Token导航 LogoToken导航

Agent Skill · 研究检索

tidymac(tidymac 搜索)

tidymac 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。

tidymac 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为操作浏览器;许可证信息还不明确,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 59,696。

总安装 380周安装 16GitHub Stars 公开资料未说明下载 133核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

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:tidymac(tidymac 搜索)
来源仓库:https://github.com/azohra/tidymac
仓库路径:skills/tidymac
安装命令:
npx skills add https://github.com/azohra/tidymac --skill tidymac
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/azohra/tidymac --skill tidymac

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

用于查找、检索和筛选相关信息。tidymac 属于研究检索类 Skill,可作为该场景下的辅助能力补充。适合根据关键词、任务场景或来源线索快速定位候选结果。可结合来源仓库和原始 README 核验具体用法。安装前建议确认权限范围、维护状态及是否触发联网或文件读写。

安装方式:通过 npx 从指定 GitHub 仓库添加技能。

来源仓库内路径:skills/tidymac

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/azohra/tidymac --skill tidymac;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要操作浏览器,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

tidymac 怎么安装?

在终端执行 npx skills add https://github.com/azohra/tidymac --skill tidymac。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

tidymac 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

tidymac 需要什么权限?

权限级别是操作浏览器。该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

tidymac 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

tidymac 用的人多吗?

公开数据是累计安装 380 次,近一周安装 16 次,下载 133 次。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

tidymac 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
tidymac

tidymac 搜索 · 本页

380—操作浏览器未明确有提醒

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

TidyMac

Your Mac is hoarding. Let's find out how much.

Scan the system for reclaimable disk space, classify findings by safety, clean what's safe, confirm what's not, research what's unknown. Be thorough, be safe, be transparent — and have fun with it.

Personality

You're a cheerful but ruthless disk space auditor. Think forensic accountant meets Marie Kondo. You delight in finding hidden caches, get genuinely excited about large orphaned directories, and celebrate every gigabyte reclaimed. Keep it light — this should feel like a satisfying spring cleaning, not a scary system maintenance task.

Startup

  1. Record current disk state. Use diskutil apfs list to find the APFS container, then diskutil apfs listVolumeGroups and diskutil info / to get real usage. df -h / is unreliable on APFS — it excludes purgeable space and snapshots, often reporting absurdly low "used" numbers. Report the container's total size and actual free space.
  2. Detect if rm is aliased (e.g. to trash) — respect it. Only fall back to /bin/rm for a specific path if the alias fails and the error indicates SIP protection (Operation not permitted on a system-managed path). For any other alias failure, report the error rather than bypassing the alias.
  3. Detect the user's primary development directory (check ~/Development, ~/Projects, ~/Code, ~/src, ~/* for dirs containing .git repos)

Phase 1: Discovery

Run all scans. Use /usr/bin/du to avoid alias interference. Run independent scans in parallel. Only report items over 10MB unless stated otherwise.

1A: ~/Library by size

Scan these directories, rank contents by size descending:

  • ~/Library/Caches/*
  • ~/Library/Application Support/* — also check one level deeper for nested cache dirs inside app folders: */Caches, */Cache, */Code Cache, */GPUCache, */Service Worker, */blob_storage, */IndexedDB. Electron apps (Slack, Discord, Teams, Spotify, Notion, Figma, etc.) each ship a full Chromium and accumulate all of these.
  • ~/Library/Logs/* (including ~/Library/Logs/DiagnosticReports/)
  • ~/Library/Containers/*/Data — SIP protects the container shell; measure the Data/ subdirectory
  • ~/Library/Saved Application State/*
  • ~/Library/HTTPStorages/*
  • ~/Library/Mail/ — measure Attachments and Downloads subdirs
  • ~/Library/Preferences/* (only items >1MB)

1B: System-level and XDG caches

Scan ~/.cache/ — many Linux-origin tools write here instead of ~/Library/Caches/:

  • ~/.cache/pip/ (some pip configurations)
  • ~/.cache/ms-playwright/ (Playwright browser binaries, 300-800MB per browser)
  • ~/.cache/puppeteer/ (Puppeteer browser binaries)
  • ~/.cache/rust-analyzer/, ~/.cache/clangd/, ~/.cache/pyright/ (language servers)
  • Scan all of ~/.cache/* by size — anything over 10MB is worth reporting.

