flink-udf

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the confluent CLI and docker to perform infrastructure management, artifact creation, and function registration. These operations are restricted to the intended developer workflow.
  • [EXTERNAL_DOWNLOADS]: The skill fetches environment configurations and CLI plugins from Confluent's official GitHub repositories and domains. These are trusted vendor resources associated with the skill's author.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection because it ingests user-defined business logic to generate and deploy Java code.
  • Ingestion points: User-provided logic for scalar UDFs, UDTFs, or PTFs in SKILL.md.
  • Boundary markers: The instructions include a mandatory safety workflow (Step 5) requiring the agent to display the full deployment plan, including exact commands and JAR paths, and wait for explicit user confirmation before proceeding.
  • Capability inventory: Subprocess calls for Maven/Gradle builds, docker management, and confluent CLI operations across all reference guides.
  • Sanitization: The primary protection against malicious generated code is the required human-in-the-loop verification step.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 03:27 PM
Security Audit — agent-trust-hub — flink-udf