Smart changelog

Smart Changelog Generator

Generate professional changelogs and release notes from your git history.

When to Activate

• User says /smart-changelog, "generate changelog", "release notes", or "what changed"
• User wants to prepare a release
• User asks about version bumping or semantic versioning

Instructions

Step 1: Determine the Range

Figure out what commits to include:

# List recent tags
git tag --sort=-version:refname | head -10

# Find the latest tag
git describe --tags --abbrev=0 2>/dev/null || echo "no tags"

# If no tags, use the first commit
git rev-list --max-parents=0 HEAD

Range logic:

• If user specifies a range (e.g., "since v1.2.0"): use that
• If tags exist: use <latest-tag>..HEAD
• If no tags: use all commits (first release)

Ask the user if unclear:

• "I see the latest tag is v1.3.0. Should I generate changes since then?"
• "No tags found — should I include all commits, or specify a starting point?"

Step 2: Gather Commits

# Full commit log with hash, subject, body, and author
git log <range> --format='%H|%s|%b|%an' --no-merges

# Or with merge commits if the project uses merge-based workflow
git log <range> --format='%H|%s|%b|%an' --first-parent

Also check for:

# Any breaking changes?
git log <range> --grep='BREAKING' --format='%H %s'

# Linked issues/PRs?
git log <range> --grep='#[0-9]' --format='%H %s'

Step 3: Categorize Commits

If the project uses Conventional Commits:

Parse the type prefix to categorize:

Type	Category	Emoji
feat	Added	✨
fix	Fixed	🐛
perf	Performance	⚡
refactor	Changed	♻️
docs	Documentation	📚
test	Tests	🧪
chore	Maintenance	🔧
ci	CI/CD	🏗️
style	Style	🎨
revert	Reverted	⏪
BREAKING CHANGE	Breaking	💥

If NOT using Conventional Commits:

Analyze each commit's diff and message to categorize:

# Get the diff stat for context
git show --stat <hash>

Use these heuristics:

• New files created → likely "Added"
• Test files changed → likely "Tests"
• Config/CI files → likely "Maintenance"
• fix, bug, patch, resolve in message → "Fixed"
• add, new, implement, introduce → "Added"
• update, improve, enhance, refactor → "Changed"
• remove, delete, deprecate, drop → "Removed"
• docs, readme, comment → "Documentation"
• perf, optimize, speed, cache → "Performance"

Step 4: Determine Version

If the user hasn't specified a version, suggest one based on semver:

MAJOR (X.0.0) — if any:

• Breaking changes found
• Incompatible API changes
• Major feature overhauls

MINOR (x.Y.0) — if any:

• New features (feat commits)
• New capabilities or endpoints
• Backward-compatible additions

PATCH (x.y.Z) — if only:

• Bug fixes
• Documentation updates
• Performance improvements
• Refactoring

Present the suggestion:

Based on 3 features and 5 fixes (no breaking changes), I suggest:
  Current: v1.3.2 → Next: v1.4.0

Want to use v1.4.0 or specify a different version?

Step 5: Generate Changelog

Use Keep a Changelog format:

## [1.4.0] - 2026-02-14

### Added
- Rate limiting for API endpoints (#123)
- Export to CSV from dashboard
- Dark mode support

### Fixed
- Login redirect loop on Safari (#456)
- Memory leak in WebSocket handler
- Timezone offset in scheduled reports

### Changed
- Upgraded auth library to v3.0
- Refactored database connection pooling

### Performance
- 40% faster search with new index strategy
- Lazy-load dashboard widgets

### Documentation
- Added API reference for /webhooks endpoint
- Updated deployment guide for Docker

Formatting rules:

• Group by category (Added, Fixed, Changed, Removed, Performance, etc.)
• Most important categories first (Breaking → Added → Fixed → Changed)
• Each entry is a single line starting with -
• Include issue/PR references when found: (#123)
• If a commit has a scope, include it: **auth**: Add rate limiting
• Skip trivial commits (merge commits, "WIP", version bumps) unless they contain useful info
• Write entries in past tense for consistency within the changelog, but present tense is also fine — match the project's existing style if a CHANGELOG.md exists

Step 6: Present and Apply

Show the changelog first and ask the user which actions to take:

1. Update CHANGELOG.md: Prepend the new section to the existing file (or create it)
2. Create git tag: git tag -a v1.4.0 -m "Release v1.4.0"
3. Create GitHub Release: Use gh release create with the changelog as notes
4. Just show it: Display without modifying any files

Updating CHANGELOG.md

If the file exists, insert the new version after the header:

# Check existing format
head -20 CHANGELOG.md

Preserve the existing format. If creating a new CHANGELOG.md:

# Changelog

All notable changes to this project will be documented in this file.

The format is based on [Keep a Changelog](https://keepachangelog.com/),
and this project adheres to [Semantic Versioning](https://semver.org/).

## [Unreleased]

## [1.4.0] - 2026-02-14
...

Creating a GitHub Release

gh release create v1.4.0 \
  --title "v1.4.0" \
  --notes "$(cat <<'EOF'
<changelog content here>
EOF
)"

For pre-releases:

gh release create v2.0.0-beta.1 --prerelease --title "v2.0.0-beta.1" --notes "..."

Edge Cases

• Monorepo: Ask which package/directory to scope the changelog to. Use git log -- <path> to filter.
• No new commits: Report "No changes since last release" rather than generating an empty changelog.
• Very large changelog (100+ commits): Summarize by category with counts, show top 5 most impactful per category, and offer to show the full list.
• Squash-merged PRs: The squash commit message often contains PR details — parse these for richer entries.
• Release candidates: Support pre-release tags like v1.4.0-rc.1.
• Existing CHANGELOG.md with different format: Match the existing format rather than forcing Keep a Changelog.

Tips for Great Changelogs

A changelog is for humans reading it, not for machines parsing git log. Every entry should answer: "What does this mean for me as a user/developer?"

Bad: refactor: extract helper function Good: Improved startup time by lazy-loading configuration

Bad: fix: null check in processOrder Good: Fixed crash when processing orders with empty shipping address (#789)

Transform implementation details into user-facing impact.