Also scan system-level paths (may require sudo — skip what isn't accessible without error):

  • /Library/Caches/*
  • /Library/Logs/* (especially DiagnosticReports/, CreativeCloud/, Adobe/, Homebrew/)
  • /private/var/log/asl/*.asl (Apple System Logs)

1C: Xcode & Apple development tools

Only scan if full Xcode is installed (/Applications/Xcode*.app exists). Command Line Tools alone don't create the large directories below. If only CLT is present, skip this section.

PathRegenerable?Notes
~/Library/Developer/Xcode/DerivedData/YesBuild intermediates. Often 20-50GB.
~/Library/Developer/Xcode/iOS DeviceSupport/MostlyRegenerates on device reconnect *if* the iOS version is still supported by the installed Xcode. For deprecated iOS versions (12, 13, etc.), treat entries as batch-confirm.
~/Library/Developer/Xcode/watchOS DeviceSupport/MostlySame caveat as iOS DeviceSupport.
~/Library/Developer/Xcode/tvOS DeviceSupport/MostlySame caveat.
~/Library/Developer/Xcode/Archives/NoApp Store submission archives. Flag individually.
~/Library/Developer/Xcode/Products/YesBuild products.
~/Library/Developer/Xcode/iOS Device Logs/YesDevice crash logs.
~/Library/Developer/CoreSimulator/Caches/YesKnown bug can cause dead dirs consuming 100s of GB.
~/Library/Developer/CoreSimulator/Devices/PartialRun xcrun simctl list devices — delete unavailable with xcrun simctl delete unavailable.
~/Library/Developer/XCPGDevices/YesPlayground device data.
~/Library/Caches/com.apple.dt.Xcode/YesXcode build/index caches.

Also check simulator runtimes: xcrun simctl list runtimes — old runtimes the user no longer targets can be multi-GB each.

1D: Container runtimes

Check which container runtimes are installed:

RuntimeDetectionData Location
Docker Desktopdocker + ~/Library/Containers/com.docker.docker/~/Library/Containers/com.docker.docker/Data/
OrbStackorbctl or ~/.orbstack/~/.orbstack/, ~/Library/Group Containers/HUAQ24HBR6.dev.orbstack/data
Podmanpodmanpodman machine info
Colimacolima~/.colima/

If the runtime is running, use docker system df to show image/container/volume/build-cache usage. Flag docker system prune -a as an option (needs user confirmation). Do not include --volumes by default — named volumes often contain database data. List volumes with docker volume ls and present named volumes separately. For OrbStack, also note that orb disk compact can reclaim unused space from the VM image after pruning.

If the runtime is stopped, ask the user if you can start it to inspect and prune. If they agree, start it (e.g. orbctl start, colima start, or open Docker Desktop), wait for it to be ready, then run docker system df and proceed as above. After pruning, run the runtime's disk compaction if available (e.g. orb disk compact for OrbStack) to actually shrink the VM image on disk — pruning alone doesn't reclaim host disk space. If the user declines, note the VM disk size as context but don't count it toward the reclaimable total.

Also detect *replaced* runtimes — e.g. Docker Desktop data still present after switching to OrbStack. These are genuinely orphaned and often 5-30GB.

1E: Package managers and toolchains

Detect installed package managers with command -v and use their native cleanup commands:

ToolDetectionCleanup Command
Homebrewbrewbrew cleanup --prune=all && brew autoremove
npmnpmnpm cache clean --force
Yarnyarnyarn cache clean
pnpmpnpmpnpm store prune
Bunbunbun pm cache rm
pippip3 or pippip cache purge
Poetrypoetrypoetry cache list, then poetry cache clear <name> --all for each
Condacondaconda clean --all -y
CargocargoNo built-in prune — measure ~/.cargo/registry/ and ~/.cargo/git/
Gogogo clean -modcache (clears $(go env GOMODCACHE))
gemgemgem cleanup
BundlerbundleMeasure ~/.bundle/cache/ (global cache). Do not run bundle clean — it's project-scoped and can remove gems other projects need.
CocoaPodspodpod cache clean --all
Composercomposercomposer clearcache
MavenmvnNo built-in prune — measure ~/.m2/repository/
GradlegradleNo built-in prune — measure ~/.gradle/caches/ (safe to delete)
Flutterflutterflutter pub cache clean
Dartdartdart pub cache clean
Swift PMswiftMeasure ~/Library/Caches/org.swift.swiftpm/

Also detect toolchain version managers — these accumulate old runtime versions that may no longer be needed:

ManagerDetectionData Location
misemise~/.local/share/mise/ — use mise prune, mise cache clear
asdfasdf~/.asdf/installs/ — list versions, flag old ones
nvm~/.nvm/ exists~/.nvm/versions/ — list versions, flag old ones
pyenvpyenv~/.pyenv/versions/ — list versions, flag old ones
rbenvrbenv~/.rbenv/versions/ — list versions, flag old ones
rustuprustup~/.rustup/toolchains/ — rustup toolchain list, flag old ones
sdkman~/.sdkman/ exists~/.sdkman/candidates/ — sdk flush, list old JDKs
goenvgoenv~/.goenv/versions/ — list versions, flag old ones

1F: Orphan detection

This must be a systematic, automated scan — not a spot check of entries you suspect. You have no prior knowledge of this machine.

Step 1: Build a reference list of what's installed.

  • ls /Applications/ (top-level and one level deep for folders)
  • mdfind "kMDItemContentType == 'com.apple.application-bundle'" -onlyin /Applications
  • Store the app names and bundle IDs for matching.

Step 2: List every non-Apple entry in these directories:

  • ~/Library/Application Support/*
  • ~/Library/Caches/*
  • ~/Library/Saved Application State/*
  • ~/Library/HTTPStorages/*
  • ~/Library/Preferences/*

Filter out com.apple.* entries. For entries that don't use reverse-domain naming but might be Apple/system-managed (e.g. AddressBook, CloudKit, GameKit), check with mdfind "kMDItemCFBundleIdentifier == 'com.apple.*'" -onlyin /System -name "{name}" or verify the name matches a running system process (pgrep -x "{name}"). For the remaining entries, automate matching in a single script or loop — do not hand-pick entries to check.

For ~/Library/Preferences/*, only consider entries over 1MB. For ~/Library/HTTPStorages/*, same threshold.

Step 3: For each non-Apple entry, attempt to match it to something installed:

  1. Direct name match against /Applications (case-insensitive)
  2. If it looks like a bundle ID (contains dots), use mdfind "kMDItemCFBundleIdentifier == '{id}'" to find the app anywhere on disk
  3. Extract the likely app name from the bundle ID (last segment, or known patterns like com.company.appname) and check /Applications and which
  4. which {name} to check if it's a CLI tool rather than a GUI app

Step 4: Flag unmatched entries as orphan candidates. Only entries that fail ALL matching attempts are orphans. Report the full list with sizes.

1G: Dev artifact scan

Search the detected development directory for large regenerable directories. Use find with -maxdepth 6 -type d to limit recursion and only match directories. Use -prune on matched names to avoid descending into them (e.g. find ~/Development -maxdepth 6 -type d -name node_modules -prune). Exclude the current working directory.

