kafka-stream-processing
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documentation and examples focus on building Kafka consumers that ingest records from external data topics. This constitutes an attack surface for indirect prompt injection where data processed from a stream could contain instructions targeting the LLM.
- Ingestion points: Kafka consumer implementations using
consumer.poll()inEXAMPLES.md(Examples 6-9, 16, 20) andSKILL.md. - Boundary markers: Not explicitly defined in the provided code snippets to isolate data from agent instructions.
- Capability inventory: The skill demonstrates capabilities for network communication (Kafka protocol), file system/database interactions (Kafka Connect Sink connectors), and search engine indexing (Elasticsearch Sink).
- Sanitization: While the skill uses schema-based deserialization (Avro), it does not include logic for sanitizing data content for potential prompt-level instructions.
- [SAFE]: The skill uses industry-standard practices for Kafka administration and security, including SSL/TLS and SASL authentication configurations. All shell commands are for local cluster management (
kafka-topics.sh,kafka-console-producer.sh) and all code is standard Java/Maven boilerplate. No obfuscation, hardcoded secrets, or persistence mechanisms were detected.
Audit Metadata