Token导航 LogoToken导航TokenDH.com
研究检索需要联网github未标认证来源可访问许可证需确认审计通过

release-notes发行说明

Agent Skill

release-notes 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

192

周安装

8

GitHub Stars

公开资料未说明

下载量

64
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:release-notes(发行说明)
来源仓库:https://github.com/devarfeen/agent-skills-kit
仓库路径:skills/release-notes
安装命令:
npx skills add https://github.com/devarfeen/agent-skills-kit --skill release-notes
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/devarfeen/agent-skills-kit --skill release-notes

简介

生成结构化的发行说明,聚焦业务影响、用户接触点和具体结果描述。

  • 适合在 Codex、Claude、Cursor、Gemini CLI 中需要总结变更时使用。
  • 强调用产品语言命名功能,避免内部技术细节和文件路径。
  • 安装方式:npx skills add 指定 GitHub 仓库路径;输出 Markdown 至 changelog/ 目录。
  • 仅包含有实际更改的项目,并按日期或特征分类保存。

SKILL.md

Release Notes

Purpose

Generate clear, structured release notes or session summaries for Project Managers based on development activity.

The skill can produce summaries from:

  • Git commits
  • Current development session
  • Completed feature work

Outputs must explain what changed, why it changed, and the improvement delivered, using simple language suitable for non-technical stakeholders.

Default audience is Project Managers and operations stakeholders, not engineers.

Writing Style Rules

The #1 priority of this skill is clarity for non-technical readers. Every sentence must pass this test: "Would a PM or QA person understand this without asking a developer?"

Absolute Rules

  1. Use plain, everyday words. No jargon. No corporate-speak. No engineering terms.

- BAD: "Standardized how scanning is prepared before each RFID lookup" - GOOD: "The app now checks the scanner is ready before starting a scan"

  1. Say what the user sees or does. Name the screen, button, field, or page.

- BAD: "Concrete touchpoints called out in the workstream docs include the signin page" - GOOD: "Affects the Login page, Sign Up page, and Reset Password page"

  1. One idea per bullet. Each bullet = one short sentence. Never a paragraph.

- BAD: "Two implementation workstreams were documented and broken down for execution covering username lowercasing and consistent password show/hide behavior across login, signup, reset, invite, and user-management screens." - GOOD: - "Usernames will be saved as lowercase so 'John' and 'john' are treated the same" - "All password fields now have a show/hide toggle"

  1. Feature names must describe what changed, not how it was done.

- BAD: "Auth Hardening Workstreams Prepared" - GOOD: "Login and Password Improvements Planned" - BAD: "Workspace Process Rules Clarified" - GOOD: "Updated Internal Team Workflow Guide"

  1. No filler phrases. Remove words that add no meaning.

- Remove: "formally prepared", "in a clear scope", "as a result of", "in order to" - Remove: "it should be noted that", "this ensures that", "with a clear"

  1. Avoid abstract language. Be specific about what happened.

- BAD: "reduces the risk of one screen being fixed while another screen keeps the same login or password-typing problem" - GOOD: "All login screens now behave the same way"

  1. Problem section = what the user experienced. Not what the code lacked.

- BAD: "Without a written design and task breakdown, this type of security-sensitive work is easier to implement inconsistently" - GOOD: "Some login pages let you toggle password visibility, others didn't. Usernames 'John' and 'john' were treated as different accounts."

  1. Impact section = what is better now. One short sentence per bullet. Concrete outcome.

- BAD: "Engineering now has approved-ready implementation notes for sensitive auth work instead of relying on ad hoc fixes" - GOOD: "Login will work the same way across all apps"

  1. Max 2 bullets per section (Problem, Change, Impact). If you need more, you're being too detailed — combine or simplify.
  2. Write like you're telling a coworker what you did today over a coffee. Not like you're writing a formal document.

Banned Words and Phrases

Never use these in release notes output:

  • workstream, artifact, canonical, process drift, touchpoint
  • formally, standardized, operationally, implementation
  • ad hoc, scope, aligned, resolution
  • "it is suitable for", "this reduces the chance of"
  • "without X, Y is easier to Z" (rewrite as a direct statement)
  • "not explicitly visible in commit history"
  • Any phrase longer than 15 words in a single bullet

Writing Test

Before finalizing each feature entry, mentally check:

  • Can a PM read this in 10 seconds and know what changed?
  • Can a QA person read this and know what to test?
  • Would you understand this if you had never seen the code?

If any answer is no, rewrite it simpler.

Supported Generation Modes

1. Date-Based Release Notes (Git History)

Generate release notes from Git commits for a specific date.

Example prompts:

  • Generate release notes for 11 March 2026
  • Create changelog for 11 March for all projects
  • Generate release notes for Partners App on 11 March

Behavior:

  1. Read git commit history
  2. Filter commits by the requested date
  3. Group commits by product, app, or module
  4. Combine related commits into a single logical improvement summary

2. Session-Based Development Summary

Generate a summary of changes made in the current development session.

