The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill generates Python service classes that are intended to ingest and process external data (e.g., via the create and update methods). This creates a vulnerability surface where malicious instructions embedded in the processed data could attempt to influence the agent's behavior.
Ingestion points: Untrusted data enters the generated service logic through the data dictionary arguments in the create and update methods provided in the template in SKILL.md.
Boundary markers: The template includes placeholders for validation logic such as _validate_create_data and _validate_submission, which serve as potential boundary markers if properly implemented.
Capability inventory: The generated code has the capability to write to the database (repo.create, doc.save, doc.submit), execute workflow actions, and perform external network operations if the integration pattern is used.
Sanitization: The skill provides a specific 'Security Guidelines' section that instructs the user to use parameterized queries to prevent SQL injection and to avoid using eval() or exec() with user input.

frappe-service