changelog · shpigford — Claude Code skill: what it does, how to install
Drafts customer-facing release notes from merged pull requests and inserts them directly into static site/changelog.html.
MIT 135 installs ★ 653 updated 2026-11-02
npx skills add Shpigford/nurb --skill changelog -a claude-codeWhat changelog does
This skill writes release updates for nurb by checking the gap between published versions and git history. It reads the highest version ID in site/changelog.html, converts SemVer names to PEP 440 git tags, and gathers commits using git log and GitHub CLI. Instead of dumping raw commits, it extracts only the improvements a non-technical 3D printing user can see in the app or viewport.
The workflow inserts an <article class="release"> block at the top of the main container, assigning new, improved, or fixed tags to each item. Internal changes like CI updates, test suites, and internal refactors are omitted. To verify output, it spins up python3 -m http.server 7878 -d site, checks a browser capture, and runs grep checks against duplicate IDs and em-dashes.
This skill expects the exact directory layout of the nurb project; without site/changelog.html and its specific HTML markup, the automation halts.
What it helps with
- filter merged pull requests down to visible user-facing changes
- format release entries into static HTML with category tags
- translate PEP 440 git tags into SemVer version anchors
- verify changelog markup using a local HTTP preview server
Example prompts for the agent
- Update the changelog for the latest release.
- Draft release notes from recently merged pull requests.
- Write the changelog entry for the upcoming version.
How to install changelog
1Skills CLI (Vercel)
npx skills add Shpigford/nurb --skill changelog -a claude-code2Manually: put the files in place
~/.claude/skills/changelog/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/changelog/
1Skills CLI (Vercel)
npx skills add Shpigford/nurb --skill changelog -a codex2Manually: put the files in place
.agents/skills/changelog/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Shpigford/nurb --skill changelog -a cursor2Manually: put the files in place
.cursor/skills/changelog/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Shpigford/nurb .claude/skills/changelog2Manually: put the files in place
.github/skills/changelog/SKILL.md1Skills CLI (Vercel)
npx skills add Shpigford/nurb --skill changelog -a gemini-cli2Platform CLI
gemini skills install https://github.com/Shpigford/nurb --path .claude/skills/changelogManually: put the files in place
.devin/skills/changelog/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add Shpigford/nurb --skill changelog -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about changelog
How do I install changelog in Claude Code?
Run npx skills add Shpigford/nurb --skill changelog -a claude-code. The skill configuration will be stored in ~/.claude/skills/changelog/SKILL.md.
What does the changelog skill do?
It inspects git tags and merged PRs, filters out technical debt or internal commits, and appends user-friendly release notes into site/changelog.html.
How does it verify the generated HTML?
It launches python3 -m http.server 7878 -d site, inspects a local screenshot, and runs grep checks to catch duplicate IDs or forbidden em-dashes.
Source text
SKILL.md · MIT
---
name: changelog
description: Write nurb.app changelog entries from the releases since the last published one. Filters the release's merged PRs to what a user would actually notice, drafts each in the site's voice, and inserts the release into site/changelog.html. Use when the user says "changelog", "update the changelog", "announce this", or after cutting a release.
---
# nurb changelog
One surface: `site/changelog.html`, the static page behind nurb.app/changelog. One `<article class="release">` per released version, newest first. There is no build step and no data file; the HTML is the data.
Releases are the unit. A nurb release is a version bump merged to main; `desktop/scripts/release.sh` tags it and publishes it as a GitHub release. Final releases use the same spelling everywhere (`v1.0.0`); prerelease git tags use PEP 440 (`v1.0.0a1`) while the app and changelog use SemVer (`v1.0.0-alpha.1`). The changelog entry for a version is written from the commits merged between one git tag and the next, filtered hard.
## Step 1: Find the gap
The newest `id="vX.Y.Z"` in `site/changelog.html` is the high-water mark in SemVer spelling. Map prereleases to PEP 440 before looking for their git tag: `v1.0.0-alpha.1` maps to `v1.0.0a1`, `beta` to `b`, and `rc` loses only the hyphen and dot (`v1.0.0-rc.1` to `v1.0.0rc1`).
