Changelog Weaver#

Transform git history into polished, human-readable changelogs and release notes. Parse Conventional Commits, auto-categorize changes by type, deduplicate similar commits, and generate output in multiple formats — from CHANGELOG.md to messaging-platform announcements.

First-Success Path (30 seconds)#

bash

# 1. Collect commits since your last git tag
python3 scripts/weaver.py collect --from-latest-tag -o commits.json

# 2. Generate a CHANGELOG.md
python3 scripts/weaver.py generate -i commits.json -f changelog

# 3. That's it — a structured changelog ready for review

To add AI-powered semantic rewriting (human-readable descriptions instead of raw commit messages), load commits.json into chat and say: "Rewrite all commit descriptions to be user-facing" — copy the ai_rewrites back into the JSON, then regenerate.

Workflow#

Step 1: Collect Change Sources#

Decide what to include in the changelog:

bash

# Between two tags
weaver.py collect --from v2.3.0 --to v2.4.0 -o changes.json

# Since the latest tag (auto-detected)
weaver.py collect --from-latest-tag -o changes.json

# By date range
weaver.py collect --since "2026-06-01" --until "2026-06-15" -o changes.json

# All commits on a branch
weaver.py collect --from main --to feature/big-refactor -o changes.json

The script reads git log, filters merge commits by default, and outputs structured JSON.

If PR/issue data is available (GitHub/GitLab), the AI can enrich the changelog with PR titles and linked issue numbers. Mention this in your prompt.

Step 2: Smart Classification#

The script automatically classifies every commit by Conventional Commits type:

PrefixCategoryExamplefeat:Featuresfeat(auth): add OAuth2 login``fix:Bug Fixesfix: resolve race condition in cache``docs:Documentationdocs: update API reference``chore:Choreschore: bump dependencies``refactor:Refactoringrefactor: extract payment module``perf:Performanceperf: optimize query planner``test:Teststest: add integration tests``build:Build Systembuild: migrate to vite``ci:CI/CDci: add deploy workflow``revert:Revertsrevert: rollback auth changes``security:Securitysecurity: patch XSS vector``deps:Dependenciesdeps: update react to 19.0

Commits that don't follow the convention are grouped under other — the AI can still reclassify them by analyzing the commit message content.

Step 3: AI Semantic Rewrite#

Load the generated JSON into chat for the AI to perform semantic rewriting:

Prompt: "I have a changelog JSON at commits.json. Rewrite all commit descriptions to be user-facing, replace technical jargon with plain language, and suggest category corrections for any misclassified commits. Output as ai_rewrites mapping."

The AI will:

• Convert fix: resolve race condition in user auth token refresh → "Fixed login sessions expiring prematurely during token refresh"
• Shorten verbose descriptions to one line
• Merge related entries that span multiple commits
• Flag entries that need human review

Copy the ai_rewrites object back into the JSON under the top-level key "ai_rewrites", then regenerate.

Step 4: Deduplicate & Merge#

The script merges adjacent commits of the same type with ≥70% description token overlap. For manual merging, tell the AI:

Prompt: "Merge any commits in commits.json that clearly belong to the same feature or fix. Update the count field."

The AI will identify related commits (e.g., 3 commits all fixing the same date-picker bug) and consolidate them.

Step 5: Identify Breaking Changes#

Breaking changes are detected automatically via:

• ! suffix: feat(api)!: remove deprecated endpoint
• Footer: BREAKING CHANGE: drops support for Node 16

The AI also scans other commits for breaking-change language:

Prompt: "Scan all commit descriptions in commits.json for breaking changes not marked with ! or BREAKING CHANGE footer. Add breaking: true and a breaking_detail for each."

Step 6: Generate Output Formats#

Once the JSON is ready (with optional AI rewrites), generate output in any format:

bash

# Keep a Changelog format (CHANGELOG.md)
weaver.py generate -i commits.json -f changelog -o CHANGELOG.md

# GitHub Release body
weaver.py generate -i commits.json -f release -o RELEASE_NOTES.md

# Plain text for 飞书/钉钉
weaver.py generate -i commits.json -f plain -p feishu -o announcement.txt

# JSON for CI/CD pipelines
weaver.py generate -i commits.json -f json -o changelog.json

Step 7: Quality Check#

Before publishing, run a quality pass:

Prompt: "Review CHANGELOG.md for issues: missing entries (compare to git diff --stat), broken links, misclassified items, entries that need human confirmation. Output a QA checklist."

