Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add Kokxi/qa-test-skills --skill qa-ai-output-critiquegit clone --depth 1 https://github.com/Kokxi/qa-test-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/kokxi/qa-test-skills/qa-ai-output-critique)<a href="https://agentmods.dev/skills/kokxi/qa-test-skills/qa-ai-output-critique"><img src="https://agentmods.dev/badge/skills/kokxi/qa-test-skills/qa-ai-output-critique/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/kokxi/qa-test-skills/qa-ai-output-critique"><img src="https://agentmods.dev/badge/skills/kokxi/qa-test-skills/qa-ai-output-critique.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00181 | $0.03531 |
| Opus 5 | $0.00090 | $0.01766 |
| Sonnet 5 | $0.00036 | $0.00706 |
| Haiku 4.5 | $0.00018 | $0.00353 |
Grade A, and why
qa-ai-output-critique scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 294 lines — stays where its author put it; the contents beside it link to each section on GitHub.
⚠️ 安全警告：本技能的示例可能涉及测试用例整理、合并或删除建议。 实际使用时请勿直接执行批量删除操作，先备份原数据并确认非关键用例。 本技能仅在 workspace/ 输出评估文件，不持久化、不外传、不跨会话复用。
AI 输出评审
使用逻辑
本技能分为四个阶段，按顺序执行：
阶段一：评审框架（八维评审法）
│
│ 使用八维评审法对AI输出进行系统打分
│ 输出：评审报告（评分 + 问题清单）
│
▼
阶段二：报告模板（输出格式）
│
│ 按标准格式输出评审结果
│ 输出：结构化评审报告
│
▼
阶段三：追问问法（深入挖掘）
│
│ 使用追问技巧引导发现问题
│ 输出：更深层的质量洞察
│
▼
阶段四：辅助工具（假设挖掘/反驳机制/投入产出评估）
│
│ 补充评审维度，提升评审深度
│ 输出：完整评审结论
关键点：阶段三的"追问问法"是在评审报告输出后使用的，用于引导发现更深层的问题，不是独立的评审体系。
输入要求
必填输入
- AI生成的测试用例
推荐输入（提供则评审更准确）
- 场景树（来自qa-scenario-tree）：用于完整性评审，判断场景覆盖率
- 风险清单（来自qa-risk-intuition）：用于风险覆盖评审，判断高风险区域是否深测
- 需求ID列表：用于追溯性评审，判断用例是否可追溯
降级策略
如果缺少推荐输入：
| 缺少的输入 | 评审策略 | 降级处理 |
|---|---|---|
| 场景树 | 完整性评审 | 基于通用场景清单（主流程/分支/异常/边界） |
| 风险清单 | 风险覆盖评审 | 基于默认风险类型（资金/安全/并发/数据一致性） |
| 需求ID列表 | 追溯性评审 | 跳过，标记为"需补充追溯信息" |
评审模式
模式A：完整评审（有场景树和风险清单）
- 八维评审全部启用
- 输出完整评审报告
- 适用场景：工作流中的标准评审
模式B：快速评审（无上游数据）
- 启用六维评审（跳过追溯性，简化完整性）
- 输出简化评审报告
- 标注"建议补充场景树后重新评审"
- 适用场景：独立使用、快速检查
模式选择规则
有场景树 AND 有风险清单 → 模式A（完整评审）
否则 → 模式B（快速评审）
核心原则
AI输出看起来都对，但专家能看出哪里不够。
八维评审法速查
评分标准：每维度10分，总分≥64分为合格。详见 references/review-dimensions.md。
| 维度 | 评分核心 | 核心关注 |
|---|---|---|
| 完整性 | 场景覆盖是否完整 | 主路径+分支+异常+边界 |
| 正确性 | 业务规则和预期是否正确 | 内容正确性 |
| 可执行性 | 步骤是否清晰可执行 | 可操作性 |
| 风险覆盖 | 高风险区域是否深测 | 资金/安全/并发/数据 |
| 规范性 | 格式是否符合标准 | 编号/表格/字段 |
| 追溯性 | 需求/风险ID是否完整 | 需求追溯链 |
| 一致性 | 用例间是否自相矛盾 | 前置与步骤匹配 |
| 冗余度 | 是否有无价值用例 | 重复/低价值用例 |
每维度的详细评分标准、评审清单和追问问法参见
references/review-dimensions.md。
中文/国产模型特有评审点
国产模型（DeepSeek/通义/文心/豆包等）输出除八维通用评审外，需额外检查以下特有风险：
| 评审点 | 典型表现 | 处理 |
|---|---|---|
| 中文语义漂移 | 用例标题与步骤语义不一致（如标题"验证登录"步骤却在测注册） | 用"步骤是否支撑标题"反向核查，漂移即降分 |
| 数字/单位幻觉 | 超时值、并发数、金额单位凭空编造（如"P95<100ms"无依据） | 核对来源，无依据标注"MISSING：需基准数据" |
| 政策/合规表述 | 用例预期结果含不合规表述（如支付、个人信息处理） | 标注合规风险，提示对照监管要求 |
| 格式不稳定 | 表格列错位、编号断号（TC_001→TC_003）、中文标点混入 | 按规范性维度检查编号连续性 |
| 过度泛化 | 用例写得"像测试指南"而非"可执行用例"（步骤不可操作） | 按可执行性维度扣分，要求补具体步骤 |
| 伪正确性 | 预期结果"返回正确"但未定义什么算正确 | 要求预期结果可验证（具体状态码/字段值） |
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 7d ago Changed 90ea3e4d7b7c
- 12d ago First seen · 294 lines · 181 tokens per session scan A 73310e9afc33
qa-ai-output-critique is a skill published in the GitHub repository Kokxi/qa-test-skills (27 stars, last pushed 9d ago), licensed MIT. It adds 181 tokens to every session and 3,531 once invoked, about $0.0009 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
code-review
Reviews Spec Kit code changes for positive and negative test coverage, regression evidence for bug fixes, and consistent repository terminology. Use when reviewing a diff or pull request. Do not use for implementing changes or posting GitHub review actions.
agent-harness-fault-injection
Use when an agent workflow needs deterministic recovery evidence for sandbox, MCP/tool, worker, checkpoint, memory, or orchestration failures.
octocode-benchmark
Use when planning, running, grading, or reporting the by-hand Octocode research benchmark — pairwise matchups (Octocode anchor vs one baseline: gh+RTK, gh+Headroom, or plain gh) over markdown questions, with a fresh isolated runner agent per (question, arm, pass), one blind judge per question grading two answers X/Y…
octocode-graph-eval
Use when you need a measurable keep/discard loop — goal→KPI, baseline vs target, held-out checks, eval suites, or don't-stop-till-done against a runnable sensor. Not for ordinary ship checks where 'tests passed' is enough.
plugin-test
A testing guide for Zhin.js plugins using Vitest, a JavaScript and TypeScript testing framework. It focuses on checking command and tool behavior, ordinary business logic, and the plugin package’s required structure.
modernize-test-starter
Modernize QUnit unit tests and OPA5 integration tests to the UI5 Test Starter concept. Use this skill when: The linter reports prefer-test-starter for .qunit.html or .qunit.js files Test HTML files use manual sap-ui-core.js bootstrapping instead of Test Starter's runTest.js/createSuite.js Test JS files use…