跳到正文
原文
Simon Willison 博客· Simon Willison 博客·· 2026-05-09精选AI 评分71

Claude Code实践：HTML输出格式的卓越效果

AI 导读

Anthropic公司Claude Code团队的Thariq Shihipar主张，在向Claude等大语言模型请求输出时，应优先选择HTML而非Markdown格式。HTML允许模型直接生成包含SVG图表、交互式组件和页面内导航等丰富元素的文档，显著提升信息呈现的交互性与清晰度。作者以GPT-5.5生成一个Linux安全漏洞的交互式HTML解释页面为例，展示了该方法的实际效果。这促使长期习惯使用Markdown的作者重新评估输出格式，并计划在提示工程中更多尝试富HTML输出。

推荐理由

Simon 拿自己实验打脸 Markdown 惯性，HTML 输出能自动带 SVG 图表和交互导航，所有还在用纯文本解释复杂代码的人都该换个思路。

正文

8th May 2026 - Link Blog

Using Claude Code: The Unreasonable Effectiveness of HTML. Thought-provoking piece by Thariq Shihipar (on the Claude Code team at Anthropic) advocating for HTML over Markdown as an output format to request from Claude.

The article is crammed with interesting examples (collected on this site) and prompt suggestions like this one:

Help me review this PR by creating an HTML artifact that describes it. I'm not very familiar with the streaming/backpressure logic so focus on that. Render the actual diff with inline margin annotations, color-code findings by severity and whatever else might be needed to convey the concept well.

I've been defaulting to asking for most things in Markdown since the GPT-4 days, when the 8,192 token limit meant that Markdown's token-efficiency over HTML was extremely worthwhile.

Thariq's piece here has caused me to reconsider that, especially for output. Asking Claude for an explanation in HTML means it can drop in SVG diagrams, interactive widgets, in-page navigation and all sorts of other neat ways of making the information more pleasant to navigate.

I wrote about Useful patterns for building HTML tools last December, but that was focused very much on interactive utilities like the ones on my tools.simonwillison.net site. I'm excited to start experimenting more with rich HTML explanations in response to ad-hoc prompts.

Trying this out on copy.fail

copy.fail describes a recently discovered Linux security exploit, including a proof of concept distributed as obfuscated Python.

I tried having GPT-5.5 create an HTML explanation of the exploit like this:

curl https://copy.fail/exp | llm -m gpt-5.5 -s 'Explain this code in detail. Reformat it, expand out any confusing bits and go deep into what it does and how it works. Output HTML, neatly styled and using capabilities of HTML and CSS and JavaScript to make the explanation rich and interactive and as clear as possible'

Here's the resulting HTML page. It's pretty good, though I should have emphasized explaining the exploit over the Python harness around it.

Screenshot of a dark-themed technical document titled "What this Python script does". Body text: "This is a compact, deliberately obfuscated Linux-specific local privilege-escalation proof-of-concept. Its apparent goal is to tamper with the in-memory image/page cache of /usr/bin/su, then execute su to obtain elevated privileges." A yellow-bordered callout reads: "Safety note: This explanation is for code understanding, reverse engineering, and defensive analysis. Do not run this on systems you do not own or administer. On a vulnerable kernel, code like this can alter the behavior of a privileged executable." Left column heading "High-level summary": "The script opens /usr/bin/su read-only, decompresses an embedded binary payload, and then processes that payload in 4-byte chunks. For each chunk, it performs a carefully arranged sequence involving Linux's kernel crypto socket interface, AF_ALG, pipes, and splice(). The important point is that this is not ordinary file writing. It never calls write() on /usr/bin/su. Instead, it appears to rely on a kernel bug/primitive involving spliced file pages and the crypto API to get controlled bytes placed into the page-cache representation of a privileged executable." Numbered steps follow: "1. Open target executable — /usr/bin/su is opened read-only. 2. Decode hidden payload — A zlib-compressed hex blob is decompressed into bytes. 3. Patch in 4-byte chunks — The helper function is called repeatedly with offsets 0, 4, 8, ...". Right column heading "Why it looks strange" contains a table with Pattern and Purpose columns: "import os as g — Short aliasing to make the script compact and harder to read. socket(38, 5, 0) — Uses raw numeric Linux constants instead of readable names. Compressed hex blob — Hides binary payload bytes and keeps the script small. splice() — Moves file-backed pages through pipes without normal user-space copying. try: recv(...) except: 0 — Triggers the kernel operation and ignores expected errors."

来源：Simon Willison 博客 · simonwillison.net