The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential surface for indirect prompt injection by demonstrating how to interpolate user-provided data directly into the agent's prompts.
Ingestion points: Untrusted data is ingested via the data.userPrompt.prompt field in SKILL.md using variables such as customerId and issue.
Boundary markers: The prompt examples (e.g., ="Customer " + customerId + " reports: " + issue) do not include explicit delimiters or instructions to treat the variables as non-executable data.
Capability inventory: The agent has the ability to execute tools including REST connectors (io.camunda:http-json:1) and script tasks, which could be leveraged if an injection succeeds (see references/tool-modeling.md).
Sanitization: There are no recommendations or examples for sanitizing input variables before prompt construction.
[COMMAND_EXECUTION]: The documentation provides examples of using the c8ctl CLI for legitimate development tasks, such as applying element templates and linting BPMN files (e.g., c8ctl element-template apply and c8ctl bpmn lint). These are standard operations for Camunda 8 developers.
[EXTERNAL_DOWNLOADS]: The skill mentions the c8ctl element-template sync command, which fetches official connector templates from Camunda's registry. This is an expected configuration step within the vendor's ecosystem.

camunda-ai-agents