评估你的
AI Agent

from agentforge import Agent

agent = Agent(agent_id="pm_agent_v1")
tasks = ["上线推荐系统（v1）","优化搜索排序（v2）","新增画像模块（v3）"]

for task in tasks:
    output = run_your_agent(task)
    profile = agent.evaluate(output=output)
    print(f"得分: {profile.overall_score} | {profile.tier_label}")

from agentforge import get_evaluation_history

history = get_evaluation_history("pm_agent_v1")
scores = [h["overall_score"] for h in history]
print(f"平均: {sum(scores)/len(scores):.1f} | 最高: {max(scores):.1f}")

name: AgentForge Evaluation
on: [push, pull_request]
jobs:
  evaluate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Run AgentForge
        env:
          AGENTFORGE_API_KEY: ${{ secrets.AGENTFORGE_API_KEY }}
        run: |
          pip install agentforge
          python -c "
from agentforge import Agent
a=Agent(agent_id='ci',api_key='$AGENTFORGE_API_KEY')
r=a.evaluate_task('{{ github.event.head_commit.message }}')
assert r.overall_score>=70
print(f'Score: {r.overall_score}')"

# 低于以下分数 → CI 失败
THRESHOLD = {
    "overall_score":      70,
    "safety_alignment":   65,
    "hallucination_rate": 60,
}

for dim, min_score in THRESHOLD.items():
    if dim in r.radar and r.radar[dim] < min_score:
        raise AssertionError(f"{dim}: {r.radar[dim]} < {min_score}")

pip install agentforge

from agentforge import Agent
agent = Agent(agent_id="my_pm", auto_trace_llm=True)
result = agent.evaluate_task("用户要求一个月内上线推荐系统")
result.print_summary()

from agentforge import agentforge_evaluate

@agentforge_evaluate(agent_id="my_agent", evaluation_mode="deep")
def my_agent(task: str) -> str:
    return run_existing_agent(task)

profile = my_agent("完成XX任务")

from agentforge import Agent
agent = Agent(agent_id="my_pm")

with agent.trace_tool("jira_client", issue_id="PROJ-1"):
    result = jira.get("PROJ-1")

profile = agent.evaluate(output="任务计划：1.数据 2.模型 3.部署")
profile.print_summary()