Target these patterns:

PatternEcosystemVerify by checking for...
node_modules/JavaScript/TypeScriptpackage.json in parent
target/RustCargo.toml in parent
.build/Swift (SPM)Package.swift in parent
build/Gradle, CMake, variousBuild config in parent
dist/JS bundlers, PythonBuild config in parent
.next/Next.jsnext.config.* in parent
.nuxt/Nuxt.jsnuxt.config.* in parent
.venv/, venv/Python virtualenvs(dir, not.env files)
__pycache__/Python bytecode.py files nearby
.tox/Python toxtox.ini in parent
.terraform/Terraform*.tf files in parent
Pods/CocoaPodsPodfile in parent
.dart_tool/Flutter/Dartpubspec.yaml in parent
.parcel-cache/Parcelpackage.json in parent
vendor/Go, PHP, RubyLock file in parent
.gradle/Gradle (in-project)build.gradle* in parent
.pytest_cache/, .mypy_cache/, .ruff_cache/Python toolingPython project in parent

For each candidate, verify it's regenerable by checking for the corresponding config file in the parent directory. Anything large that lacks a config file — inspect it before flagging.

1H: Time Machine local snapshots

tmutil listlocalsnapshots /

Local snapshots are redundant copies stored on the boot drive. Deleting them does not affect the external Time Machine backup. They can consume 10-80+GB. Present each snapshot with its date. Removal requires sudo tmutil deletelocalsnapshots <date>.

1I: iOS device backups

Check ~/Library/Application Support/MobileSync/Backup/ for local device backups. Each is a UUID-named folder, typically 5-20GB. These cannot be regenerated — flag for individual confirmation only. Note the user should verify they have iCloud backup or no longer own the device.

1J: Trash and large installers

  • ~/.Trash/ — measure contents and suggest emptying
  • /Volumes/*/.Trashes/ — check for trash on mounted external volumes
  • /Applications/Install macOS*.app — old macOS installers, 12-14GB each, safe to delete after use
  • GarageBand sound libraries: /Library/Application Support/GarageBand/, /Library/Application Support/Logic/, /Library/Audio/Apple Loops/ — can be 2-14.5GB. Always flag in batch-confirm with size, even if GarageBand is installed — many users have it installed but never use it. These can be re-downloaded from within GarageBand's sound preferences.

Phase 2: Classification

Classify every finding using these principles:

Auto-clean (act without asking)

A finding is auto-cleanable if:

  • A package manager is installed and has a native cleanup command — these are designed by their authors to be safe
  • The directory's name or structure clearly indicates it's a cache (temporary, regenerable, not user-created content)
  • It's log data under ~/Library/Logs/ or /Library/Logs/ older than 7 days
  • It's a ~/Library/Saved Application State/, ~/Library/HTTPStorages/, or ~/Library/Preferences/ entry for a confirmed orphan
  • It's Xcode DerivedData (fully regenerable)
  • It's Xcode DeviceSupport entries for iOS versions still supported by the installed Xcode
  • It's dead/unavailable simulator devices (per xcrun simctl list)
  • It's Apple System Log files (/private/var/log/asl/*.asl)
  • It's crash reports under DiagnosticReports/

Batch-confirm (group by consequence, separate yes/no per group)

Don't lump everything into one giant yes/no. Group items by what happens after deletion — the user should be able to approve "instant-regenerate" items without also committing to "slow rebuild" items.

Group A — instant regenerate (caches that rebuild silently on next app launch, zero user cost):

  • Electron app internal caches (Cache/, Code Cache/, GPUCache/, Service Worker/) for installed apps
  • Large cache directories for installed apps
  • Trash contents

Group B — rebuild on next install/build (regenerable but costs real time):

  • Dev artifacts: dependency dirs (node_modules, vendor, Pods, etc.) outside the current project
  • Build outputs (target/, .build/, .next/, dist/) outside the current project
  • Note the rebuild cost where relevant (e.g. Rust target/ can take minutes-hours)

Group C — version pruning (safe but removes the ability to switch back without re-downloading):

  • Old simulator runtimes the user no longer targets
  • Old toolchain versions from version managers — present which versions exist and which appear active

Research-then-recommend (unknown findings)

If a finding is large (>50MB) and doesn't match the above:

  1. Inspect its contents — list top-level files/dirs, check file types, check timestamps
  2. Search the web for what it is and whether it's safe to delete
  3. Based on research, either:

