Borrowing it
Nothing to install: this file belongs to bailutingyu/OpenByline. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/bailutingyu/OpenByline/main/.claude/skills/judge-panel/SKILL.mdgit clone --depth 1 https://github.com/bailutingyu/OpenBylineWrote 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/bailutingyu/openbyline/judge-panel)<a href="https://agentmods.dev/skills/bailutingyu/openbyline/judge-panel"><img src="https://agentmods.dev/badge/skills/bailutingyu/openbyline/judge-panel/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/bailutingyu/openbyline/judge-panel"><img src="https://agentmods.dev/badge/skills/bailutingyu/openbyline/judge-panel.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 | $0.00101 | $0.01646 |
| Opus 5 | $0.00051 | $0.00823 |
| Sonnet 5 | $0.00020 | $0.00329 |
| Haiku 4.5 | $0.00010 | $0.00165 |
Grade A, and why
judge-panel 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 9d 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 — 60 lines — stays where its author put it; the contents beside it link to each section on GitHub.
择优评委手册（judge-panel）
很多写作环节本质是"从一堆候选里挑最好的几个"：标题、金句、钩子、案例、类比、选题角度、画像指纹。这类任务最适合 judge panel——但前提是做对，否则会被"自我偏好"和"长/工整=好"的系统性偏置带偏。
一、铁律：生成与评选分离
禁止 writer 自生自评、自荐 Top2。 一个实例既当运动员又当裁判，会偏好自己的产出、也分不清"工整"和"好"。正确做法是换不同实例：
- 多候选并行生成：N 个 writer 实例（不同侧重/不同公式/不同温度），各产 5-8 候选，汇成一个大候选池（落
*-cand-{i}.mdfragment，见 handoff-protocol §五）。 - 评委独立评选：M 个与生成者不同的评委实例，按下面的规程打分。
- 投票出 Top-K，连同淘汰理由写进正式 artifact（
headlines.md/punchlines.md…）。
并行落盘遵守 fragment-then-merge：生成分支各写各的候选 fragment，judge 汇聚后写正式文件。详见 skill: ultracode-orchestration。
二、评委规程（每个评委都照做）
- 先写 CoT 评判理由，再给分——不许直接吐分数。先逐条说"这条强在哪、弱在哪、命中哪条否决"，分数是理由的结论。
- 对照固定 rubric 打分（见 §三），不凭语感。
- 显式去偏（最重要）：研究表明 LLM 评委的 style bias（偏好华丽/对仗/长）远大于 position bias。每个评委打分前先自我提醒一句："我是否因为某条更长、更对仗、更像金句而高估它？好 ≠ 长 ≠ 工整。" 朴素但精准、能被读者一句话复述的，往往比辞藻华丽的更好。
- 先过否决清单：命中任一硬否决的候选直接出局，不进打分（否决清单 = 复用对应 agent 已写死的硬约束，见 §四）。
- 投票：M 个评委各出自己的 Top-K，取票数交集/多数；平票由理由质量定。Top-K 一般 K=2-3。
三、通用 rubric（各场景在此之上加专属维度）
| 维度 | 问什么 | 权重提示 |
|---|---|---|
| 切题/相关 | 是否精准服务这篇的中心意思/这个论点？ | 高 |
| 可复述 | 读者能否一句话把它转给朋友（"缝合"检验）？ | 高 |
| 具体非套路 | 有没有具体名词/数字/画面，而非空洞华丽？ | 高 |
| 像作者本人 | 是否贴合 voice-profile，无 AI 套话？ | 中高 |
| 节制 | 是否够短、不堆砌、不为工整而工整？ | 中（去偏重点） |
四、各场景的专属维度 + 否决清单（复用对应 agent 硬约束）
- 标题（headline-writer / skill: headline-craft）：专属维度=标题四问（够吸引？能更短？有被抢读欲？用了疑问/转折式？）。否决清单=headline-craft 的标题党否决清单（夸大、悬浮、与正文不符、纯猎奇）。按平台分轨（公众号/小红书/报告各一组候选 + 各自评委）。
- 金句/钩子（punchline-writer）：钩子线与金句线拆成两条独立候选池、各自评委。否决清单=对仗密度以画像密度基线为准（默认 ≤1）；AI 式空泛升华按 voice-profile §5.2 判定淘汰，真诚拔高放行；禁把开头论点复读到结尾；结尾只留 1 句砸点。评委可为破例候选写明理由保留进 Top-K，最终交人类主编定。不设生成天花板：先多生成 N≫3 → judge 按否决清单淘汰 → 按体温/可复述性排序 → 默认建议保留 3 条上下（可按内容调）。
- 案例/类比（story-curator）：专属维度=论点相关性/具体可核实/原创非套路/与作者风格。否决清单=原创类比强制查"是否烂大街"（温水煮青蛙、木桶理论、滚雪球…命中即毙）；真实案例必须带可信度/时效标注。
- 选题角度（topic-strategist）：专属维度=选题四标准（不可替代性/新颖度/可读性/贴近度，贴近度呼应心理账户）+ 是否能收成"一句话点睛角度"。否决=无法一句话说清"对谁有什么用"的角度出局。
- 画像独特标记/指纹（voice-profiler）：N 个分析实例各独立提 2-3 个指纹，投票取交集——被多数独立实例认出的才算真指纹（防单篇巧合）。
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.
- 9d ago First seen · 60 lines · 101 tokens per session scan A bb1b37c6ecad
judge-panel is a skill published in the GitHub repository bailutingyu/OpenByline (2 stars, last pushed 2mo ago), licensed MIT. It adds 101 tokens to every session and 1,646 once invoked, about $0.0005 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 2026-08-31.
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…