The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill generates Python source code and JSON definitions by interpolating user-supplied input (such as DocType names and field lists) into templates. This constitutes an injection surface if the input is crafted to escape string literals in the generated code.
Ingestion points: Step 1 (Gather DocType Requirements) ingests variable user data including names, modules, and fields.
Boundary markers: The skill includes a mandatory Step 8 (Show Preview and Confirm) which requires the user to review the generated structure before execution.
Capability inventory: The skill specifies file creation across multiple directories and shell execution of the bench CLI.
Sanitization: There is no explicit sanitization logic defined in the procedure for the user inputs; the security relies on the agent's safe code generation and the human review step.
[COMMAND_EXECUTION]: The procedure involves executing shell commands to finalize the environment setup and verify the generated code.
Evidence: Step 9 (Execute and Verify) instructs the agent to run bench --site <site> migrate and bench --site <site> run-tests --doctype "<DocType Name>" using the provided parameters.

frappe-doctype