agent 抓了一份 Python 文档，写了三段 list comprehension 示例，然后跑起来。前两段没问题第三段抛出了语法错误。它没有停在那里，而是去读错误信息、找到问题、把代码改了，再跑一次。这次过了，到这一刻"agentic" 这个词才真正落地。
单个工具是噱头，工具集才是真正的系统。其实只要有三个工具就能把 agent 从聊天机器人变成能干活的东西：
fetch_url
读取任意网页，
write_file
把结果落到磁盘，
run_python
直接执行代码。一次对话里，研究、写、测、修可以全部走完。
真正有意思的不是工具本身而是规划，让agent 自己决定："得先研究，再写代码，最后运行验证。" 这个顺序不是写死在代码里的；模型从目标推理出来，自动成形。
下面这段代码是一个完整的多工具 agent，能做网页研究、写并运行 Python，也能保存文件。
import anthropic
import subprocess, sys, tempfile, re
import urllib.request
from pathlib import Path
client = anthropic.Anthropic()
OUTPUT_DIR = Path("agent_output")
OUTPUT_DIR.mkdir(exist_ok=True)
# ── Tool implementations ──────────────────────────────────────────────────
def fetch_url(url: str) -> str:
"""Fetch a webpage and strip HTML - gives the agent access to the web."""
try:
req = urllib.request.Request(url, headers={"User-Agent": "Mozilla/5.0"})
with urllib.request.urlopen(req, timeout=10) as r:
html = r.read().decode("utf-8", errors="ignore")
text = re.sub(r"<[^>]+>", " ", html)
text = re.sub(r"\s+", " ", text).strip()
return text[:2500] # cap to avoid filling the context window
except Exception as e:
return f"Error fetching {url}: {e}"
def write_file(filename: str, content: str) -> str:
"""Save content to the output directory."""
path = OUTPUT_DIR / filename
path.write_text(content)
return f"Saved {len(content)} chars to {path}"
def run_python(code: str) -> str:
"""Execute Python code in a subprocess. Returns stdout and stderr."""
with tempfile.NamedTemporaryFile(mode="w", suffix=".py", delete=False) as f:
f.write(code)
tmp = f.name
try:
result = subprocess.run(
[sys.executable, tmp],
capture_output=True, text=True, timeout=15
)
out = result.stdout or ""
err = result.stderr or ""
return (out + ("\nERROR:\n" + err if err else "")).strip() or "Ran with no output."
except subprocess.TimeoutExpired:
return "Error: timed out after 15 seconds."
finally:
Path(tmp).unlink(missing_ok=True)
TOOLS = [
{
"name": "fetch_url",
"description": "Fetch the text content of any webpage. Use for research.",
"input_schema": {"type":"object","properties":{"url":{"type":"string"}},"required":["url"]}
},
{
"name": "write_file",
"description": "Save text to a file in the output directory.",
"input_schema": {"type":"object","properties":{"filename":{"type":"string"},"content":{"type":"string"}},"required":["filename","content"]}
},
{
"name": "run_python",
"description": "Execute Python code and return the output. Use to test code you write.",
"input_schema": {"type":"object","properties":{"code":{"type":"string"}},"required":["code"]}
}
]
def execute_tool(name, inputs):
if name == "fetch_url": return fetch_url(inputs["url"])
if name == "write_file": return write_file(inputs["filename"], inputs["content"])
if name == "run_python": return run_python(inputs["code"])
return f"Unknown tool: {name}"
# ── Agent loop ─────────────────────────────────────────────────────────────
def run_agent(task: str):
print(f"\nTask: {task}\n" + "="*50)
messages = [{"role": "user", "content": task}]
system = """You are a research and coding agent. When given a task:
1. Research first if needed (fetch_url)
2. Write any code required
3. Run the code to verify it works - fix it if it errors
4. Save the final result to a file
Be methodical. Show your reasoning before each tool call."""
for i in range(15):
response = client.messages.create(
model="claude-haiku-4-5-20251001",
max_tokens=1024,
system=system,
tools=TOOLS,
messages=messages
)
if response.stop_reason == "end_turn":
answer = next((b.text for b in response.content if hasattr(b,"text")), "Done.")
print(f"\nDone: {answer}")
return answer
tool_results = []
for block in response.content:
if block.type == "tool_use":
print(f"\n [{block.name}]")
result = execute_tool(block.name, block.input)
# Print a preview so you can follow along
print(f" → {result[:200]}{'...' if len(result)>200 else ''}")
tool_results.append({
"type": "tool_result",
"tool_use_id": block.id,
"content": result
})
messages.append({"role":"assistant","content":response.content})
messages.append({"role":"user","content":tool_results})
# Run it
run_agent(
"Write three Python list comprehension examples - basic, filtered, and nested. "
"Run each one to verify it works. Save a cheat sheet to 'list_comprehensions.md'."
)
fetch_url
用正则把 HTML 标签去掉，方法粗糙，但拿到可读文本是够用的；timeout 用来防止 agent 在慢站上卡住。
run_python
走的是临时文件加 subprocess 隔离，比
exec()
安全一些，但放到生产环境仍有风险，timeout 同样用来防死循环。
system prompt 给 agent 划了一条工作路径：研究、写代码、测试、保存。它会自然按这个节奏走。
多工具带来的真正变化
agent 本身不会写代码，也不会上网。它只会用工具。智能落在规划上。
丢一个复杂任务进去，看它的执行序列：抓 URL → 综合信息 → 写代码 → 运行 → 读错误 → 改 bug → 再跑 → 保存结果。这条工作流不是你编排出来的，是模型自己推出来的。
这就是为什么工具设计比挑模型更关键。一个设计得当的工具，能让 agent 的能力上一个台阶。
最值得学习的一点：agent 会调试自己写的代码。看到报错，理解原因，动手修。这不是预设脚本，是循环加工具自然涌现出来的行为。
改进提高
加一个
read_file
工具，从输出目录读取已有文件。然后把这个任务交给 agent："读取你之前生成的 list comprehensions cheat sheet，再加两个进阶示例，把文件更新一遍。"
看它怎么走：读 → 分析 → 扩展 → 保存。同一套规划逻辑，落到完全不同的工作流上。
https://avoid.overfit.cn/post/061983d61b58422e8436c0eece200573
by TechnoFin
