camunda-ai-agents
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes a framework for processing untrusted user input within AI-driven business processes.
- Ingestion points: User-provided data (e.g.,
customerId,issue) is interpolated into prompts inSKILL.mdand extracted as tool parameters via thefromAi()function inreferences/fromai.md. - Boundary markers: The provided examples do not use explicit delimiters (like XML tags or triple quotes) for user input inside prompts, which is a potential surface for injection.
- Capability inventory: The agent can invoke tools modeled as BPMN activities, including REST connectors (
io.camunda:http-json:1), FEEL script tasks, and user tasks. - Sanitization: The
fromAi()function provides structural validation by allowing developers to define expected types (string,number,boolean) and JSON Schema fragments (e.g., enums), which helps prevent schema confusion attacks. - [EXTERNAL_DOWNLOADS]: The skill uses the
c8ctlCLI tool to synchronize element templates from Camunda's official OOTB (Out-Of-The-Box) catalog. This is a standard operation for the vendor's ecosystem and targets trusted infrastructure.
Audit Metadata