cdc-debezium-rails
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a pattern for processing external data streams originating from database changes.
- Ingestion points: The
PostCdcConsumerRuby implementation inSKILL.mdprocesses event payloads delivered via Kafka topics. - Boundary markers: The implementation processes structured JSON data; while it does not use explicit delimiters, the data is handled within a programmatic consumer rather than being directly interpolated into a prompt.
- Capability inventory: The consumer code is granted the capability to modify search indexes via
SearchIndex.upsertandSearchIndex.deletemethods. - Sanitization: The skill demonstrates a defensive coding pattern using a
serializemethod that acts as a field whitelist, ensuring only predefined columns (id,title,body,account_id) are ingested, which prevents unexpected database schema changes from affecting the downstream index.
Audit Metadata