Key checks:

• Every significant change has an entry
• No duplicate or contradictory entries
• Breaking changes are clearly highlighted
• Links to PRs/issues are valid
• Release date and version are correct

Step 8: Deliver & Integrate#

Write the final output:

bash

# Write to project root
cp CHANGELOG.md /path/to/project/CHANGELOG.md

# Or prepend to existing changelog (keep history)
cat CHANGELOG.md <(tail -n +2 /path/to/project/CHANGELOG.md) > /tmp/merged.md

For CI/CD integration, consume the JSON output:

yaml

# GitHub Actions example
- name: Generate Changelog
  run: |
    python3 scripts/weaver.py collect --from-latest-tag -o changelog.json
    python3 scripts/weaver.py generate -i changelog.json -f release >> $GITHUB_STEP_SUMMARY

Sample Prompts#

1. Basic Release Changelog#

"Generate a changelog for the v2.4.0 release from git history since v2.3.0. Use Keep a Changelog format. Our project is at ~/projects/myapp."

What the skill does: Runs collect, loads JSON, suggests rewrites, generates CHANGELOG.md.

2. Weekly Team Report#

"Generate a weekly summary of changes to the backend repo since last Monday. Format as a plain text message for DingTalk — Chinese, concise, grouped by feature/fix/chore."

What the skill does: Runs collect --since "last Monday", generates plain -p dingtalk, AI rewrites descriptions in Chinese.

3. GitHub Release with AI Rewrites#

"Collect all commits since v1.5.0 in ~/projects/api-server. Rewrite every description to be user-facing. Generate a GitHub Release body with contributor shoutouts. Add a 'What's New' highlight section at the top."

What the skill does: Collects commits, AI rewrites descriptions, generates release format with highlights.

4. Breaking Change Audit#

"Scan commits between v3.0.0 and v3.1.0 for breaking changes — both marked and unmarked. Generate a migration guide section in the changelog."

What the skill does: Collects commits, AI scans for breaking changes, adds migration notes.

5. PR-Enriched Changelog#

"Generate a changelog from git history, but also pull PR titles from GitHub for the myorg/myrepo project. Match commits to PRs by branch name and reference numbers. Use PR titles for the changelog entries instead of commit messages."

What the skill does: Collects commits, AI cross-references with GitHub PR API, uses PR titles.

Real-World Examples#

Example 1: Open Source Library Release#

Scenario: You maintain react-data-grid and are releasing v4.2.0. The repo uses Conventional Commits. You need a professional CHANGELOG.md following Keep a Changelog format.

Input:

bash

python3 scripts/weaver.py collect \
  --repo ~/projects/react-data-grid \
  --from v4.1.0 --to v4.2.0 \
  -o v4.2.0-commits.json

The JSON contains 47 commits: 12 feat, 18 fix, 5 docs, 4 chore, 3 refactor, 2 perf, 1 breaking, 2 other.

AI Prompt: "Load v4.2.0-commits.json. Rewrite all descriptions to be concise and user-facing. The audience is React developers. Merge the 3 commits about 'column resize' into one entry. Flag the breaking change for prominent display."

Output (CHANGELOG.md excerpt):

markdown

# Changelog

## [4.2.0] - 2026-06-15

### ⚠️ Breaking Changes
- Removed `legacyRowRenderer` prop — migrate to `rowRenderer` with the new `RowConfig` interface (#892)

### ✨ Features
- Column resize handles: drag to resize any column, double-click to auto-fit (#845, #867, #881) — @sarahchen
- Virtual scrolling for 100K+ row datasets with <50ms render time (#878) — @mikez
- Multi-select cells with Shift+Click and Ctrl+Click (#903) — @alexk

### 🐛 Bug Fixes
- Fixed date filter not respecting custom timezone (#890) — @jamesw
- Fixed column reorder breaking when a column is hidden (#895) — @linday

### 🙏 Contributors
@alexk, @jamesw, @linday, @mikez, @sarahchen

---

Example 2: Team Weekly Report for 飞书#

Scenario: Every Friday, the platform team sends a summary of the week's changes to the company-wide 飞书 group. The repo ( ~/projects/platform) has 30-50 commits per week with mixed commit quality.

