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 liucongg/liucong-skills --skill liucong-model-evalgit clone --depth 1 https://github.com/liucongg/liucong-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/liucongg/liucong-skills/liucong-model-eval)<a href="https://agentmods.dev/skills/liucongg/liucong-skills/liucong-model-eval"><img src="https://agentmods.dev/badge/skills/liucongg/liucong-skills/liucong-model-eval/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/liucongg/liucong-skills/liucong-model-eval"><img src="https://agentmods.dev/badge/skills/liucongg/liucong-skills/liucong-model-eval.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00085 | $0.01710 |
| Opus 5 | $0.00043 | $0.00855 |
| Sonnet 5 | $0.00017 | $0.00342 |
| Haiku 4.5 | $0.00009 | $0.00171 |
Grade A, and why
liucong-model-eval 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 3d 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 — 63 lines — stays where its author put it; the contents beside it link to each section on GitHub.
刘聪模型测评
用固定题目测试真实接入的模型，区分模型回答、生成状态和实际验收。调度 Agent 不替被测模型答题，也不把自己修好的页面记成模型首轮成绩。本包按刘聪式实测方法组织，内置题是原创演示题，不是刘聪完整私有题库或官方榜单。
先识别当前情况
- 第一次使用 / 缺软件 / 缺 Key：读 初始化，从环境检查开始。不要假设别人的电脑已经配置好；不要沿用演示机路径。随包执行器支持 macOS，其他系统未提供已验证适配器，不降级为无隔离执行。
- 用户已经有题库或偏好：先用用户指定的题面、图片和评分方式，读 自有题库与偏好。不擅自改题，不用演示题顶替。
- 用户要找权威题 / benchmark：读 题源目录，从作者/机构官方渠道获取原题，冻结版本、题号、图片和答案。少量抽测称“某 benchmark 子集”，不得标成完整 benchmark 分数。
- 只说“简单测一下”：沿用已保存偏好；没有偏好时用内置简单演示题，默认最多3题、先1个模型。手机指令不自动扩大成长任务。
所有下列命令从本 Skill 文件夹执行；路径有空格时用双引号引用，或使用程序参数数组。脚本路径相对 Skill，自定义题库路径相对该题库。
初始化与凭据
- 运行
node scripts/setup.mjs doctor。未初始化则按初始化文档补软件、运行 init；已有配置保留，不清空用户的 Claude Code 全局配置。 - 用户在终端执行
node scripts/connect.mjs，由本人隐藏输入自己的 Agent Plan Key。Key 只在连接进程内存里；配置、Skill、题库、报告和截图均不得含真实 Key。不能安全输入时给用户这一步，不让其把 Key 发到聊天。 - 运行
node scripts/runner.mjs isolation-check。越界读写、外网必须拒绝，目录内写、Node、Claude CLI 启动必须通过；失败则停在具体错误，不换成无隔离方式。 - 对每个模型先跑 connection；要用图片再跑 visioncheck，要做代码题再跑 toolscheck。准备检查不计正式测评。工具检查核对真实 tool_result，模型自述不算证据。视觉探针失败不直接断言模型不支持视觉，先检查接口错误与原始回答。
- 连接关闭或重启后重新输入 Key，并重新做该连接的准备检查。更换模型名单后重启连接。接口拒绝、模型名错误或额度不足不自动切模型。
冻结题目，再执行
- 先用 bank validate 检查题库；用 run --dry-run 列出本轮题号、模型、预算，不调用模型。
- 模型对照使用同题面、同图片字节/顺序、同初始工程、同工具、同预算。固定 seed 和题号；保留全部选中题，不看结果后换题。不同题库、工具条件或预算分组报告。
- 支持一次 run --models=模型A,模型B 顺序执行，避免抢同一本地服务端口。默认不自动重跑已有同条件任务；补跑需明确理由并加 --repeat --reason="原因"，保留失败记录。
- 简单问答禁用全部工具；代码题限 Read、Write、Edit、Bash，默认无外网。每题新工作目录与 Claude 配置，禁用全局 Skill/MCP/记忆/历史发现。不能使用跳过权限或关闭隔离参数。
- 题库答案、解析、评分脚本不复制进被测模型目录。图片作为真实 image 内容传入，按 Image 1…顺序记录；不能用调度者的图像描述替代视觉输入。
- 隔离连接给每题单独的短期令牌，绑定单个模型；被测进程不接触真实上游 Key。系统运行库仍可读，本适配器是进程沙箱，不宣称是虚拟机。
- 生成的命令和网页视为待验收产物；后端在同样的受限环境跑测试，不能直接在宿主机执行不受限代码。
常用流程：
node scripts/runner.mjs run --cases=connection,visioncheck,toolscheck --model=glm-5.3-flash
node scripts/runner.mjs run --dry-run --tier=simple --count=3
node scripts/runner.mjs run --tier=simple --count=3
node scripts/runner.mjs run --cases=orbit_audio --model=glm-5.3-flash --seconds=1200
node scripts/runner.mjs status
node scripts/runner.mjs export
What ships with it
21 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.
