changelog-analyzer

name	changelog-analyzer
description	Analyze git commits in a range by examining actual code diffs (not commit titles) and produce a ranked list of improvements and optimizations. Use when the user wants a changelog, wants to summarize recent changes, asks what changed since a commit, or wants to review project progress.

Changelog Analyzer

Analyze all code changes between two git commits and produce a ranked summary of improvements and optimizations based on actual code diffs, not commit messages.

Parameters

• Start commit: The earliest commit hash (inclusive). Default: e0be0387a636a93a062cee621fd78b83dfccb06e
• End commit: The latest commit. Default: HEAD

If the user specifies a different range, use their values.

Workflow

Step 1: Enumerate commits and changed files

# List all commits in range (inclusive of start)
git log --oneline <start>~1..HEAD

# Get overall stats
git diff --stat <start>~1..HEAD

Step 2: Analyze code diffs in batches

The diff may be very large. Process it in logical batches by directory or feature area:

# Get diff for a specific path
git diff <start>~1..HEAD -- <path>

# Or per-commit diffs when granularity matters
git show --stat <commit-hash>
git show <commit-hash> -- <file>

Key analysis areas (check each):

1. apps/web/src/components/ — UI components, UX improvements
2. apps/web/src/services/ — Service layer, storage, transcription
3. apps/web/src/core/ and apps/web/src/lib/ — Core logic, utilities
4. apps/web/src/hooks/ — React hooks, state management
5. apps/web/src/stores/ — Zustand stores, state architecture
6. apps/web/src/types/ — Type definitions, data model changes
7. apps/web/src/app/ — Pages, routes, API endpoints
8. packages/ — Shared packages
9. Config files, CI/CD, build system

Step 3: Classify each change

For each meaningful change found in the diffs, classify it into one of these categories:

Category	Icon	Description
New Feature	✨	Entirely new capability added
Performance	⚡	Speed, memory, or rendering optimization
Bug Fix	🐛	Corrected broken or incorrect behavior
UX Improvement	🎨	Better user experience, UI polish
Refactor	♻️	Code restructuring without behavior change
Architecture	🏗️	Structural or design pattern improvement
DX / Tooling	🔧	Developer experience, build, CI/CD
i18n	🌐	Internationalization, localization
Security	🔒	Security hardening
Cleanup	🧹	Dead code removal, dependency cleanup

Step 4: Determine importance

Rate each change on impact (1-5):

Score	Criteria
5	Core feature, affects most users, or prevents data loss
4	Significant feature or major UX/perf improvement
3	Notable improvement, moderate user impact
2	Minor enhancement, affects subset of users
1	Cosmetic, cleanup, or internal-only change

Step 5: Produce the report

Output a markdown report with this structure:

# Changelog: <start-hash-short>..<end-hash-short>

**Period**: <first-commit-date> → <last-commit-date>
**Commits**: <count> | **Files changed**: <count> | **+<insertions> / -<deletions>**

---

## Key Changes (by importance)

### ⭐⭐⭐⭐⭐ Critical / High Impact

#### ✨ [Feature/Change Title]
- **What changed**: Concise description based on actual code diff
- **Files**: List of key files modified
- **Impact**: Why this matters to users or the codebase

### ⭐⭐⭐⭐ Significant

...

### ⭐⭐⭐ Notable

...

### ⭐⭐ Minor

...

### ⭐ Cosmetic / Internal

...

---

## Summary Statistics

| Category | Count |
|----------|-------|
| New Features | X |
| Performance | X |
| Bug Fixes | X |
| ... | ... |

Important Rules

1. Read actual diffs — Never rely solely on commit messages. Always git diff or git show to see what really changed.
2. Group related commits — Multiple commits may contribute to a single logical change. Merge them into one entry.
3. Be specific — Reference actual function names, components, or files that changed.
4. Skip noise — Ignore auto-generated files (lock files, build artifacts) and trivial whitespace changes.
5. Use subagents for parallelism — When the diff is large, launch multiple explore/generalPurpose subagents to analyze different directories concurrently.
6. Output in Chinese — The final report should be written in Chinese (中文), matching the user's language preference.