce CLI (sf) command patterns for data operations.
Sanitization: The skill mitigates risks by explicitly instructing the agent to include security enforcement clauses (WITH SECURITY_ENFORCED, WITH USER_MODE) and to use Apex bind variables in generated code snippets to prevent database-level injection.
[COMMAND_EXECUTION]: The skill includes extensive documentation for Salesforce CLI commands, which could be misused if an agent is permitted to execute them without restriction.
Evidence: references/cli-commands.md provides executable patterns for sf data query and sf data export bulk. These are categorized as developer references but represent a potential command execution vector for the agent.
[DYNAMIC_EXECUTION]: The skill provides patterns for generating dynamic SOQL in Apex.
Evidence: In references/selector-patterns.md, a QueryBuilder class is demonstrated. Security is addressed by using Database.queryWithBinds and AccessLevel.USER_MODE, which are Salesforce-recommended patterns for safe dynamic execution.

sf-soql