Back to search
code-walkthrough
MarketplaceWorks with Claude CodeGitHub
Agent skill: interactive walkthroughs that follow one request through real code pinned at a commit, to understand an issue before implementing it
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/liyangtseng--code-walkthrough)HTML
<a href="/listings/liyangtseng--code-walkthrough"><img src="/api/listings/liyangtseng--code-walkthrough/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Claude Code, as a plugin:
/plugin marketplace add LiyangTseng/code-walkthrough /plugin install code-walkthrough@code-walkthrough
Or clone and link the skill into your personal skills:
git clone https://github.com/LiyangTseng/code-walkthrough.git ln -s "$PWD/code-walkthrough/skills/code-walkthrough" ~/.claude/skills/code-walkthrough
Other agents that read `SKILL.md` files can use `skills/code-walkthrough/` directly. The build needs `git` and Node.js 18 or later.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 15 lines
- {
- "name": "code-walkthrough",
- "owner": { "name": "LiyangTseng" },
- "metadata": { "description": "Code-trace walkthrough skill", "version": "0.1.0" },
- "plugins": [
- {
- "name": "code-walkthrough",
- "source": "./",
- "description": "Interactive walkthrough pages that follow one request through real code pinned at a commit, to understand an issue before implementing it.",
- "version": "0.1.0",
- "author": { "name": "LiyangTseng" },
- "category": "development"
- }
- ]
- }
README.md · README · 73 lines
- English | [繁體中文](README.zh-TW.md)
- # code-walkthrough
- An agent skill that builds an interactive walkthrough of a codebase before you change it. Point it at a GitHub issue (or a PR, a bug report, a question) and it follows one concrete traveler, such as a request, a token or a batch, through the real code pinned at a commit, until you can see why the task exists and where it lands.
- **[Try the live example](https://liyangtseng.github.io/code-walkthrough/examples/sglang-decompression-queue.html)**: step through it with ← →.
- 
- Each page has:
- - **A thesis**: one sentence naming the mechanism that makes the task necessary.
- - **Pinned source**: every step highlights the lines it explains, at a fixed commit, with a link to them on GitHub.
- - **A map**: the traveler moves across the components it crosses, so the thesis is something you watch happen.
- - **Evidence tags** on every step: verified, inference, to measure, draft, schematic. You know what you can build on.
- - **Concepts taught in passing**: the language or framework idea each step needs, at the moment it needs it.
- - **The way forward**: design options, a measurement plan, who else is working on it, and the road to a PR.
- ## Install
- Claude Code, as a plugin:
- ```
- /plugin marketplace add LiyangTseng/code-walkthrough
- /plugin install code-walkthrough@code-walkthrough
- ```
- Or clone and link the skill into your personal skills:
- ```bash
- git clone https://github.com/LiyangTseng/code-walkthrough.git
- ln -s "$PWD/code-walkthrough/skills/code-walkthrough" ~/.claude/skills/code-walkthrough
- ```
- Other agents that read `SKILL.md` files can use `skills/code-walkthrough/` directly. The build needs `git` and Node.js 18 or later.
- ## Use
- Ask in your own words, in the repository you will work in:
- - "Walk me through https://github.com/owner/repo/issues/123 before I implement it."
- - "I'm taking this bug next. Build a walkthrough of where it happens."
- - "How does a request get from the HTTP layer to the scheduler? Walk me through it."
- The agent reads the issue and its discussion, writes the thesis, traces the code at the pinned commit, scripts the steps, draws the map, and builds the page. The build pulls every source file from git at that commit, rejects line ranges and map ids that do not exist, and prints a fact-check list: every step tagged "verified" with the lines it cites, which the agent re-reads before publishing. In Claude's apps the page is published as an artifact; elsewhere you get an HTML file to open in a browser.
- Pages come in English or Traditional Chinese labels out of the box, and the narration follows your language.
- ## Example
- [`examples/sglang-decompression-queue`](examples/sglang-decompression-queue) walks through an [sglang](https://github.com/sgl-project/sglang) issue: why queued request decompressions still hold their whole body in memory. [Open it live](https://liyangtseng.github.io/code-walkthrough/examples/sglang-decompression-queue.html), or rebuild it from an sglang checkout:
- ```bash
- node skills/code-walkthrough/scripts/build.mjs examples/sglang-decompression-queue/walk.js --repo ~/path/to/sglang --out docs/examples/sglang-decompression-queue.html
- ```
- GitHub Pages serves `docs/`, so the rebuilt page goes live when it is pushed.
- ## Layout
- ```
- skills/code-walkthrough/
- SKILL.md the workflow and the evidence tags
- references/ walk.js format, map design, writing the scenes
- assets/player.html the player every page is built on
- scripts/build.mjs pulls pinned sources, checks the walk, writes the page
- examples/ walkthroughs to read and copy
- ```
- ## License
- MIT. The example quotes sglang source, which is Apache-2.0.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.