Crate nenjo_tool_api 

Shared tool contracts for Nenjo agents, model providers, and runtimes.

This crate owns the common tool API surface used across the Nenjo workspace. It is deliberately independent from the rest of the workspace so model integrations, SDK code, and worker runtimes can agree on tool schemas and execution results without depending on each other.

The main entry points are:

• Tool, the async trait implemented by concrete tool runtimes.
• ToolSpec, the JSON-schema-backed metadata sent to model providers.
• ToolCategory, the side-effect classification used for guidance and filtering.
• ToolCall, ToolResult, and ToolResultMessage, the request and result payloads that flow through tool execution.
• ToolAutonomy and ToolSecurity, the SDK-level policy inputs used when constructing tools.

Example

use async_trait::async_trait;
use serde_json::json;
use nenjo_tool_api::{Tool, ToolCategory, ToolResult};

struct EchoTool;

#[async_trait]
impl Tool for EchoTool {
    fn name(&self) -> &str {
        "echo"
    }

    fn description(&self) -> &str {
        "Echoes a message back to the caller."
    }

    fn parameters_schema(&self) -> serde_json::Value {
        json!({
            "type": "object",
            "properties": {
                "message": { "type": "string" }
            },
            "required": ["message"]
        })
    }

    fn category(&self) -> ToolCategory {
        ToolCategory::Read
    }

    async fn execute(&self, args: serde_json::Value) -> anyhow::Result<ToolResult> {
        Ok(ToolResult {
            success: true,
            output: args["message"].as_str().unwrap_or_default().to_string(),
            error: None,
        })
    }
}

Re-exports

pub use async_ops::AsyncOperationKind;

pub use async_ops::AsyncOperationSignalKind;

pub use async_ops::AsyncOperationStatus;

pub use async_ops::INSPECT_OPERATIONS_TOOL_NAME;

pub use async_ops::InspectOperationsArgs;

pub use async_ops::SEND_OPERATION_INPUT_TOOL_NAME;

pub use async_ops::STOP_OPERATIONS_TOOL_NAME;

pub use async_ops::SendOperationInputArgs;

pub use async_ops::StopOperationsArgs;

pub use async_ops::WAIT_OPERATIONS_TOOL_NAME;

pub use async_ops::WaitOperationsArgs;

pub use async_ops::inspect_operations_parameters_schema;

pub use async_ops::send_operation_input_parameters_schema;

pub use async_ops::stop_operations_parameters_schema;

pub use async_ops::wait_operations_parameters_schema;

Modules

async_ops

Shared contracts for model-visible async operation tools.

Structs

ToolCall

A tool call requested by the LLM.

ToolResult

Result of a tool execution.

ToolResultMessage

A tool result to feed back to the LLM.

ToolSecurity

SDK-level tool construction policy.

ToolSpec

Full specification of a tool for LLM registration.

Enums

ToolAutonomy

High-level autonomy requested while constructing runtime tools.

ToolCategory

Classifies a tool’s side-effect profile for filtering and model guidance.

ToolOrigin

Runtime ownership surface for a tool.

Traits

Tool

Core tool trait for agent capabilities.

Functions

sanitize_tool_name

Sanitize a tool function name to match the strict OpenAI pattern ^[a-zA-Z0-9_-]+$.

sanitize_tool_name_lenient

Light sanitization for lenient providers (Ollama) while preserving dots used in MCP namespaced tool names.