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 pwya/md-paper --skill md-triagegit clone --depth 1 https://github.com/pwya/md-paperWrote 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/pwya/md-paper/md-triage)<a href="https://agentmods.dev/skills/pwya/md-paper/md-triage"><img src="https://agentmods.dev/badge/skills/pwya/md-paper/md-triage/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/pwya/md-paper/md-triage"><img src="https://agentmods.dev/badge/skills/pwya/md-paper/md-triage.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 | $0.00320 | $0.07903 |
| Opus 5 | $0.00160 | $0.03952 |
| Sonnet 5 | $0.00064 | $0.01581 |
| Haiku 4.5 | $0.00032 | $0.00790 |
Grade A, and why
md-triage 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 — 243 lines — stays where its author put it; the contents beside it link to each section on GitHub.
md-triage — 任意修订意图 → 离散条目清单（第三代·改稿起头）
⛔ 路由铁律：本 skill 属第三代
md-*。它把【任意输入】整理成md_triage.md（离散修订条目清单），停在人工闸；下游md-swarm据此批量并行改manuscript.md。对应第一代reviewer-triage，但通用于任意来源（不止审稿意见），且不碰 docx 占位符。 🧭 定位铁律（决定什么时候用它）：md-triage 是**【大批量乱意见】的专属入口**——一堆审稿意见 / 一份长重构方案 / 一摞语音记录，手改扛不动时才上。小修不经过它：① 改错字/调一两句 → 直接手改manuscript.md→/md-build；② 让 AI 改某一两处（你直接指哪儿）→md-iterate；③ 大批量乱意见 → 本 skill →md-swarm。别为小修启动这套重机械。
🩺 环境自检（可选·WARN 级）：md-triage 不碰
manuscript.md，风险低，但开工前可顺手跑一次共享环境闸，及早发现保护钩子被 cc-switch 刷掉：py "<md-swarm skill>\preflight.py" --mode warn --context md-triage它只提醒不阻断（绿则继续，红则照设计兜底：下游
md-unpack/md-swarm会硬拦）。原理见开发手册 §7.6。
它干什么 / 不干什么
- ✅ 干：读任意输入 → 归一成离散条目 → 分类 / 合并 / 分批 → 写
md_triage.md→ 停。 - ❌ 不干：不改
manuscript.md（那是md-swarmPhase 2 的事）、不出 Word（那是md-build）、不下载外部文献（只列清单提示你去取）。
总流程
第一段【归一】 任意输入 → ①确定性前闸(认格式/原生读/瘦身/分块) → ②LLM抽离散条目(+置信度) → ③确定性后检
第二段【整理】 条目清单 → 五维分类 / 真实章节 / 纯评价过滤 / 合并簇 / 外部资源 / 分批方案
▼
写 swarm/md_triage.md（顶部「人工确认：待确认」令牌 + 行序约束）→ 自检 → 停！汇报、交还用户
▼ 人工闸：用户审清单（标跳过 / 写「人类修改思路」/ 改批次 / 把令牌改「已确认」）→ 交 md-swarm
⛔ DeepSeek-proof 铁律：你（AI）永远不许自己把令牌从「待确认」改成「已确认」。第一段+第二段跑完就停、汇报、交还控制权。 ⛔ 人类修改思路铁律：每条的
**人类修改思路：**是作者的批注槽——AI 永不填写，一律只留（你填）占位。往里写任何内容 = 越权替作者说话（实测弱模型会犯这个错）。自检会查它是否全为占位。 ⛔ 令牌格式铁律：顶部令牌行必须逐字是**人工确认：** 待确认（单独一行、最顶），别改写成别的样子——下游 gate 钩子和用户都只认这一行。
两条出错铁律（任何时候不可破）
- 🌐 语言铁律：条目的「原文」字段逐字保留输入原语言（英文意见就留英文原文）；摘要用中文。改稿语言由下游守，但本步绝不翻译原文。
- 🚫 不瞎猜铁律：模型读不了的东西（读花的扫描件、非多模态模型遇到的图）→ 停下问用户要文字，绝不凭空编一条意见（见 0.1 末）。
第一段 · 归一（把任意输入变成离散条目）
核心原则：别按"输入长啥样"分类，按"产出 = 一条条离散可操作条目"定义这一步。 输入无穷，产出唯一形状。唯一让 AI 放手的是 0.2「抽条」，它被 0.1 的确定性前闸 + 0.3 的确定性后检 + 后面的人工闸夹住。
0.1 确定性前闸：认格式 → 原生读（text-first）→ 瘦身 → 分块
读取 cascade（查表选路，先试最便宜的原生读法，读不出才升级）：
| 输入 | 怎么读（依次尝试） |
|---|---|
.md / .txt |
直接 Read |
.pdf |
① pdftotext -layout "<文件>" -（Bash/PowerShell）；② 抽不出 / 满屏乱码方块 → Skill anthropic-skills:pdf（自带 OCR） |
.docx 正文 |
三级：① mcp__docx-mcp__（open→extract→close，只读文字；多数版本只回结构计数、抽不到正文是正常的）→ ② Skill minimax-docx（提全部段落文字）→ ③ Skill anthropic-skills:docx |