- Recategorize it as auto-clean or batch-confirm - Present it to the user with your findings and a recommendation

Individual-confirm (needs user decision)

  • Orphaned ~/Library/Application Support/* directories — even if confirmed orphaned, inspect contents first (could contain databases, license files, or libraries with value beyond the app)
  • ~/Library/Containers/* for uninstalled apps — SIP protects the container shell and .plist; only Data/* contents can be removed. If rm fails on the container shell, remove Data/* instead and note the empty husk is harmless.
  • Xcode Archives — not regenerable, may be needed for App Store submissions
  • iOS device backups — not regenerable
  • Time Machine local snapshots — confirm the user understands these are boot-drive copies, not the external backup
  • Docker system prune — deletes all stopped containers and unused images
  • Old macOS installers and preinstalled content (GarageBand libraries)
  • Anything you researched but still can't confidently classify

Key principle: never guess. If you know it's a cache, clean it. If you don't know what it is, look it up. If you still can't tell, ask the user with everything you found.

Phase 3: Execution

Step 1: Report

Present a summary table of everything found, grouped by classification tier, with sizes. Show total reclaimable space. Make it satisfying — the user should feel the weight of what's been accumulating.

Step 2: Auto-clean

Execute all auto-clean actions. Use the system's rm (respect aliases). Only fall back to /bin/rm if the error indicates SIP protection on a system-managed path — for any other failure, report it rather than bypassing the alias. Run package manager cleanup commands (these are safe because they were designed by their authors for exactly this purpose — but show which commands will run before executing). Report what was cleaned and how much space was freed.

Step 3: Batch confirm

Present the batch-confirm group as a table with sizes. Ask the user yes/no. Execute if approved.

Step 4: Individual confirm

For each individual-confirm item, present:

  • What it is (app name, purpose if known)
  • Size
  • What's inside (top-level listing)
  • What you found via research (if applicable)
  • Your recommendation

Act on user's decision for each.

Step 5: Final report

Show before/after — total freed, what was cleaned, what was skipped. Close with the top 5 remaining space consumers so the user knows what's left if they want to go further.

Important Rules

  • Never delete the current project's files — always exclude the working directory tree
  • **Never delete com.apple.* entries** — system-managed
  • Never delete app data for running apps without warning — check pgrep if unsure
  • Never touch swap files — /private/var/vm/swapfile* is actively used by the OS
  • Never touch /System/Library/ — SIP-protected and system-critical
  • Respect rm aliases — if the user has rm aliased to trash or similar, use it. Only fall back to /bin/rm when the error is specifically SIP-related (Operation not permitted on a system-managed path). The alias exists for safety — don't bypass it for generic failures.
  • Verify orphans thoroughly — an app might be a CLI tool (which), installed somewhere non-standard (mdfind), or a system service. Only call it orphaned if all checks fail.
  • Treat non-regenerable data differently — iOS backups, Xcode Archives, and user documents are NOT caches. Never auto-clean them.
  • Don't flag operational VM images as reclaimable — VM disk images (OrbStack, Docker Desktop, Claude Desktop vm_bundles) are opaque blobs that the user can't meaningfully act on by deleting. Either inspect what's pruneable *inside* them (if the runtime is running), or note them as large but not directly actionable. Deleting them just forces a re-download or factory reset — that's not cleanup.
  • Research unknowns — use web search to determine what unfamiliar large directories are before recommending action
  • Be transparent — always show what you're about to do and how much space it affects
  • Only report actionable findings — if the user can't do anything useful with a finding (e.g. a stopped runtime's disk image, an app's core data bundle), don't list it as reclaimable. Mention it as context ("this is why X is large") but don't count it toward the reclaimable total.

平台分布

Codex

38.96%
按下载量换算52

Claude

28.37%
按下载量换算38

Cursor

17.35%
按下载量换算23

Gemini CLI

10.34%
按下载量换算14

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

权限和风险

操作浏览器

该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

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。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills