The Agent Skills Directory

[SAFE]: The skill serves as a configuration guide and reference for AI-driven import steps. It defines how data flows from records into AI models and back into integration pipelines.
[SAFE]: All external references, such as model providers (OpenAI, Google Gemini, Anthropic) and CLI commands (celigo), are standard industry services or vendor-specific tools related to the skill's purpose.
[SAFE]: The documentation includes security-focused components, such as GuardrailImport for PII detection and content moderation, which provide built-in safety mechanisms for the configured agents.
[PROMPT_INJECTION]: The skill describes an architecture where AI agents ingest potentially untrusted record data and have access to tools (MCP, web search). While this represents a standard attack surface for LLM-based systems, the skill provides the configuration necessary to implement guardrails as a mitigation strategy.
Ingestion points: Records are mapped to text, record, or files fields in SKILL.md.
Boundary markers: None explicitly required in the input mapping, though system instructions are emphasized as the primary behavior definition.
Capability inventory: Support for web_search, mcp (Model Context Protocol) tools, celigo tools, and image_generation across SKILL.md and aiagent.yml.
Sanitization: The skill documents GuardrailImport specifically for filtering PII and moderated content in guardrail.yml.

configuring-ai-agents