The Agent Skills Directory

[Indirect Prompt Injection] (SAFE): The skill implements a data ingestion surface via database triggers.
Ingestion points: SQL triggers in scripts/create_broadcast_trigger.sql read data from database rows (NEW/OLD variables).
Boundary markers: None; data is transmitted as standard JSON payloads.
Capability inventory: The skill enables broadcasting data from the database to connected clients via realtime.broadcast_changes in scripts/create_broadcast_trigger.sql.
Sanitization: Data is broadcast as-is from the database tables. While this creates a surface if the database contains malicious instructions, the skill does not use an LLM to process this data.
[Privilege Escalation] (SAFE): SQL functions use SECURITY DEFINER to enable cross-schema interaction with the Supabase realtime schema. This is an idiomatic practice for Supabase Realtime triggers and is properly mediated by application-level RLS policies provided in the guides.
[Data Exposure & Exfiltration] (SAFE): No hardcoded secrets, sensitive file paths, or unauthorized network operations were detected. All code examples use placeholders for configuration.

supabase-realtime