Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/Light0305/Light-skillsnpx agentmods add skills/light0305/light-skills/light-file-readingWrote 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/light0305/light-skills/light-file-reading)<a href="https://agentmods.dev/skills/light0305/light-skills/light-file-reading"><img src="https://agentmods.dev/badge/skills/light0305/light-skills/light-file-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/light0305/light-skills/light-file-reading"><img src="https://agentmods.dev/badge/skills/light0305/light-skills/light-file-reading.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 | $0.00376 | $0.08337 |
| Opus 5 | $0.00188 | $0.04168 |
| Sonnet 5 | $0.00075 | $0.01667 |
| Haiku 4.5 | $0.00038 | $0.00834 |
Grade A, and why
light-file-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 10d 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 — 322 lines — stays where its author put it; the contents beside it link to each section on GitHub.
多格式文件深度理解（file-reading）—— 常驻横切
你是 Light 技能包的文件理解归属方：任何任务一旦涉及"用户给的文件 / 已有材料"，你后台自动启用， 把它读懂再交给下游。头部同类已经能做结构抽取、论文深读或 claim↔evidence 分析，不能把它们统称为 "只会抽取"。Light 的可验证组合是：先分诊输入 → 只解析一次并先建结构地图 → 用页/节/表/单元格定位 claim 与证据 → 显式记录覆盖缺口 → 产五面理解笔记 + 下游动作映射，而不是文本堆叠。
一句话定位：把"读文件"升级成「先判宿主能否原生读 → 输入分诊 → 结构地图先行 → 带定位与覆盖记录的五面笔记 → 映射到下游技能动作」；把"确定性脏活"（抽版面文本 / 表→DataFrame / 读模板格式约束 / 数据画像）自己干净利落做掉。它是横切 overlay，不是 DAG 节点（orchestrator-spec §3.1），是大量主线技能 的前置基础。它产读取覆盖状态、固定 fixture 抽取质量证据与"能否宣称读懂"的状态机报告，
document_status复用共享状态契约； 不产 findings（读取状态/benchmark 不是light.findings.v1），也不冒充 C1/C2 内容门。 对标判据唯一真相源 =docs/competitors/file-reading.md。
何时启动（触发信号）
常驻后台：任何任务里出现"已有文件 / 用户上传的材料 / 让你看一份东西"，自动启用、无需显式调用。
硬触发点（必须先读懂再动手，不是扫一眼就开干）：命中任一，在执行下游动作前先产理解笔记：
| 硬触发点 | 为什么 | 动作 |
|---|---|---|
| 用户给论文 / 让你"看看这篇" | 不抓 claim↔证据结构就提不出好评/好 idea | 抽章节骨架 + 论证链 + 最像的前作信号 → 喂 literature-search / idea-critique |
| 用户给模板 / 投稿要求 / 格式规范 | 模板的价值是硬约束（页数/字体/章节/引用风格），不是内容 | docx_read layout/runs 抽页边距/字号/编号 → 喂 paper-writing / typesetting |
| 用户给数据集 / Excel / CSV | 先判规模/质量/明显红旗，免得下游在烂数据上白干 | xlsx_read profile 出 shape/dtypes/describe → 规模质量初判 → 喂 data-engineering 做深度泄漏查 |
| 用户给审稿意见 / 修订稿 | 必须分清"必改 vs 可商榷"，不能一锅端 | pandoc --track-changes=all 读修订/批注（保作者+时间）→ 分级 → 喂 review-rebuttal |
| 用户给 PPT / 截图 / 设计稿 | 视觉风格要"真看一眼"，纯文本盲读丢版式 | markitdown 抽文本 + 渲染成图喂宿主多模态 Read 看版式/配色 → 喂 frontend-design / figure |
| 用户给压缩包 / 代码库 | 结构、依赖、可复用模块比单文件更重要 | 解包递归按类型处理；代码读结构/依赖/逻辑 → 喂对应技能 |
if 用户说"这个文件讲了啥 / 帮我看看这份 / 按这个模板写 / 这些数据能做什么 / 回应一下审稿意见" then 先按"决策第一步"判怎么读；PDF 先
triage，长文档先建结构地图；再产带定位与覆盖记录的理解笔记并映射下游， 不把"我大概扫了一下"当读懂。
决策第一步：先问宿主能不能原生读（省依赖，别为脚本而脚本）
Claude Code 等宿主的 Read 工具能直接读 PDF / 图片 / Jupyter notebook。 能原生读就别先写 pdfplumber 绕远路。
| 你要什么 | 怎么读 | 例 |
|---|---|---|
| ① 轻任务"看懂内容"（讲了啥 / 提要点 / 读图表） | 宿主原生 Read 直喂，零依赖最快 | "这篇 PDF 讲了什么" → 直接 Read，别上脚本 |
| ② 结构化抽取（表→DataFrame / 批量 / 改 XML/redline / 扫描 OCR / 公式不求值） | 才上专用脚本/库 | "把这 PDF 里 12 张表抽成 CSV" → pdf_ops extract-tables + verify-tables |
| ③ 宿主读不了的格式（PPTX / Excel / 视频 / 压缩包） | 按下面"按格式选工具" | "这个 pptx 什么风格" → markitdown 抽文 + 渲染图 |
What ships with it
17 files 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.
- assets/extraction-benchmark.example.json 782 B
- assets/reading-contract.example.json 5.1 KB
- assets/understanding-note.template.md 6.3 KB
- references/DOCX-REF.md 3.9 KB
- references/IMG-REF.md 5.4 KB
- references/PDF-REF.md 5.8 KB
- references/PPTX-REF.md 2.5 KB
- references/reading-resource-map.md 13 KB
- references/tools.md 20 KB
- references/XLSX-REF.md 3.2 KB
- scripts/document_status.py 6.3 KB runs code
- scripts/docx_read.py 13 KB runs code
- scripts/extraction_benchmark.py 11 KB runs code
- scripts/pdf_ops.py 26 KB runs code
- scripts/reading_contract.py 33 KB runs code
- scripts/understanding_note_gate.py 29 KB runs code
- scripts/xlsx_read.py 6.7 KB runs code
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.
- 10d ago First seen · 322 lines · 376 tokens per session scan A 3d5c61182a1d
light-file-reading is a skill published in the GitHub repository Light0305/Light-skills (617 stars, last pushed 2mo ago), licensed MIT. It adds 376 tokens to every session and 8,337 once invoked, about $0.0019 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 2026-08-30.
Other skills, from other repositories
doc-reader
Read any common document/data file — PDF, Word (.docx), Excel (.xlsx/.xls), PowerPoint (.pptx), images (OCR), CSV/TSV, plain text, JSON/YAML/TOML, HTML/XML, and most source-code files. Use the readdocument tool.
skill-doc-delivery
Convert markdown to DOCX, PPTX, XLSX, PDF office documents — use when you need exportable deliverables.
document-python-direct-exec
Use direct Python execution for reliable spreadsheet and document/PDF generation operations.
document-direct-python
Use direct Python execution for reliable document creation including spreadsheets, PDFs, and structured reports.
document-python-direct
Use direct Python execution for reliable spreadsheet and document/PDF generation operations.
meta-multi-format-export-pack
From one piece of source content, render four deliverables: .docx report, .pptx slides, .xlsx data, and an HTML/PDF public version.