Example prompts:

  • Summarize what we changed this session
  • Create PM update for today's work
  • Generate session changelog

Behavior:

  1. Review files modified in the session
  2. Review developer notes, markdown previews, and code diffs
  3. Combine related edits into logical improvements
  4. Present them grouped by product or app

3. Feature-Based Summary

Generate release notes for a completed feature or task.

Example prompts:

  • Write release notes for the RFID scanning improvements
  • Summarize the Stock Take lookup changes
  • Create PM update for the asset lookup improvements

Behavior:

  1. Identify commits related to the feature
  2. Combine commits into a single feature explanation
  3. Explain Problem → Change → Impact

Git Data Collection

The skill must NOT run git fetch, git pull, or any command that modifies the local git state. It only reads what is already available locally.

Multi-Repo Workspaces

If the workspace contains multiple repositories (multi-root workspace with several git repos):

  1. Find all git roots. Walk the workspace folder tree and identify every directory that contains a .git folder. Each is a separate project repo.
  2. Run git log in each repo separately. Do not assume the workspace root is the only git repo.
  3. Collect commits from all repos, then group by product/project.

Missing a repo means missing an entire project's release notes. Always scan for all git roots.

Git Log Command

Use this command pattern in each repo to get commits for a specific date:

git log --all --after="YYYY-MM-DDT00:00:00" --before="YYYY-MM-DDT23:59:59" --oneline --no-merges
  • --all includes commits from all local branches (catches work merged into any branch).
  • --no-merges skips merge commit noise like "Merge pull request #123".
  • For richer detail, use --format="%H %s" instead of --oneline.

Finding Repos Automatically

find <workspace-root> -name ".git" -type d -maxdepth 3 | sed 's/\/.git$//'

Then loop through each path and run git log inside it.

Important

If a user mentions a specific project/repo and the skill finds no commits, report that explicitly. Do not silently omit projects — say "No commits found for on. The local branch may not be up to date — try running git pull in that repo."

Automatic Commit Clustering

When generating release notes from Git history, do not summarize each commit separately unless the commits clearly represent unrelated work.

Instead, automatically cluster related commits into one logical change so the output reads like PM-facing release notes rather than raw engineering history.

Goal

Convert multiple implementation commits into one stakeholder-friendly summary using:

Problem → Change → Impact

Cluster Commits Together When They Share

  • the same product, app, or module
  • the same feature or workflow
  • the same bug or user problem
  • the same file area or code path
  • the same implementation objective across several commits

Keep Commits Separate When They Represent Different

  • products
  • workflows
  • features
  • bugs with unrelated causes
  • independent user-facing changes

Stakeholder Summary Section

Every release notes file must start with a Stakeholder Summary section at the top, before the detailed notes.

This section is the quick-read version for PMs and stakeholders. It combines the Summary and Change from each feature entry into short bullet points grouped by date and project code.

Format

# Stakeholder Summary

Date: DD Month YYYY

PRODUCT-CODE-1

- What changed + why it matters in one sentence
- What changed + why it matters in one sentence

PRODUCT-CODE-2

- What changed + why it matters in one sentence

Rules

  • Start with Date: DD Month YYYY (omit for session summaries that have no date).
  • Product codes in ALL CAPS (e.g., PARTNERS-APP, PORTFOLIO-WEB, SERVER-REVERSE-PROXY). No markdown heading syntax — plain text on its own line.
  • Each change is a markdown bullet (-). One bullet per feature.
  • Each bullet combines Summary + Change into one sentence that answers: "What changed and what does that mean for the user?"
  • Write bullets so a PM can scan the entire list in under 30 seconds.
  • Group by product/app, same grouping as the detailed section below.
  • Separate the Stakeholder Summary from the Detailed Release Notes with a horizontal rule (---).
  • The detailed section follows under a # Detailed Release Notes heading.

Manual QA Steps Section

Every feature entry in the detailed release notes must include a Manual QA Steps subsection.

Finding Existing QA Steps

Before generating QA steps, search for an existing 5-manual-qa-steps.md file:

  1. Look in the feature's specs folder (e.g., specs/5-manual-qa-steps.md, specs/<feature-name>/5-manual-qa-steps.md, or any **/specs/5-manual-qa-steps.md path relative to the feature's code).
  2. If found, link to it: See [Manual QA Steps](relative/path/to/specs/5-manual-qa-steps.md)
  3. Also include the key steps from that file inline so the release note is self-contained.

Generating QA Steps

If no 5-manual-qa-steps.md file exists, generate 3–5 practical manual QA steps based on the change:

  • Each step must be: Action → Expected Result
  • Steps should cover the primary happy path and one edge case.
  • Write steps a manual tester can follow without reading the code.
  • Keep language simple and specific (name the screen, button, or field).

