👁️ 代码协作训练

👁️ 模块一：读代码训练

拿到一段 Agent 代码，先别跑。用笔标注每个变量的输入输出、数据怎么流转、哪里可能炸。

import json

def run_agent(task, memory=[]):
    # 简单的 ReAct 循环
    prompt = f"任务：{task}\n历史：{memory}"
    resp = call_llm(prompt)
    thought = resp["thought"]
    action = resp["action"]
    if action == "search":
        result = search_web(resp["query"])
        memory.append(str(result))
    elif action == "done":
        return resp["answer"]
    return run_agent(task, memory)

✂️ 模块二：拆函数训练

一段 25 行能跑但难读的脚本。目标：拆成职责单一的小函数，并画出数据流箭头。

import csv, json

def process(path):
    rows = []
    with open(path) as f:
        for r in csv.DictReader(f):
            if r["age"].isdigit() and int(r["age"]) > 0:
                rows.append({"name": r["name"].strip(),
                             "age": int(r["age"])})
    total = sum(x["age"] for x in rows)
    avg = total / len(rows) if rows else 0
    adults = [x for x in rows if x["age"] >= 18]
    report = {
        "count": len(rows),
        "avg_age": round(avg, 1),
        "adults": len(adults),
        "names": [x["name"] for x in adults]
    }
    out = path.replace(".csv", "_report.json")
    with open(out, "w") as f:
        json.dump(report, f, ensure_ascii=False)
    return out

📦 模块三：调包训练

面对需求，三步走：选对库 → 查准方法签名 → 验证 AI 给的代码是否真的对。