.docx 里的批注（Word comments） |
意见直接来自 Word 批注（导师/审稿人在稿子上批的）。先看 swarm/comments_raw.json 在不在：在 → md-unpack 摄取同一篇时已抠好，直接 Read 它，别再解析一次；不在（批注在另一份后到的 docx 上）→ 当场调共享解析器 py "%USERPROFILE%\.claude\skills\md-unpack\read_docx_comments.py" --docx <文件> --out swarm\comments_raw.json --print（纯 zip+XML、不用 Word COM）。每条 = author（来源）+ comment（=修订诉求，进 0.2 抽条）+ anchor（=被批注的原文片段，直接当该条「原文」+ 拿去 manuscript.md 里子串匹配定「目标位置/章节」）。 |
图片 .png/.jpg / 扫描件 |
① 先让（多模态）模型直接读图（Read 图片，模型能看图就直接看）；② 模型非多模态（如 DeepSeek）/ 读花了 → 上 OCR（Skill anthropic-skills:pdf 或 tesseract）；③ 还不行 → 停下，告诉用户"这是图片我读不可靠，请把内容转成文字给我"。绝不凭空编。 |
| 多文件 / 一个文件夹 | 逐个按上述读；只读修订意图相关文件（跳过明显的归档 / LEGACY / 演示文稿目录），但仍以"逐字原文片段"为准防漏。 |
What ships with it
1 file 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.
- 9d ago First seen · 243 lines · 320 tokens per session scan A 1776ed91c83b
md-triage is a skill published in the GitHub repository pwya/md-paper (10 stars, last pushed 3d ago), licensed Apache-2.0. It adds 320 tokens to every session and 7,903 once invoked, about $0.0016 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
research-academic-plotting
Generates publication-quality figures for ML papers from research context. Given a paper section or description, extracts system components and relationships to generate architecture diagrams via G...
pandoc-docx-template
Use this skill when converting Markdown to Word DOCX or DOCX back to Markdown with Pandoc, especially when the output should use the bundled Chinese Word reference templates, heading numbering variants, list indentation variants, SCI paper templates, and Lua filters for HTML tags, image captions, font color, and…
md-to-office
This skill enables conversion from Markdown to various Office formats using Pandoc - the universal document converter. Convert your Markdown files to professional Word documents, PowerPoint presentations, PDFs, and more while preserving formatting and structure.
Генерация PDF из Markdown с точным академическим форматированием через Pandoc и LaTeX
A document workflow that turns Markdown text into PDF files using Pandoc and a custom LaTeX template. It applies specified academic page settings such as 14-point text, 1.5 line spacing, and A4 paper.
latex-workflow
Write, compile, and submit LaTeX papers: IMRaD structure, key packages, bibliography management, arXiv preparation, and common error fixes.
slide-excellence
Multi-agent comprehensive slide review (visual + pedagogy + proofreading, plus TikZ / parity / substance conditionally). Use when user says "full review", "excellence pass", "comprehensive check", "review everything", "pre-release review", "slide excellence", or before teaching / shipping a deck. Fanout wrapper — for…