Clustering Heuristics

Infer that commits belong together when several commits:

  • touch the same files or folders
  • use similar wording in commit messages
  • refer to the same screen, flow, or process
  • appear to be iterative work on one feature
  • are clearly part of one bugfix sequence

Summarization Rules

For each cluster:

  1. identify the shared problem
  2. combine all related commits into one logical summary
  3. explain the change in non-technical language
  4. describe the resulting improvement
  5. avoid commit-by-commit narration unless necessary

Project Discovery

Project names and codes may be found from:

  • repository documentation
  • workspace metadata
  • git repository paths
  • folder structure
  • user-provided context

Output Format

Output must follow the exact structure below. Do not reorder sections.

For date-based release notes, include a date line at the top.

Date: <DD Month YYYY>

<Product / App Name>

Summary

  • One sentence: what is now better in day-to-day operations.

Problem

  • What users or operations teams were experiencing before this change
  • Include the business consequence (delays, rework, missed scans, confusion, etc.)

Change

  • Explain the solution in product/workflow language
  • Name affected workflows or screens in plain language
  • Include 1 to 3 concrete user-visible touchpoints when available: setting, view/page, component/screen section, URL/route, or visible UI element
  • If logic changed, add one simple sentence in plain language for a 5th grader
  • Do not include file paths, class names, package names, or internal module names

Impact

  • State concrete outcomes, not generic claims
  • Prefer specific statements such as "fewer failed scans during stock take" over "improved reliability"

Scope

  • Who is affected (teams, roles, or flows) and where the change applies

Commits Included

  • List commit hashes only at the end for traceability

If there are multiple features under one project, repeat the feature block:

  • Summary
  • Problem
  • Change
  • Impact
  • Scope
  • Commits Included

Do not place feature sections above the project header.

Project Inclusion Rule

Include a project section only if that project has at least one confirmed change in the selected commits/session scope.

Do not include unchanged projects in the final generated file.

File Output Rules

Always save the generated markdown under the root changelog/ folder.

  1. Date-based summary or release notes:

- Path: changelog/DD-Month-YYYY.md - Example: changelog/12-March-2026.md

  1. Feature-based summary:

- Path: changelog/Feature-Name.md - Use Title-Case words joined with hyphens - Example: changelog/RFID-Scanner-Reliability.md

  1. Session summary without explicit feature name:

- Path: changelog/DD-Month-YYYY.md

If changelog/ does not exist, create it before writing output.

Section Hierarchy Rule

The hierarchy must be:

  1. Date line (date-based only)
  2. Project Name
  3. Feature Name
  4. Child sections under feature: Summary, Problem, Change, Impact, Scope, Commits Included

Never output child sections directly under the date line or without a feature heading.

Writing Guidelines

  • Write for non-technical stakeholders
  • Translate implementation details into user-facing outcomes
  • Avoid unnecessary engineering terminology
  • Focus on what improved and how it affects operations
  • Prefer short bullet points
  • Avoid vague wording such as "enhanced", "optimized", or "improved" without saying what changed in behavior
  • If exact metrics are unavailable, use directional impact grounded in observed behavior

Git Detail Extraction Rules

When generating from git commits/diffs, explicitly extract and summarize user-visible change locations when present.

Look for and report in plain language:

  • Main setting changed (for example scanner power, timeout, toggle defaults)
  • View/page/screen affected
  • Component or visible screen section affected
  • URL or route changed (if available and user-visible)
  • Visible element changed (button label, status text, warning, banner, modal, etc.)

Prefer naming what users see over code object names.

If details are not available in commits/diffs, say "Not explicitly visible in commit history" instead of guessing.

Simple Logic Explanation Rule

If a feature includes programming logic changes, add a plain-language explanation inside the Change section.

Format:

  • Simple logic explanation: <one sentence a 5th grader can understand>

Example:

  • Simple logic explanation: Raised scanner power from 20 to 30 and made the app switch it on every time before scanning.

Translation Rules

When source input contains engineering details, rewrite them as operational meaning.

Examples:

  • "added API classes/interfaces" -> "connected scanner capabilities so app screens use the same scan process"
  • "refactored useRfidScanner" -> "standardized scanner setup to reduce inconsistent scan starts"
  • "added test coverage" -> "reduced regression risk by validating scanner setup behavior"

Do not copy code identifiers directly into PM output unless the user explicitly asks for technical detail.

Quality Check Before Finalizing

Before returning output, verify:

  1. A PM can understand it without code context.
  2. Each impact bullet describes an observable operational outcome.
  3. At least one bullet names affected workflows or teams.
  4. Technical identifiers are removed from the main narrative.
  5. Commit hashes, if included, appear in a final traceability line only.
  6. Output follows Date -> Project -> Feature -> Child sections hierarchy.
  7. File is saved in changelog/ with correct naming convention.
  8. Change section includes user-visible touchpoints when commit history provides them.
  9. Logic changes include one simple sentence understandable by non-technical readers.
  10. Final file includes only projects that have confirmed changes.

Fallback

If clustering is uncertain, keep commits separate rather than merging unrelated work. Do not speculate about changes not present in the workspace, git history, or provided context.

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

34.61%
按下载量换算22

Claude

31.67%
按下载量换算20

Cursor

20.25%
按下载量换算13

Gemini CLI

9.09%
按下载量换算6

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills