The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes arbitrary Salesforce Apex code via the sf apex run command. While powerful, the skill mitigates risk by requiring explicit developer confirmation of all snippets and recommending Database.setSavepoint() and Database.rollback() wrappers to ensure the organization state remains untouched during verification tests.
[EXTERNAL_DOWNLOADS]: The skill relies on the sf (Salesforce CLI) tool. This is the official tool provided by the vendor (Salesforce/forcedotcom) and is a standard requirement for Salesforce development workflows.
[PROMPT_INJECTION]: The skill includes instructions to ignore authoring tasks for other file types (e.g., .cls or .trigger) and redirects to appropriate specialized skills, reducing the chance of the agent being misused for unintended code generation.
[DYNAMIC_EXECUTION]: The skill generates temporary .apex files from user-pasted snippets to avoid shell-escaping issues. It uses the platform's Write tool rather than shell heredocs to ensure the content is preserved verbatim and not subjected to shell expansion.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect injection as it processes external code snippets.
Ingestion points: Pasted Apex snippets and local .apex files.
Boundary markers: The skill explicitly instructs the agent to echo back snippets and wait for a "yes, run it" confirmation from the user before execution.
Capability inventory: The agent uses the Bash tool to run the CLI and the Write tool to create temporary files.
Sanitization: The instructions explicitly forbid the use of shell heredocs for script generation to prevent unintended variable expansion or injection at the shell level.

platform-apex-anonymous-run