kibana-agent-builder

Warn

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill relies on Bash execution to run Node.js scripts (scripts/agent-builder.js) that interact with the Kibana Agent Builder REST API for agent and tool management.
  • [CREDENTIALS_UNSAFE]: The scripts scripts/agent-builder.js and scripts/kibana-client.js implement a bypass for TLS certificate verification. If the environment variable KIBANA_INSECURE is set to true, the scripts set process.env.NODE_TLS_REJECT_UNAUTHORIZED = "0". This disables security checks for HTTPS connections, making Kibana API keys, usernames, and passwords vulnerable to interception via man-in-the-middle (MITM) attacks.
  • [PROMPT_INJECTION]: The skill facilitates the creation and management of agents that ingest untrusted data and follow custom instructions, creating a surface for indirect prompt injection.
  • Ingestion points: User-provided system instructions during agent creation/update and external data retrieved from Elasticsearch indices via ES|QL and index search tools.
  • Boundary markers: No explicit boundary markers or "ignore embedded instructions" warnings are implemented in the management scripts to isolate instructions from data.
  • Capability inventory: The management scripts can create, modify, and delete agents and tools, and can execute tool tests via the Kibana API. The agents created can further execute tools like platform.core.execute_esql or platform.core.search.
  • Sanitization: The scripts use standard JSON stringification and URI encoding for parameters, but do not provide specific sanitization for content passed into agent instructions or tool queries.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 13, 2026, 04:50 PM