Tool Creator

tool-creator

Overview

AnimaWorks tools are categorized into three types:

{data_dir} is typically ~/.animaworks/. Personal and shared tools are re-scanned by ExternalToolDispatcher via refresh_tools and can be hot-reloaded. ToolHandler checks the tool creation permission in permissions.json when writing to tools/*.py with write_memory_file.

Personal and shared tools are invoked via Bash with animaworks-tool <tool> <subcommand> [args]. Schema name format is {tool_name}_{action} (e.g., my_tool + query → my_tool_query).

Procedure

Step 1: Design the Tool

1. Decide the tool name (snake_case, e.g., my_api_tool)
2. Define the schema(s) (operations) to provide
3. Define required parameters

Step 2: Create the Module File

Create a Python file following the template below.

Single-Schema Tool (Simple)

For file my_tool.py, animaworks-tool my_tool action [args] invokes it; schema name my_tool_action is passed to dispatch.

from __future__ import annotations

import logging
from typing import Any

logger = logging.getLogger("animaworks.tools")


def get_tool_schemas() -> list[dict]:
    """Return tool schema(s) (required)."""
    return [
        {
            "name": "my_tool_action",
            "description": "Description of what this tool does",
            "input_schema": {
                "type": "object",
                "properties": {
                    "param1": {
                        "type": "string",
                        "description": "Parameter description",
                    },
                    "param2": {
                        "type": "integer",
                        "description": "Optional parameter",
                        "default": 10,
                    },
                },
                "required": ["param1"],
            },
        }
    ]


def dispatch(name: str, args: dict[str, Any]) -> Any:
    """Execute handling by schema name (recommended)."""
    args.pop("anima_dir", None)  # Injected by framework; unused in this tool
    if name == "my_tool_action":
        return _do_action(
            param1=args["param1"],
            param2=args.get("param2", 10),
        )
    raise ValueError(f"Unknown tool: {name}")


def _do_action(param1: str, param2: int = 10) -> dict[str, Any]:
    """Actual logic implementation."""
    # Implement here
    return {"result": f"Processed {param1} with {param2}"}

Multi-Schema Tool (API Integration, etc.)

Schema names use {tool_name}_{action} format. animaworks-tool myapi query [args] passes myapi_query to dispatch. File name: myapi.py.

from __future__ import annotations

import logging
from typing import Any

logger = logging.getLogger("animaworks.tools")


def get_tool_schemas() -> list[dict]:
    return [
        {
            "name": "myapi_query",
            "description": "Send query to API and get results",
            "input_schema": {
                "type": "object",
                "properties": {
                    "query": {"type": "string", "description": "Search query"},
                    "limit": {"type": "integer", "description": "Max results", "default": 10},
                },
                "required": ["query"],
            },
        },
        {
            "name": "myapi_post",
            "description": "Send data to API",
            "input_schema": {
                "type": "object",
                "properties": {
                    "data": {"type": "string", "description": "Data to send"},
                },
                "required": ["data"],
            },
        },
    ]


class MyAPIClient:
    """API client."""

    def __init__(self) -> None:
        from core.tools._base import get_credential
        self._api_key = get_credential(
            "myapi", "myapi_tool", env_var="MYAPI_KEY",
        )

    def query(self, query: str, limit: int = 10) -> list[dict]:
        import httpx
        resp = httpx.get(
            "https://api.example.com/search",
            params={"q": query, "limit": limit},
            headers={"Authorization": f"Bearer {self._api_key}"},
            timeout=30,
        )
        resp.raise_for_status()
        return resp.json()["results"]

    def post(self, data: str) -> dict:
        import httpx
        resp = httpx.post(
            "https://api.example.com/data",
            json={"data": data},
            headers={"Authorization": f"Bearer {self._api_key}"},
            timeout=30,
        )
        resp.raise_for_status()
        return resp.json()


def dispatch(name: str, args: dict[str, Any]) -> Any:
    args.pop("anima_dir", None)  # Injected by framework; unused in this tool
    client = MyAPIClient()
    if name == "myapi_query":
        return client.query(
            query=args["query"],
            limit=args.get("limit", 10),
        )
    elif name == "myapi_post":
        return client.post(data=args["data"])
    raise ValueError(f"Unknown tool: {name}")

Step 3: Save the File

Save as a personal tool (path in write_memory_file is relative to anima_dir):

write_memory_file(path="tools/my_tool.py", content=<code>)

Writing to tools/ requires personal tool permission in the "Tool creation" section of permissions.json.

