| name | blog-writing |
| description | Use when Codex needs to write publish-ready Paper Radar Jekyll posts as two separate English and Chinese Markdown files with thematic titles, institutions, key figures, and a human research-note voice. |
Blog Writing
Write publish-ready Paper Radar posts as two separate Markdown files: one English, one Chinese.
File rules:
- Create English post:
_posts/YYYY-MM-DD-HHMM-paper-radar-en.md.
- Create Chinese post:
_posts/YYYY-MM-DD-HHMM-paper-radar-zh.md.
- Do not overwrite an existing post.
- Do not save PDFs or large images to the repo; only save essential figure/table screenshots under
images/paper-radar/YYYY-MM-DD-HHMM/.
English front matter:
---
title: "A thematic English title, not a date"
date: YYYY-MM-DD HH:MM:SS +0800
permalink: /posts/YYYY/MM/paper-radar-YYYY-MM-DD-HHMM-en/
lang: en
translation_url: /posts/YYYY/MM/paper-radar-YYYY-MM-DD-HHMM-zh/
author_profile: false
tags:
- paper-digest
- paper-radar
- AI
- agents
- research
---
Chinese front matter:
---
title: "中文主题标题,不要用日期"
date: YYYY-MM-DD HH:MM:SS +0800
permalink: /posts/YYYY/MM/paper-radar-YYYY-MM-DD-HHMM-zh/
lang: zh
translation_url: /posts/YYYY/MM/paper-radar-YYYY-MM-DD-HHMM-en/
author_profile: false
tags:
- paper-digest
- paper-radar
- AI
- agents
- research
---
Style:
- English post contains English only.
- Chinese post contains Chinese only.
- Both posts must be independently readable; do not make one a stub or append one language after the other.
- Use a human research-note voice: concrete, reflective, lightly personal, not promotional.
- The title must summarize the issue's theme, not the timestamp.
- Avoid hype and vague claims.
- It is fine to say “我会优先看这篇的原因是...” when useful.
English structure:
- Short TL;DR.
- "What I am watching this round" or similar human-feeling framing.
- Paper notes.
- Reading priority and next questions.
Chinese structure:
- 简短 TL;DR。
- “本期我在看什么”或类似真人研究笔记式开场。
- 论文细读笔记。
- 阅读优先级和下期问题。
Each paper section must include:
- title
- authors
- institutions
- date or venue
- links
- key figures: 1-3 images with captions; every figure/table referenced in text must be shown (remote or local), no pointer-only references
- quick idea: one or two sentences that let a reader immediately understand the paper's core move
- why it matters: the concrete research or product bottleneck the paper addresses
- method walkthrough: 2-4 concrete steps covering mechanism, system design, training signal, data construction, or evaluation design
- evidence: source-backed datasets, benchmarks, baselines, ablations, key numbers, result tables, or curves
- why I care
- limitations/questions
- connection to the tracked themes
Depth and figure rules:
- Treat each selected paper as a mini explainer, not an abstract rewrite.
- Prefer fewer papers with deeper reading over more papers with shallow notes.
- English paper sections should usually be 450-800 words; Chinese paper sections should usually be 700-1200 Chinese characters.
- Include the main figure/method diagram whenever possible, plus one key result table/curve when it materially helps the reader; every figure/table mentioned must be embedded as an image.
- After every image, write 2-4 sentences explaining what it shows, which claim it supports, and what caveat matters.
- If figure extraction is unreliable, capture a local screenshot or crop from the open PDF/HTML; if you still cannot obtain an image, do not reference that figure/table.
Do not expose internal labels:
- Do not write
原文读取状态:fulltext_read.
- Do not write
partial_read or metadata_only as raw fields.
- If source access was limited, explain naturally.