```bash
grep -m1 'id="v' site/changelog.html
git fetch --tags && git tag --sort=-v:refname | head -20
```
Every tagged release newer than the high-water mark needs an entry. If the user describes a feature in free text instead, write that one entry into the version it shipped in (or the pending version if it hasn't shipped yet) and skip the sweep.
When /release invokes this before the bump PR merges, the newest version has no tag or GitHub release yet. Write its entry anyway: source it from the PRs merged since the last release (`gh pr list --state merged --base main --json title,number,mergedAt`) plus whatever the release branch itself ships, date it today, and write the version as plain text, like every entry already on the page. The entry lands as part of the release PR, so commit it there instead of leaving it uncommitted.
For each missing version:
```bash
git for-each-ref --format='%(creatordate:iso-strict)' refs/tags/vPEP440
git log --format='%h %s' vPREVIOUS_PEP440..vPEP440
```
The tag date is the publish date and the commits between tags are the release contents. Read a PR (`gh pr view N`) whenever a commit or PR title alone doesn't tell you what a user would notice.
## Step 2: Filter to what earns a line
Most PRs are not changelog lines. A PR earns one only if **someone using nurb could notice the difference without reading the code.**
Yes: a new capability in the viewer or the modeling vocabulary, a new check, a behavior change they'd feel, a fix for something that visibly broke, a change to how their AI works with them (asks for measurements, names sliders in their words).
No: CI, tests, refactors, repo docs, CLAUDE.md, release plumbing, skill-file mechanics with no visible effect, marketing site tweaks.
Group by capability, not by PR. Several PRs polishing one feature collapse into one line. A version whose every PR is internal gets no entry at all; say so in the report rather than padding.
## Step 3: Draft the copy
The audience is a hobbyist with a printer, not a programmer. They talk to their AI and judge parts in a browser.
- Never mention Python, module names, function names, build123d, OCCT, or agent-harness jargon. `stand()` is "diagonal printing"; `nurb inspect` is "your AI can inspect a part".
- "Your AI" is the actor for agent-side changes, matching the landing page.
- Sentence case, second person, plain English. Say what changed and what they can now do, then stop. No rationale, no design defense.
- No em-dashes, no exclamation points, no emoji, no marketing fluff.
- Each line gets a tag: `new`, `improved`, or `fixed`. Only `new` also carries the class (`<b class="tag new">`); the other two are `<b class="tag">improved</b>` and `<b class="tag">fixed</b>`.
- A headline capability may lead its line with a bold two-or-three-word name: `<b>Assemblies.</b>`.
## Step 4: Write the page
Insert the new `<article class="release" id="vX.Y.Z">` at the **top** of `<main>`, right after the `.lede` paragraph. Copy the markup shape of the entry below it exactly:
```html
<article class="release" id="v0.9.0">
<div class="meta"><span class="ver">v0.9.0</span><time datetime="2026-07-31">July 31, 2026</time></div>
<ul>
<li><b class="tag new">new</b><span><b>Diagonal printing.</b> A part that prints better tilted...</span></li>
<li><b class="tag">fixed</b><span>...</span></li>
</ul>
</article>
```
The date is the annotated tag's creator date, not today (a pre-merge entry uses today, which becomes the release date when the PR merges). Never reword or delete an already-published entry; its `id` is a live anchor.
## Step 5: Verify
Look at the page, don't just lint it: serve `site/` (`python3 -m http.server 7878 -d site`), open `http://localhost:7878/changelog.html` with whatever browser tooling is available, and screenshot it. Check the new entry renders like its neighbors, then kill the server. Also, silence is pass on both of these:
```bash
git diff -U0 site/changelog.html | grep '^+.*—' # em-dashes in what you added
grep -o 'id="v[^"]*"' site/changelog.html | uniq -d # duplicate versions
```
And confirm by eye that versions still descend down the page.
## Step 6: Report
List each version with its lines and tags, plus one line for what was skipped and why. Leave the change uncommitted unless asked.
© Shpigford · MIT · Shpigford/nurb
Similar skills
All skills for Git and GitHub →changelog by other authors comes first.