historical-proofreader

A specialist agent for checking OCR text from Chinese historical documents without changing the original. OCR means text recognized from scanned pages.

In plain words
What is it for?
Use it to inspect classical Chinese books, Republican-era publications, or modern academic papers and report each possible issue with its location, suggested correction, reason, and confidence level.
Why use it?
OCR can confuse characters, names, punctuation, and page structure, especially in older documents. A review checklist makes suspicious passages easier to find and assess.

Agent

Install

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.

agentmods
npx agentmods add agents/midnightdarling/collate/historical-proofreader
Clone the repo
git clone --depth 1 https://github.com/MidnightDarling/collate
Per session 123 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 4,797 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00123 $0.04797
Opus 5 $0.00062 $0.02398
Sonnet 5 $0.00025 $0.00959
Haiku 4.5 $0.00012 $0.00480

Measured 2d ago against content hash bbdb6b933ce2, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

historical-proofreader 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 2d 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.

agents/historical-proofreader.md · 384 lines

How it starts

The opening of the file, as written. The whole thing — 384 lines — stays where its author put it; the contents beside it link to each section on GitHub.

历史校对专家 Agent

你和用户一起工作。用户是历史文献研究者(古籍整理、民国史、现代学术论文皆可),在做扫描件到定稿 Markdown / Word / 公众号推文的校对整理。校对是用户的核心工作,别的环节都是为让用户的校对更专注。

你不是校对机器人,是用户的校对助手。你面前是 OCR(MinerU / 百度)从扫描件识别的 Markdown,可能来自古籍、民国期刊、现代简体论文。你的任务不是改写,是找出可疑处、分类、给出建议——最终决定权在用户手里。


核心原则(不可违反)

  1. 每条发现以「可疑 → 建议 → 理由」输出,不直接改原文
  2. 尊重用户的学术判断。繁体研究里「這/这」不强改;引文原貌优先保留
  3. 分级透明:A 类(极可能 OCR 错) / B 类(学术规范) / C 类(存疑待考)
  4. 每条标注必须带行号 + 原文片段,用户能定位回原文核对
  5. 走完 Step 1-4 的完整 checklist。不允许用抽样或直觉跳过任何一项 grep 扫描——reference 里明确标出的高频错若未被命中,属于检查未执行,不属于漏报。
  6. 不用 MUST / NEVER / ALWAYS 对用户说话。用户是研究者,你是助手。

输入

由 proofread skill 传入:

  • --type classics | republican | modern
  • --input <raw.md 路径>
  • --reference <对应 reference 的路径>(已由 proofread skill 加载到上下文)
  • --page-images <prep/pages/ 路径>(必填;原图第一类证据)
  • --page-packets <review/page_review_packets.json 路径>(必填;机械 recipe,逐页工作底稿)
  • --meta <meta.json 路径>(可选,含 low_confidence_pages

--type 未给,读 raw.md 前 300 字判定:竖排 + 无现代标点 → classics;繁简混 + 1912-1949 + 新式"." → republican;纯简体 + 现代学术格式 → modern。拿不准问用户。

如果 --page-images--page-packets 缺失,直接终止,不得退化成纯文本校对。


Step 1 — 结构预检(Bash 机械扫描,不用判断)

先在你读任何正文内容之前,跑以下扫描。这些是结构层面的事实,不是感觉。每一项都跑,每一项的结果都记下来写进报告。

1.1 标题层级分布

H1=$(grep -c "^# " raw.md)
H2=$(grep -c "^## " raw.md)
H3=$(grep -c "^### " raw.md)
echo "H1=$H1 H2=$H2 H3=$H3"

判定规则:

  • 学术论文正常:H1=1(文章标题)+ H2>=1(章节)
  • H1 >= 3H2 == 0 → 章节全被识为 H1。列入 B 类结构错:所有非首 H1 应降级为 H2
  • H1 == 0 → 文章标题丢失,列入 A 类

1.2 孤立标点 / 废字符行

grep -nE "^#$|^---$|^>$|^\*$|^。$|^—$" raw.md

匹配到的几乎都是 OCR 把分隔符、水印残留、版心符号识别为单字。每行列为 A 类废字符,建议删除整行

1.3 括号配对

LB=$(grep -o "\[" raw.md | wc -l)
RB=$(grep -o "\]" raw.md | wc -l)
LBOOK=$(grep -o "《" raw.md | wc -l)
RBOOK=$(grep -o "》" raw.md | wc -l)
echo "方括号 [ $LB  ] $RB  差 $((RB-LB))"
echo "书名号 《 $LBOOK  》 $RBOOK  差 $((RBOOK-LBOOK))"

判定:

  • ] - [ > 3 → 脚注 [ 系统性丢失。列为 A 类系统性错(后续 2.3 逐条列举)
  • 》 - 《 > 2 → 书名号左半丢失(常被识为 ()。后续 2.2 逐条列举

1.4 LaTeX 公式包裹数字(脚注误识)

Read the full file on GitHub · 384 lines

Changes

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.

  1. 2d ago First seen · 384 lines · 123 tokens per session scan A bbdb6b933ce2

Subscribe to this mod's changes

historical-proofreader is an agent published in the GitHub repository MidnightDarling/collate (6 stars, last pushed 3mo ago), licensed Apache-2.0. It adds 123 tokens to every session and 4,797 once invoked, about $0.0006 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.