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 Azzygoatcoder/agent-useful-skills --skill paper-readinggit clone --depth 1 https://github.com/Azzygoatcoder/agent-useful-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/azzygoatcoder/agent-useful-skills/paper-reading)<a href="https://agentmods.dev/skills/azzygoatcoder/agent-useful-skills/paper-reading"><img src="https://agentmods.dev/badge/skills/azzygoatcoder/agent-useful-skills/paper-reading/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/azzygoatcoder/agent-useful-skills/paper-reading"><img src="https://agentmods.dev/badge/skills/azzygoatcoder/agent-useful-skills/paper-reading.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 | $0.00084 | $0.01906 |
| Opus 5 | $0.00042 | $0.00953 |
| Sonnet 5 | $0.00017 | $0.00381 |
| Haiku 4.5 | $0.00008 | $0.00191 |
Grade A, and why
paper-reading 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 — 62 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Paper Reading — 论文阅读工作流
科研骨架的论文阅读模块。目标：把一篇论文读深读透，产出符合自家格式、经得起对抗质疑的阅读报告。
第 0 步：场景判定（先定深度，再选流程）
阅读需求不同，深度和产出完全不同。先判定场景，再走对应流程：
| 场景 | 触发 | 深度 | 产出 | 状态 |
|---|---|---|---|---|
| A. 搜索 | 找新论文 / 调研方向 | L1 | 候选表 + 入库 | ✅ 本轮 |
| B. 防撞车 | 组会前 related work | L1-L2 | 风险分级对比表 | ✅ 本轮 |
| C. 快速读 | 判断是否值得精读 | L1 | 一句话总结 + 定位 | ✅ 本轮 |
| D. 精读报告 | 正式阅读 / 写报告 | L2-L3 | 六节模板 + 置信度 frontmatter | ✅ 本轮 |
| E. 深挖对抗 | 跟进 / 定 PRP 方向 | L3-L4 | 脆弱假设 → 反例 → idea | 后续轮 |
| F. 引用审计 | 投稿 / 引用核实 | — | 每条引用三层验证 | 后续轮 |
本轮实现 A（搜索入库）+ B（防撞车）+ C（快速读）+ D（精读）。遇到 E/F 触发词，先确认是否真需要精读；不需要就按浅场景快速处理，不强行走 D。
场景指南
选择场景后，加载对应详细流程：
- A 搜索与入库：自动搜题、多信源、状态机 → references/scenario-guides.md
- B 防撞车检查：定靶心 → 定范围 → 多源搜索 → 风险分级 → 产出 → references/scenario-guides.md
- C 快速读：抓摘要 → 五问定位 → 一句话总结 → 处置三选一 → references/scenario-guides.md
- D 精读闭环：定位抓原文 → 六节报告 → 置信度 frontmatter → 信息卫生 → 批判性验证 → references/scenario-guides.md
核心不可丢的规则已在场景指南中；主文件只负责分流和索引。
自进化日志
每次阅读实践吸收的模式记录于此，skill 随之进化：
| 日期 | 学习来源 | 吸收的模式 |
|---|---|---|
| 2026-08-03 | 初始 17 篇 prp 报告 + RubricsTree 范本 | 六节模板=格式规格；RubricsTree 是唯一完整执行 frontmatter+六节的范本；prp 基线失败=缺 frontmatter、节被合并/删除 |
| 2026-08-03 | ARIS / PaperForge / kill-argument 调研 | 渐进阅读省 context；四类信息卫生映射置信度；承诺式攻击替代平衡弱点清单 |
| 2026-08-03 | CircuitFusion D 闭环首测 | D1 抓原文路径可用（元数据/数字全对上）；代码链接不在 arXiv abs 页需直接访问 repo 核实；批判从方法依赖链推导（summary 摘要谁写）；信息卫生落地=推断集中个人体会节；一句话总结设为可选 |
| 2026-08-05 | T²-GRPO GREEN 全量验证 | D 闭环端到端跑通、无需迭代：D1 元数据/数字核对、D3 frontmatter、D4 推断标注（可考虑/推测）、D5 承诺式批判（模拟器保真度循环论证）全部可执行；产出实习模板报告 |
| 2026-08-05 | T²-GRPO 去数学化试验 | 公式策略按受众分：专业/技术→保留公式+规范 LaTeX（$...$/$$...$$）；一般组会→去公式抓核心思想（组会分享进度非数学课）。受众判定先于写作，与 figure-drawing 分场景同理 |
