agent-tool-integration

What problem does it solve?

Bridge the gap between agents and external tools by safely exposing read-only capabilities through system-prompt injection. This approach reduces the agency gap and enables informed decision-making without modifying core LLM invocation logic.

Core Features & Use Cases

• Role-based tool allowlists to limit which tools are accessible per agent role.
• Tool metadata injection into the system prompt to describe available tools, side effects, and example inputs.
• Tool invocation via portable XML tags to signal tool usage across providers, with deterministic handling.
• Local tool catalog discovery to enable agents to discover available capabilities without hardcoding tool names.
• Use cases include fact-checking, data enrichment, and safe publishing workflows with read-only data access.

Quick Start

Enable the agent-tool-integration pattern by wiring the local tool catalog into the system prompt and start emitting <tool_use> tags for safe, read-only tools.