Loading ToolBench
Universal QA Skills - Model Context Protocol (MCP) Server. Exposes 40 QA subdivision domain skills as native Tools for Claude Code, Cursor, Antigravity and other AI Agents. Pure Python 3.9+ standard library implementation with zero external dependencies, supporting JSON-RPC 2.0 stdio protocol.
Server has 6 tools with explicit schemas and descriptions, but quality is inconsistent. Tool names lack clear action verbs (list_qa_skills is acceptable, but run_quality_gate, run_tia_analysis, run_red_blue_swarm are vague). Descriptions are present but brief (avg ~80 chars, below 194-char baseline). Schemas are defined but lack depth: most parameters have minimal descriptions, no enums for constrained inputs, no output schemas documented. Error handling is minimal, responses lack recovery guidance. No tool annotations (readOnlyHint, destructiveHint). Chinese descriptions reduce LLM clarity for English-speaking agents.
动态生成合规脱敏的测试数据 (合规中国手机号/身份证/姓名/邮箱/UUID)
获取当前仓库中全部 40 个细分 QA 技能的分类列表与能力定义
运行质量红线门禁评估，检查指定产物目录中的 signals 信号是否 blocking
启动红蓝对抗自主进化测试，红队生成变异攻击载荷，蓝队生成加固防御用例与补丁
运行 TIA 精准测试分析，根据 Git 变更计算受影响的测试用例文件
对输入的文本或测试数据进行主动脱敏清洗 (手机号/身份证/银行卡号自动掩码打码)
Vague action verbs in tool names. 'run_quality_gate', 'run_tia_analysis', 'run_red_blue_swarm' use generic 'run', LLMs cannot infer what these tools do from names alone. Should be 'check_quality_gate', 'analyze_test_impact', 'generate_adversarial_tests'.
Descriptions are in Chinese and too brief (avg 80 chars vs 194-char baseline). LLMs trained primarily on English struggle with Chinese descriptions. Descriptions lack WHEN/WHY context. E.g., 'run_quality_gate' does not explain when to call it vs other tools.
No output schemas documented. Callers cannot predict response structure. E.g., 'run_red_blue_swarm' returns 'attacks_generated', 'hardened_cases', 'patch_suggestions', but LLMs have no schema to validate or plan downstream calls.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | <=2025-11-25 | v2 |
Parameter descriptions are minimal or missing. 'category' in list_qa_skills has a description, but 'signals_dir' in run_quality_gate lacks format guidance (is it relative? absolute? must it exist?). No constraints on 'count' in generate_test_data (max 100 is enforced in code but not documented).
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All tools are marked READ_ONLY in metadata but not in schema. 'run_red_blue_swarm' may have side effects (generates test cases), should be marked destructiveHint or clarified.