The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from a contact CSV which could potentially contain malicious instructions or injection attempts intended to influence the agent.
- Ingestion points: The contacts_csv file is read and processed in the main function of references/render_emails.py.
- Boundary markers: The system uses a specific {{field}} placeholder syntax and enforces a mandatory check (assert_rendered) to ensure no unsubstituted placeholders remain in the output.
- Capability inventory: The skill performs local file read and write operations on CSV files. It does not possess network access, subprocess execution, or dynamic code evaluation capabilities.
- Sanitization: The renderer implements html.escape() on the final assembled string for HTML targets, which effectively sanitizes substituted field values against HTML injection/XSS before they are written to the output file.

email-generation