Loading ToolBench
B2B Lead Intelligence API — search, enrich & score leads from 10M+ global companies database. All report generation runs locally. No data leaves the user's machine.
Golddigger has 4 tools with complete input schemas and descriptions, but quality is uneven. Tool names lack action verbs (health_check is acceptable, but search_leads, score_lead, generate_report are weak). Descriptions are present but verbose and include marketing language rather than LLM-optimized guidance. Parameters have types and descriptions, but several lack constraints (e.g., limit unbounded, output_dir unconstrained). Output schemas are not documented. Error handling is minimal, most tools return generic error dicts without recovery guidance. The server exposes API keys as parameters (critical security issue), violating secret-injection pattern. Overall: functional but below production baseline.
生成 Golddigger 风格的 B2B 客户画像雷达图 Word 报告。 数据主权声明： - 雷达图生成：matplotlib 本地渲染，PNG 不上传任何服务器 - Word 报告：python-docx 本地生成，.docx 存储在客户本地路径 - 不调用任何外部存储服务 ⚠️ Token 费用：搜索调用使用客户提供的 api_key，费用由客户自行承担。 雷达图生成和报告排版完全免费（无 API 调用）。
检查 Golddigger API 连接状态和配额。
对采购商进行六维质量评分（本地计算，不调用外部 API）。 数据主权：评分逻辑完全在客户本地运行，原始数据不离开用户机器。
搜索 Golddigger B2B 客户数据库。 ⚠️ 注意：免费版 db_rows=0，搜索永远返回空结果。 客户需升级 SUPER 套餐（¥499+/月）才有真实数据。
API keys exposed as tool parameters (api_key, host in all tools). Credentials must use server-side secret injection via environment variables. Agent traces log all parameters, secrets leak into logs and prompt history.
Tool names lack action verbs. 'search_leads' is acceptable, but 'score_lead' (should be 'rate_lead' or 'evaluate_lead') and 'generate_report' (should be 'create_report') are vague. LLMs infer intent from names, weak names cause misselection.
Output schemas not documented. Descriptions mention return structure (e.g., 'Returns: {"status": "ok"|"error", ...}') but are embedded in docstrings, not in formal schema. LLMs cannot parse unstructured docstring returns, they need explicit output type definitions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
Numeric parameter 'limit' in search_leads has no bounds. Unbounded integers let LLMs pass absurd values (e.g., limit=999999) that break APIs or cause timeouts. Should declare min=1, max=100.
Error handling is minimal. Tools return generic error dicts (e.g., {"error": str(e), "leads": [], "count": 0}) without recovery guidance. LLMs cannot determine if errors are retryable, user-fixable, or fatal.