configuring-ai-agents
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill defines a configuration for processing external, untrusted data via LLMs, establishing a surface for indirect prompt injection.
- Ingestion points: External records are ingested through the
text,record, andfilesinput fields as defined inSKILL.mdandaiagent.yml. - Boundary markers: The documentation does not specify mandatory delimiters or "ignore previous instructions" markers for the model input, though it leverages structured JSON input.
- Capability inventory: The configuration allows agents to invoke tools including
web_search,googleSearch, MCP servers, and internal Celigo Tools. - Sanitization: The skill provides built-in mitigations through
GuardrailImportschemas, enabling PII masking (guardrail.pii) and content moderation (guardrail.moderation) to filter harmful or sensitive content. - [COMMAND_EXECUTION]: The documentation instructs the agent on the use of the
celigoCLI for account discovery and resource management (e.g.,celigo ai-agents list,celigo account search). These are first-party vendor-provided tools for the Celigo platform. - [SAFE]: The skill correctly emphasizes secure credential management by instructing users to use
_connectionIdfor BYOK (Bring Your Own Key) scenarios, avoiding the exposure of sensitive API keys within the skill configuration.
Audit Metadata