- Home
- Skills
- Technical Documentation
- Release Notes Writer (CHANGELOG + GitHub Release from git hi
Release Notes Writer (CHANGELOG + GitHub Release from git hi
Ask: **"Write the release notes for v1.0.0..v1.1.0"** (or "since the last tag", "for September", "the last 20 commits").
$1
Works with the AI Agents you already use
Release Notes Writer (CHANGELOG + GitHub Release from git hi
Example session with this skill installed
Repo: rf-marketplaces-mcp (our public MCP server for Wildberries analytics, local clone). Write the release notes for v1.0.0..v1.1.0: the CHANGELOG.md entry, the GitHub Release body, and a short X post announcing it. Credit contributors.
- Read your context and instructions
- Compiled the release notes writer
Real output from 23 Oct 2026. A fresh agent got only this skill's files and the request, and ran only read-only git commands. Shortened for the listing.
Range: v1.0.0..v1.1.0, 4 commits, 2026-09-29 → 2026-10-23. PR titles aren't available (gh not installed, no merge commits), so this is written from the commits.
Suggested version: MINOR → 1.1.0. Reason: fabea44 adds optional output fields (last_known_price_rub, last_known_price_date, unavailable_fields, data_note, cheapest_last_known) and makes the broken tools work again; no tool, parameter, env var or entry point is removed or renamed. Judgement call: price and stock fields can now be null, which is strictly breaking for typed clients, so it is flagged under "Before you upgrade". Matches the v1.1.0 tag and package.json.
CHANGELOG.md entry (excerpt)
## [1.1.0] - 2026-10-23
### Added
- `last_known_price_rub` and `last_known_price_date` in `wb_product`, `wb_compare`, `wb_price_watch` and `wb_card_audit` results when the current price is unavailable. <!-- fabea44 -->
- `unavailable_fields` and `data_note` explaining which data WB no longer publishes. <!-- fabea44 -->
- `npm run test:offline` runs all 17 tools against recorded WB responses without network access. <!-- fabea44 -->
### Changed
- While `card.wb.ru` is unavailable, price, discount, stock, sizes, delivery time and seller rating are `null`; clients that expect numbers must handle `null`. <!-- fabea44 -->
- The server still tries `card.wb.ru` first and switches back to full data automatically if WB reopens it. <!-- fabea44 -->
### Fixed
- `wb_product`, `wb_compare`, `wb_price_watch`, `wb_reviews`, `wb_review_insights` and `wb_card_audit` work again after Wildberries started answering `card.wb.ru` with 403: cards are now built from public CDN files and the feedbacks API, without bypassing WB's anti-bot protection. <!-- fabea44 -->
Release notes (GitHub Release body)
## What's new in 1.1.0
**Wildberries product tools work again.** In October 2026 WB closed its card API to outside requests, which broke six tools. They now build product cards from WB's open CDN files and reviews instead. Still no token needed.
### New
- When the current price isn't available, you get the last known price from WB's price history, with its date.
- Responses say what is missing: `unavailable_fields` lists the empty fields and `data_note` explains why.
### Before you upgrade
- Until WB reopens its card API, current price, discount, stock, sizes, delivery time and seller rating come back as `null`. Sellers can still see their own current prices with `wb_seller_prices`.
Announcement — X (274/280 with the link counted as 23)
rf-marketplaces-mcp 1.1.0: the Wildberries product tools work again after WB closed its card API. Cards are now built from WB's open CDN files and reviews.
Current price and stock are null for now; you get the last known price and its date instead.
[release link]
Sources (shortened): every line → fabea44 (diff: src/wb-public.js, README). Omitted (internal): a88b435, ae5eaf8 README badges; 6b608ac README contact section.
Second real run, same date, on a repo with no changelog and messy messages: 10 commits → 6 backfilled sections + Unreleased; "Release 1.1.0" recognised as a version bump only; a default change (HTTP server now listens on 127.0.0.1) caught as
BREAKING with the fix (MCP_HTTP_HOST=0.0.0.0) and a question instead of a guess; "since the last tag" → "nothing worth releasing: one README-only commit".
Connects securely to your tools. The creator never sees your data.
About this skill
Writing release notes means rereading commit messages like "fix", "update" and "Release 1.1.0" and turning them into something a user understands. This skill teaches your AI agent to do that from the real history. It reads the commits, merged PR titles (if gh is available) and the diffs where the messages don't say enough.
Ask: "Write the release notes for v1.0.0..v1.1.0" (or "since the last tag", "for September", "the last 20 commits"). You get:
- A CHANGELOG.md entry in Keep a Changelog 1.1.0 format: Added / Changed / Deprecated / Removed / Fixed / Security, ISO dates and compare links. Breaking changes are marked BREAKING with what to do about them. A range that spans several tags gets one section per release, and commits after the last tag go under
Unreleased. - A suggested SemVer bump with the reason, and whether the existing tag matches.
- GitHub / GitLab Release notes in plain language, grouped by what users notice: a lead line, New, Improved, Fixed, and "Before you upgrade" for breaking changes. CI, tests, refactors, typo fixes, badges and version-bump commits are left out.
- A store "What's new" on request, within the App Store's 4000-character and Google Play's 500-characters-per-language limits (checked against Apple's and Google's documentation on 23 Oct 2026), with character counts.
- An announcement post on request, for X (280 weighted characters), LinkedIn or Telegram, built only from what is in the release notes.
It never invents features. Every bullet carries its source (short SHA or PR number) in a hidden HTML comment, and a Sources appendix maps each line to its commits. Commits that produced no line are listed as "omitted (internal)" with the reason. If the effect of a change on users isn't clear, it keeps the change out of the user-facing notes and asks you.
Read-only by design. It may run only git log, show, diff, tag --list, describe, rev-parse, rev-list, shortlog, for-each-ref and remote -v. It never commits, tags, checks out, fetches or pushes, and never creates the GitHub Release for you. It updates CHANGELOG.md only if you ask, and never commits it.
Works with any commit style. Conventional Commits are mapped directly; messy histories are handled too: one commit with several changes is split, a fix and its follow-up are merged, reverted pairs are dropped, and "WIP" / "Release 1.1.0" commits are read from the diff. Contributor credits use names and handles, never email addresses.
Pure instructions in Markdown: it works in any agent that supports the SKILL.md standard and can run git.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Technical Documentation.
- It only knows what's in git (and PR titles if gh is available). Changes made outside the repo, such as server config or backend flags, are invisible to it.
- With very large ranges (hundreds of commits) it reads diffs selectively, so check the "omitted" list in the Sources appendix.
- The App Store (4000) and Google Play (500 per language) limits were checked against official docs on 23 Oct 2026. Other stores' limits and LinkedIn and Telegram post lengths were not verified, and it says "check the current limit" for them.
- SemVer suggestions involve judgement for defaults, null-able fields and security hardening. It explains its reasoning, but the call is yours.
- It doesn't publish anything. You paste the texts into GitHub, the stores or social media yourself.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 25 days ago
- Software
GitTerminal access
- Optional
GitHub CLI
Any recent Git works. GitHub CLI, signed in, adds merged PR titles.
Permissions
- Reads your local repository with read-only git commands only. It never commits, tags, checks out, resets, fetches or pushes, and never changes git config. - Writes no files unless you ask it to update CHANGELOG.md. Even then it only inserts the new section, and it never commits. - No network access needed. The optional gh pr list is a read-only call to GitHub using your own login. - It never prints contributor email addresses. - The skill is four Markdown files of instructions with no code and no secrets.