changelog-entry

Draft the ## [Unreleased] entry for what just changed. Keep a Changelog format, written for the person upgrading.

Read the CHANGELOG's own rules first

Always. Every repo of mine states its conventions in its own CHANGELOG.md header, above the first ##, and they differ per project. That header outranks everything below.

Two real examples of what lives up there:

• poops — the [Unreleased] heading takes a short title after an em dash, and the entry opens with one paragraph saying what was wrong before. Those two become the title and description of a generated changelog post, so they are not decoration. Write plain Markdown; template tags get fenced automatically.
• code-preview-element — call out anything that changes the DOM the element produces, the CSS an author may already be targeting, or the contents of the preview iframe. None of the three shows up in a function signature, so none of them is visible in the diff to a reader.

If the header asks for something not listed here, do that instead.

Steps

1. git diff (or git diff --staged, or against the merge base) to see what landed.
2. Read CHANGELOG.md's header block for the house rules.
3. Read the last two or three released entries for bolding and grouping. Match their shape, not their length — an existing entry that runs long is not a licence to.
4. Write the entry under ## [Unreleased], in the right group.

Groups

### Added, ### Changed, ### Deprecated, ### Removed, ### Fixed, ### Security. Only the ones you need, in that order. Create ## [Unreleased] if it is missing.

Voice

Bare. One sentence saying what changed, one saying why. Two sentences is the ceiling, not the target — most bullets are one.

• Write for the person upgrading, not the person who wrote the code. "Describe what changed for someone using this, not which functions moved."
• Lead each bullet with a bold sentence stating the change. Add the why only when the diff cannot show it — a fix, a rework, a reversed default.
• No commit hashes, no file paths, no function names unless they are public API.
• No AI attribution.

Shape of a bullet that works:

- **The console strip moved from under the preview to under the code block.** The lines are
  logged by the js in the pane, so they belong against the pane.

Cut before writing

An entry is not a design note, a test report or a post-mortem. None of these belong in it:

• How it was verified. Browsers driven, scenarios run, numbers measured — that is the PR.
• Alternatives rejected, and why the shape is what it is — that is the commit body.
• Internals a reader cannot touch — a listener removed, a helper inlined, a selector constant. Only what they can call, write in markup, style or catch.
• The second telling. One narrative per bullet; a rename table replaces its prose, not joins it.

Long is the failure, not the effort. If a bullet needs three paragraphs to land, the thing it describes needs a docs page — link that instead.

Breaking changes

Say so in the bullet, in the words someone hits it with — the error they will see, the attribute that stopped working, the migration. A [Unreleased] entry becomes the GitHub release body verbatim via script/publish, so this is the only place the migration gets written.

Boundaries

Writes the entry. Does not bump the version, does not cut the release, does not commit — script/publish does all three, and it reads what this wrote. See the release skill.