Input:

bash

python3 scripts/weaver.py collect \
  --repo ~/projects/platform \
  --since "2026-06-08" --until "2026-06-15" \
  -o weekly-2026-06-15.json

AI Prompt: "Load weekly-2026-06-15.json. Rewrite all descriptions in Chinese, grouping by feature/fix/infra. Focus on business impact. Generate plain text for 飞书, limit to top 10 entries."

Output (announcement.txt):

text

📦 平台周报 (2026-06-08 ~ 2026-06-15)

【新功能】
  • 用户画像页新增「活跃度趋势」图表，支持按周/月切换
  • 数据导出支持自定义字段选择，减少导出体积 60%

【问题修复】
  • 修复 Safari 浏览器下 Dashboard 加载白屏问题（影响约 5% 用户）
  • 修复大文件上传偶发超时，现在支持 500MB 以下文件稳定上传
  • 修复深夜时段定时任务漏执行的问题

【基础设施】
  • CI 构建时间从 12 分钟优化到 6 分钟
  • 升级 Redis 集群至 7.2，内存使用降低 30%

👥 本周贡献者: 张三, 李四, 王五, 赵六

---

Example 3: Breaking Change Migration Guide#

Scenario: Your API server v3.0 introduces several breaking changes. The changelog must include a detailed migration guide for downstream consumers.

Input:

bash

python3 scripts/weaver.py collect \
  --repo ~/projects/api-server \
  --from v2.9.0 --to v3.0.0 \
  -o v3.0.0-commits.json

AI Prompt: "This is a major version bump. Scan all commits for breaking changes — both marked and unmarked. For each breaking change, write a migration note with 'Before' and 'After' code snippets. Generate the changelog with a prominent 'Migration Guide' section at the top."

Output (CHANGELOG.md excerpt):

markdown

## [3.0.0] - 2026-06-15

### 📋 Migration Guide

#### 1. `/api/v1/users` → `/api/v2/users`
**Before:**
```http
GET /api/v1/users?role=admin

After:

http

GET /api/v2/users?filter[role]=admin

2. Authentication header format

Before: Authorization: Token abc123 After: Authorization: Bearer abc123

3. Webhook payload structure

The event.data field is now always an object (was sometimes a string). Update your webhook handlers to expect typeof event.data === 'object'.

---

⚠️ Breaking Changes#

• Migrated user endpoints to v2 with filter-based query params (#1204)
• Changed authentication scheme from Token to Bearer (#1189)
• Normalized webhook payload structure — event.data is always an object (#1215)

✨ Features#

• Added bulk user import API with CSV validation (#1198)
• New webhook retry mechanism with exponential backoff (#1207)

text

## Requirements

- **Python 3.9+** — standard library only (no pip dependencies)
- **git** — available on `$PATH`

## Safety

- All processing is local — no git history is sent to external services
- The script runs read-only git commands (`git log`, `git describe`)
- AI rewriting happens in-chat with explicit user review before final output
- No destructive operations — output files are written to specified paths only

## Compare with Related Skills

| Skill | Focus | How Changelog Weaver Differs |
|-------|-------|------------------------------|
| **Release Guard** | Pre-release quality checks (files, secrets, scripts) | Changelog Weaver **generates the changelog content**, not checks |
| **Git Blame Auditor** | Code attribution, contributor stats, tech debt | Changelog Weaver summarizes **what changed**, not who wrote what |
| **Doc Weaver** | Markdown → Word/PDF with templates | Changelog Weaver is **git-aware**, specialized for changelogs |

These skills are **complementary**: use Release Guard to validate a release, Changelog Weaver to document it, and Doc Weaver to format it into a polished PDF.