- agents/openai.yaml 264 B
- assets/custom-example/bank.json 578 B
- assets/demo/bank.json 15 KB
- assets/demo/board-source.html 17 KB
- assets/demo/clock.png 53 KB
- assets/demo/illusion.png 22 KB
- references/benchmark-sources.md 3.8 KB
- references/initialization.md 6.1 KB
- references/local-machine.md 717 B
- references/methodology.md 2.0 KB
- references/question-bank.md 4.9 KB
- references/visual-prompts/board-redesign.md 3.3 KB
- references/visual-prompts/orbit-audio.md 3.4 KB
- scripts/bank.mjs 1.5 KB runs code
- scripts/broker.mjs 3.9 KB runs code
- scripts/connect.mjs 2.3 KB runs code
- scripts/core.mjs 7.7 KB runs code
- scripts/import-benchmark.py 7.3 KB runs code
- scripts/runner.mjs 13 KB runs code
- scripts/self-test.mjs 7.8 KB runs code
- scripts/setup.mjs 3.3 KB runs code
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.
- 3d ago First seen · 63 lines · 85 tokens per session scan A 7104ea8a7258
liucong-model-eval is a skill published in the GitHub repository liucongg/liucong-skills (245 stars, last pushed 3d ago), licensed Apache-2.0. It adds 85 tokens to every session and 1,710 once invoked, about $0.0004 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 2026-09-08.
Other skills, from other repositories
systematic-debugging
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes.
local-ai-agents
Build local-first AI agents that run entirely on a developer workstation with Microsoft Foundry Local and Qwen function-calling models. Covers Small Language Models (SLMs), the OpenAI-compatible local endpoint, sandboxed local tools, local RAG with Chroma, local MCP servers, hybrid cloud/local routing, and the…
next-cache-components-adoption
Turn on Cache Components in a Next.js app and resolve the blocking routes it surfaces. Use when the user wants to enable, adopt, or migrate to Cache Components, flip the cacheComponents flag, work through a flood of blocking-prerender / instant validation errors, run the cache-components-instant-false codemod, or…
insight-error-page
Write or audit an insight-kind error page for the Next.js dev overlay. Use when creating a new errors/ .mdx page, auditing an existing one, or checking that a page matches the framework fix cards. Covers page structure, title alignment, FixCard cards with Copy prompt button, code snippets, terminology verification…
next-cache-components-optimizer
Drive a Next.js route to instant navigation by setting up an agentic loop, under Cache Components / PPR, on initial load (hard navigation) and client-side navigation (soft navigation). Encode the goal as a failing @next/playwright instant() e2e and work it to green, one verified route at a time; the shipped test then…
next-partial-prefetching-adoption
Turn on Partial Prefetching in a Next.js app and work through the insights it surfaces. Use when the user wants to enable or adopt Partial Prefetching, flip the partialPrefetching flag, opt routes in with export const prefetch = 'partial', audit Link prefetch={true} behavior, preserve existing prefetched UI with…