Step 4: Enable the Tool

After saving, call refresh_tools for hot reload:

refresh_tools()

The tool becomes available immediately without restarting the session. Personal tools do not need to be listed in permissions.json external_tools; once discovered by refresh_tools, they are callable via Bash with animaworks-tool <tool> <subcommand>.

Step 5: Share (Optional)

To let other Anima use it, share the tool:

share_tool(tool_name="my_tool")

This copies it to ~/.animaworks/common_tools/ and makes it available to all Anima. Sharing requires shared tool permission in permissions.json.

Required Interface

Bash Invocation

Anima invokes personal/shared tools via Bash with animaworks-tool <tool> <subcommand>:

animaworks-tool myapi query "search term" [--limit 10]

schema_name = tool_name + "_" + action is passed to dispatch(name, args). In the example above, name="myapi_query".

Schema Definition Conventions

Both input_schema and parameters are supported and normalized (core/tooling/schemas._normalise_schema).

{
    "name": "tool_action_name",       # snake_case. Format: {tool_name}_{action}
    "description": "1-2 sentence description",  # Used by LLM for tool selection
    "input_schema": {                  # JSON Schema format (parameters also accepted)
        "type": "object",
        "properties": { ... },
        "required": [ ... ],
    },
}

Credential Retrieval (get_credential)

Obtain API keys etc. via get_credential(). Never hardcode.

from core.tools._base import get_credential

api_key = get_credential(
    credential_name="myapi",   # Key in config.json credentials
    tool_name="myapi_tool",   # For error messages
    key_name="api_key",       # Default. Can specify other keys in keys
    env_var="MYAPI_KEY",      # Fallback environment variable
)

Resolution order: config.json → vault.json (encrypted vault) → shared/credentials.json → environment variable. ToolConfigError if none found.

Tool Creation Permission in permissions.json

Add the following to permissions.json for tool creation and sharing:

## Tool creation
- Personal tools: yes
- Shared tools: yes

OK, enabled, or true are also valid instead of yes.

Validation Checklist

• Filename: snake_case, .py extension (e.g., my_tool.py)
• from __future__ import annotations at top of file
• get_tool_schemas() exists and returns a list
• Schema names follow {tool_name}_{action} format (animaworks-tool integration)
• Schema has name, description, input_schema (or parameters)
• dispatch() or same-name function exists
• args.pop("anima_dir", None) in dispatch when passing args to other functions
• Handler exists for all schemas
• Appropriate exceptions raised on error
• Timeout set for external APIs

Security Guidelines

1. Credentials: Obtain via get_credential(). Never hardcode

2. Access control: Do not access other Anima's directories

3. Timeout: Always set timeout for external APIs (recommended: 30 seconds)

4. Logging: Use logging.getLogger("animaworks.tools")

5. Dependencies: Import external libraries inside functions (lazy import)

Mode S (MCP) exposure for core tools

When adding a module under core/tools/, only names listed in core/mcp/server.py _EXPOSED_TOOL_NAMES are exposed to Claude Code via MCP (curated subset). As of 2026-03, that set includes: search_memory, read_memory_file, write_memory_file, archive_memory_file, send_message, post_channel, call_human, delegate_task, submit_tasks, update_task, create_skill. External-service core tools such as Slack, Gmail, and web_search are not on the MCP list; they are reached via use_tool, Bash (animaworks-tool), or skills.

Notes

• Tools are Python code, different from Skills (Markdown procedure documents)
• Tool creation requires personal tools: yes in the "Tool creation" section of permissions.json
• Sharing tools requires shared tools: yes permission
• Created tools are discovered immediately on refresh_tools call (hot reload)
• Personal tools do not need to be in permissions.json external_tools; once discovered, they are available via Bash with animaworks-tool <tool> <subcommand>
• Schema names use {tool_name}_{action} format; keep them unique across tools
• Personal or shared tools with the same name as core tools are shadowed and skipped (core/tools/__init__.py)