| 2026-08-05 | 现有防撞车检查模板（prp/notes） | B 场景工作流：定靶心→定范围→多源搜索→风险分级（高/中/低）→产出（分级表/全景矩阵/建议）；撞车判定=任务×方法维度组合全覆盖（非"提到类似词"）；中高风险升 D 精读确认；摘要级≠全文确认 |
| 2026-08-05 | arxiv_fetch.py 实战（A 场景） | A 场景工作流：arxiv_fetch.py（search/add/download/status/view）+ 状态机 candidate→fetched→read→report→cited；实战挖出 TACO 新撞车候选（手工漏掉）——自动化信源收集 > 手工检索 |
| 2026-08-05 | 时间线边界讨论 | 撞车时点基准 = 目标投稿 deadline（审稿可见性分水岭）；相似不互相影响三条件：时间错位(concurrent)/贡献点不重叠/上下游互补；灰色带保守当可见处理 |
| 2026-08-06 | DBLP + Semantic Scholar 接入 | arxiv_fetch.py 升级为多信源：--sources arxiv,dblp,semantic + cite(引用数) + bibtex(DBLP 权威)；S2 无 key 优雅降级（429 静默）教训：外部 API 必须有降级路径 |
| 2026-08-06 | S2 key 被拒 → OpenAlex 替代 | S2 个人申请被拒（优先学术/非营利）；OpenAlex 稳定无 key 但引用数对近期论文滞后（2025-2026 多 0）；cite=OpenAlex 底 + S2 兜底补当前数；教训：免费引用数对近期论文不可靠，够用即可 |
| 2026-08-06 | PubMed 接入 | pubmed 子命令（esearch+esummary 免费）：实习线临床证据源；--save 追加 markdown 表；教训：PowerShell GBK 显示乱码≠文件编码错（用 Read 验证 UTF-8）；f-string 表达式不能含反斜杠（3.11 限制） |
| 2026-08-06 | C 场景 GREEN 验证（TACO） | C 端到端可执行：抓摘要→五问→一句话→处置；防撞车中风险候选强制升 D 的判定成立；C5 边界「摘要≠全文」实战有效 |
| 2026-08-06 | D 配图嵌入需求 | D7 配图嵌入：extract 免费滤 + classify 精选 1-4 张重要原图 → markdown 嵌入报告对应章节；报告配原图可读性大增（组会/导师直接见图） |
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 · 62 lines · 84 tokens per session scan A 352bcbb2bc0a
paper-reading is a skill published in the GitHub repository Azzygoatcoder/agent-useful-skills (6 stars, last pushed 11d ago), licensed MIT. It adds 84 tokens to every session and 1,906 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 2026-08-31.
Other skills, from other repositories
dsh-plugin-dev
An authoritative development guide for DeepSeek Harness plugins. DeepSeek Harness is a system where models, tools, services, events, and adapters are built as plugins.
dsh-plugin-client
A development guide for the browser interface of a DSH plugin, including its package structure, registered interface locations, and build setup.
dsh-plugin-dev
A guide for creating and debugging DeepSeek Harness plugins, which are add-ons packaged as npm modules with configuration files.
dsh-plugin-i18n
A pattern for giving a DSH plugin’s user interface English and Chinese text that changes immediately when the selected language changes.
dsh-git-commit
A Git commit guide for writing standardised commit messages and performing basic checks before committing code. Git commits are saved checkpoints in a project’s history.
dsh-dependency-audit
A guide for reviewing and upgrading project dependencies, which are third-party packages the project relies on. It covers security warnings, installation scripts, lock files, breaking changes, tests, rollback plans, and reducing unnecessary dependencies.