Rohan03's picture
Track 1: purpose_agent/tools.py
40b8915 verified
raw
history blame
18.9 kB
"""
Tool Framework — Structured tool definitions with schema, validation, and retry.
Tools are the agent's interface to the world. This module provides:
- Tool base class with JSON schema, input validation, retry logic
- Built-in tools (search, calculator, python_exec, file_read)
- Tool registry with semantic search (for Tool RAG with SLMs)
- Automatic tool documentation generation for prompts
SLM optimization: Tools are formatted with minimal token overhead.
For small context windows, use ToolRegistry.get_relevant_tools()
to retrieve only the k most relevant tools (TinyAgent pattern).
"""
from __future__ import annotations
import json
import logging
import math
import os
import re
import sys
import time
import traceback
from abc import ABC, abstractmethod
from dataclasses import dataclass, field
from typing import Any, Callable
logger = logging.getLogger(__name__)
# ---------------------------------------------------------------------------
# Tool Base Class
# ---------------------------------------------------------------------------
@dataclass
class ToolResult:
"""Result of a tool execution."""
output: str
success: bool = True
error: str | None = None
execution_time_s: float = 0.0
metadata: dict[str, Any] = field(default_factory=dict)
class Tool(ABC):
"""
Abstract tool that an Agent can invoke.
Every tool has:
- name: Unique identifier
- description: What it does (used in agent prompts)
- parameters: JSON schema of expected inputs
- execute(): The actual implementation
Built-in retry logic, input validation, and timeout handling.
Example:
class SearchTool(Tool):
name = "web_search"
description = "Search the web for information"
parameters = {
"type": "object",
"properties": {
"query": {"type": "string", "description": "Search query"}
},
"required": ["query"]
}
def execute(self, query: str) -> str:
return do_search(query)
"""
name: str = "unnamed_tool"
description: str = "No description"
parameters: dict[str, Any] = {}
max_retries: int = 2
timeout_seconds: float = 30.0
@abstractmethod
def execute(self, **kwargs) -> str:
"""Execute the tool with the given parameters. Return a string result."""
...
def run(self, **kwargs) -> ToolResult:
"""
Run the tool with validation, retry, and error handling.
This is the method the Orchestrator calls — it wraps execute()
with production safeguards.
"""
# Validate inputs against schema
validation_error = self._validate_inputs(kwargs)
if validation_error:
return ToolResult(
output="", success=False,
error=f"Input validation failed: {validation_error}",
)
# Retry loop
last_error = None
for attempt in range(self.max_retries + 1):
start = time.time()
try:
result = self.execute(**kwargs)
return ToolResult(
output=str(result),
success=True,
execution_time_s=time.time() - start,
metadata={"attempt": attempt + 1},
)
except Exception as e:
last_error = e
elapsed = time.time() - start
logger.warning(
f"Tool '{self.name}' failed (attempt {attempt + 1}/{self.max_retries + 1}): {e}"
)
if attempt < self.max_retries:
time.sleep(0.5 * (attempt + 1)) # Exponential backoff
return ToolResult(
output="",
success=False,
error=f"Tool '{self.name}' failed after {self.max_retries + 1} attempts: {last_error}",
execution_time_s=time.time() - start,
)
def _validate_inputs(self, kwargs: dict) -> str | None:
"""Validate inputs against the JSON schema. Returns error string or None."""
if not self.parameters:
return None
required = self.parameters.get("required", [])
properties = self.parameters.get("properties", {})
for req in required:
if req not in kwargs:
return f"Missing required parameter: '{req}'"
for key, value in kwargs.items():
if key in properties:
expected_type = properties[key].get("type")
if expected_type == "string" and not isinstance(value, str):
return f"Parameter '{key}' should be string, got {type(value).__name__}"
elif expected_type == "integer" and not isinstance(value, int):
return f"Parameter '{key}' should be integer, got {type(value).__name__}"
elif expected_type == "number" and not isinstance(value, (int, float)):
return f"Parameter '{key}' should be number, got {type(value).__name__}"
return None
def to_schema(self) -> dict[str, Any]:
"""Return OpenAI-compatible tool schema."""
return {
"type": "function",
"function": {
"name": self.name,
"description": self.description,
"parameters": self.parameters or {"type": "object", "properties": {}},
},
}
def to_prompt(self, compact: bool = False) -> str:
"""
Format tool for inclusion in agent prompts.
compact=True: Minimal format for SLMs (fewer tokens)
compact=False: Full description with parameter details
"""
if compact:
params = ", ".join(
f"{k}: {v.get('type', 'any')}"
for k, v in self.parameters.get("properties", {}).items()
)
return f"- {self.name}({params}) — {self.description}"
lines = [f"### {self.name}"]
lines.append(f" {self.description}")
if self.parameters.get("properties"):
lines.append(" Parameters:")
for pname, pinfo in self.parameters["properties"].items():
req = "REQUIRED" if pname in self.parameters.get("required", []) else "optional"
lines.append(f" - {pname} ({pinfo.get('type', 'any')}, {req}): {pinfo.get('description', '')}")
return "\n".join(lines)
# ---------------------------------------------------------------------------
# Function Tool — Create tools from plain functions
# ---------------------------------------------------------------------------
class FunctionTool(Tool):
"""
Create a Tool from a plain Python function.
Usage:
def search(query: str) -> str:
'''Search the web for information.'''
return requests.get(f"https://api.search.com?q={query}").text
tool = FunctionTool.from_function(search)
# or
tool = FunctionTool(
name="search",
description="Search the web",
parameters={"type": "object", "properties": {"query": {"type": "string"}}, "required": ["query"]},
fn=search,
)
"""
def __init__(
self,
name: str,
description: str,
parameters: dict[str, Any],
fn: Callable[..., str],
**kwargs,
):
self.name = name
self.description = description
self.parameters = parameters
self._fn = fn
super().__init__(**kwargs)
def execute(self, **kwargs) -> str:
return str(self._fn(**kwargs))
@classmethod
def from_function(cls, fn: Callable, name: str | None = None, description: str | None = None) -> "FunctionTool":
"""Auto-create a Tool from a function's signature and docstring."""
import inspect
func_name = name or fn.__name__
func_desc = description or fn.__doc__ or f"Execute {func_name}"
func_desc = func_desc.strip().split("\n")[0] # First line only
sig = inspect.signature(fn)
properties = {}
required = []
type_map = {str: "string", int: "integer", float: "number", bool: "boolean"}
for pname, param in sig.parameters.items():
ptype = "string"
if param.annotation != inspect.Parameter.empty:
ptype = type_map.get(param.annotation, "string")
properties[pname] = {"type": ptype, "description": f"Parameter {pname}"}
if param.default == inspect.Parameter.empty:
required.append(pname)
return cls(
name=func_name,
description=func_desc,
parameters={"type": "object", "properties": properties, "required": required},
fn=fn,
)
# ---------------------------------------------------------------------------
# Built-in Tools
# ---------------------------------------------------------------------------
class CalculatorTool(Tool):
"""Safe math expression evaluator — no eval(), no arbitrary code."""
name = "calculator"
description = "Evaluate a mathematical expression. Supports +, -, *, /, **, sqrt, sin, cos, abs."
parameters = {
"type": "object",
"properties": {
"expression": {"type": "string", "description": "Math expression to evaluate (e.g. '2 + 3 * 4')"}
},
"required": ["expression"],
}
# Whitelist of safe tokens
_SAFE_PATTERN = re.compile(
r'^[\d\s+\-*/().,%e]+$|'
r'(abs|round|min|max|sqrt|sin|cos|tan|log|pi)\b'
)
def execute(self, expression: str) -> str:
import ast
import operator
# Only allow safe characters and function names
cleaned = expression.replace("^", "**").strip()
# Validate: reject anything with letters that aren't known functions
tokens = re.sub(r'(abs|round|min|max|sqrt|sin|cos|tan|log|pi|e)\b', '', cleaned)
if re.search(r'[a-zA-Z_]', tokens):
return f"Error: expression contains disallowed characters: '{expression}'"
allowed = {
"abs": abs, "round": round, "min": min, "max": max,
"sqrt": math.sqrt, "sin": math.sin, "cos": math.cos,
"tan": math.tan, "log": math.log, "pi": math.pi, "e": math.e,
}
try:
# Use compile + eval with empty builtins — no code execution
code = compile(cleaned, "<calc>", "eval")
# Verify AST contains only safe nodes
tree = ast.parse(cleaned, mode="eval")
for node in ast.walk(tree):
if isinstance(node, (ast.Call,)):
if isinstance(node.func, ast.Name) and node.func.id not in allowed:
return f"Error: function '{node.func.id}' not allowed"
result = eval(code, {"__builtins__": {}}, allowed)
return str(result)
except Exception as e:
return f"Error evaluating '{expression}': {e}"
class PythonExecTool(Tool):
"""Execute Python code in a subprocess with timeout and temp directory."""
name = "python_exec"
description = "Execute Python code and return the output. Use print() to output results."
parameters = {
"type": "object",
"properties": {
"code": {"type": "string", "description": "Python code to execute"}
},
"required": ["code"],
}
timeout_seconds: float = 10.0
def execute(self, code: str) -> str:
import subprocess
import tempfile
import os
# Write code to temp file in isolated temp directory
with tempfile.TemporaryDirectory(prefix="pa_exec_") as tmpdir:
script = os.path.join(tmpdir, "script.py")
with open(script, "w") as f:
f.write(code)
try:
result = subprocess.run(
[sys.executable, script],
capture_output=True, text=True,
timeout=self.timeout_seconds,
cwd=tmpdir,
env={**os.environ, "HOME": tmpdir}, # isolate HOME
)
output = result.stdout
if result.stderr:
output += f"\nSTDERR:\n{result.stderr}"
if result.returncode != 0:
output += f"\n(exit code {result.returncode})"
return output or "(no output)"
except subprocess.TimeoutExpired:
return f"Error: execution timed out after {self.timeout_seconds}s"
except Exception as e:
return f"Error: {e}"
class ReadFileTool(Tool):
"""Read a local file — sandboxed to allowed root directory."""
name = "read_file"
description = "Read the contents of a file at the given path."
parameters = {
"type": "object",
"properties": {
"path": {"type": "string", "description": "File path to read"}
},
"required": ["path"],
}
def __init__(self, sandbox_root: str = ".", **kwargs):
self.sandbox_root = os.path.abspath(sandbox_root)
super().__init__(**kwargs)
def execute(self, path: str) -> str:
import os
resolved = os.path.abspath(path)
if not resolved.startswith(self.sandbox_root):
return f"Error: path '{path}' is outside sandbox root '{self.sandbox_root}'"
try:
with open(resolved, "r") as f:
content = f.read()
if len(content) > 10000:
return content[:10000] + f"\n...[truncated, {len(content)} chars total]"
return content
except Exception as e:
return f"Error reading '{path}': {e}"
class WriteFileTool(Tool):
"""Write content to a local file — sandboxed to allowed root directory."""
name = "write_file"
description = "Write content to a file. Creates the file if it doesn't exist."
parameters = {
"type": "object",
"properties": {
"path": {"type": "string", "description": "File path to write"},
"content": {"type": "string", "description": "Content to write"},
},
"required": ["path", "content"],
}
def __init__(self, sandbox_root: str = ".", **kwargs):
self.sandbox_root = os.path.abspath(sandbox_root)
super().__init__(**kwargs)
def execute(self, path: str, content: str) -> str:
import os
resolved = os.path.abspath(path)
if not resolved.startswith(self.sandbox_root):
return f"Error: path '{path}' is outside sandbox root '{self.sandbox_root}'"
try:
os.makedirs(os.path.dirname(resolved) or ".", exist_ok=True)
with open(resolved, "w") as f:
f.write(content)
return f"Written {len(content)} chars to {path}"
except Exception as e:
return f"Error writing '{path}': {e}"
# ---------------------------------------------------------------------------
# Tool Registry — with semantic retrieval for SLMs (Tool RAG)
# ---------------------------------------------------------------------------
class ToolRegistry:
"""
Registry of available tools with semantic retrieval.
For SLMs with small context windows, you can't list all tools in the prompt.
Instead, use get_relevant_tools() to retrieve only the k most relevant
tools for the current task (TinyAgent pattern, arxiv:2409.00608).
Usage:
registry = ToolRegistry()
registry.register(CalculatorTool())
registry.register(SearchTool())
registry.register(PythonExecTool())
# Get all tools (for LLMs with large context)
all_tools = registry.get_all()
# Get top-k relevant tools (for SLMs)
relevant = registry.get_relevant_tools("calculate 2+2", top_k=3)
"""
def __init__(self):
self._tools: dict[str, Tool] = {}
self._embeddings: dict[str, list[float]] = {}
def register(self, tool: Tool) -> "ToolRegistry":
"""Register a tool."""
self._tools[tool.name] = tool
# Compute embedding for Tool RAG
text = f"{tool.name} {tool.description}"
self._embeddings[tool.name] = self._embed(text)
return self
def get(self, name: str) -> Tool | None:
return self._tools.get(name)
def get_all(self) -> list[Tool]:
return list(self._tools.values())
def execute(self, name: str, **kwargs) -> ToolResult:
"""Execute a tool by name."""
tool = self._tools.get(name)
if not tool:
return ToolResult(output="", success=False, error=f"Unknown tool: '{name}'")
return tool.run(**kwargs)
def get_relevant_tools(self, query: str, top_k: int = 5) -> list[Tool]:
"""
Retrieve the k most relevant tools for a query.
Uses lightweight trigram embedding + cosine similarity (same as
ExperienceReplay). For production, swap in sentence-transformers.
"""
if len(self._tools) <= top_k:
return list(self._tools.values())
query_emb = self._embed(query)
scored = []
for name, emb in self._embeddings.items():
sim = self._cosine_sim(query_emb, emb)
scored.append((sim, name))
scored.sort(key=lambda x: -x[0])
return [self._tools[name] for _, name in scored[:top_k]]
def format_for_prompt(self, tools: list[Tool] | None = None, compact: bool = False) -> str:
"""Format tools for inclusion in agent prompts."""
tools = tools or list(self._tools.values())
return "\n".join(t.to_prompt(compact=compact) for t in tools)
def to_schemas(self, tools: list[Tool] | None = None) -> list[dict]:
"""Get OpenAI-compatible tool schemas."""
tools = tools or list(self._tools.values())
return [t.to_schema() for t in tools]
@staticmethod
def _embed(text: str) -> list[float]:
"""Lightweight embedding (same as ExperienceReplay)."""
dim = 64
vec = [0.0] * dim
text_lower = text.lower()
for i in range(len(text_lower) - 2):
trigram = text_lower[i:i + 3]
h = hash(trigram) % dim
vec[h] += 1.0
magnitude = math.sqrt(sum(x * x for x in vec))
if magnitude > 0:
vec = [x / magnitude for x in vec]
return vec
@staticmethod
def _cosine_sim(a: list[float], b: list[float]) -> float:
if not a or not b or len(a) != len(b):
return 0.0
dot = sum(x * y for x, y in zip(a, b))
return dot # Vectors are already normalized