Awesome Skills

Independent Agent Skills packaged according to the Anthropic Agent Skills specification. These packages provide modular workflows for software engineering, code quality review, network tunneling, and UI components.

Compatible with OpenAI Codex, Claude Code, ZCode, Cursor, Gemini CLI, and Windsurf.

English | 中文说明

Acknowledgments: LINUX DO

This repository acknowledges the LINUX DO community. Engineering solutions throughout these packages—including log-first reproduction loops, code review criteria, and cross-NAT tunnel configurations—draw directly from technical sharing within the LINUX DO community.

Architecture and Design Principles

Reproduction before modification: Packages such as evidence-driven-bugfix and production-code-quality-review require capturing reproducible failure evidence and root-cause verification before applying code changes.
Progressive disclosure: Each package exposes only metadata during initial discovery, costing roughly 100 tokens per skill. Operational scripts and reference documentation load only when invoked.
Process isolation: Packages such as codex-agent-worktree-setup bind agent executions to isolated Git worktrees, keeping branch state decoupled from the primary working directory.
Modular distribution: Each skill is self-contained with dedicated documentation, reference materials, and optional automation scripts.

Skill Categories

Software Engineering and Quality Assurance

evidence-driven-bugfix: Captures reproducible failure evidence, isolates root causes, and re-verifies post-fix state.
production-code-quality-review: Evaluates pull requests, merge readiness, and architecture-sensitive diffs.
codex-agent-worktree-setup: Creates branch-bound isolated Git worktrees for agent tasks.
best-project-memory: Maintains cross-session project context, decision logs, and handoff state.

Networking and Infrastructure Operations

cloudphone-adb-tunnel: Sets up FRP STCP relays and Termux keepalive for remote ADB and scrcpy connections to non-root Android cloud phones.
windows-ssh-stcp: Configures SSH access to NAT-isolated Windows machines over existing FRP STCP relays with loopback sshd.
muse-reverse-ssh: Configures reverse SSH tunnels and VPS port forwarding with process keepalive.

UI Engineering and Assets

awesome-ui-kit: Web components for AI chat interfaces, RAG search results, split-pane canvases, and agent execution monitors.
split-image-assets: Splits source images into reusable asset packages with alpha masks, layers, metadata, and quality checks.

Search, Knowledge Base, and Automation

grok-search: Executes web searches, URL content extraction, and site discovery via search APIs.
obsidian-doc-router: Provides routing queries and anti-orphan documentation workflows for Obsidian vaults.
yunzhi-cloudphone-checkin: Automates daily check-ins and cloud phone renewals via browser console, Chrome CDP, or HTTP CLI.

Choose a Skill

Skill	When to use	Best for	Docs
`awesome-ui-kit`	You are assembling AI chat, RAG, canvas, or agent interfaces	copy-paste AI web components, multi-framework support	Guide
`best-project-memory`	You need durable repo memory across long-running work	restoring context, recording decisions, keeping TODOs and handoffs current	Guide
`cloudphone-adb-tunnel`	You need ADB/scrcpy to a non-root cloud phone over the internet	FRP STCP relay, Termux frpc deployment, phantom-process hardening	Guide
`codex-agent-worktree-setup`	You need an isolated Codex development thread bound to a branch	protected main worktrees, branch-bound agents, detached HEAD repair	Guide
`evidence-driven-bugfix`	You need a real bugfix loop instead of a guess-fix	logs-first debugging, failing evidence, root cause, re-verification	Guide
`grok-search`	You need live web access instead of offline knowledge	web search, current facts, URL reading, site page discovery	Guide
`muse-reverse-ssh`	You need a machine without a public IP reachable over SSH from the internet	reverse SSH tunnels, VPS port forwarding, tunnel keepalive	Guide
`obsidian-doc-router`	You are reading/writing Obsidian vault ops notes or topologies	router-first queries, anti-orphan documentation workflow	Guide
`production-code-quality-review`	You want a production-minded review of changes	PR review, merge readiness, architecture-sensitive diffs	Review Workflows
`split-image-assets`	You need to turn one image into reusable package assets	masks, transparent layers, previews, metadata, QA	Guide
`windows-ssh-stcp`	You need SSH to a Windows host with no inbound public port	reuse an existing frps over STCP, loopback-only sshd, NSSM keepalive	Guide
`yunzhi-cloudphone-checkin`	You need automated check-in and cloud phone renewal for Yunzhi Cloudphone	browser console one-click script, Chrome CDP automation, direct CLI	Guide

If you are not sure which one to use, go straight to the Skill Matrix.

Recommended Starting Points

I do not know which skill to use: Skill Matrix
I want the fastest install path: Quickstart
I prefer Chinese docs: 中文说明
I want a repo walkthrough first: Golden Path
I want to review code changes: Review Workflows
I want real-world usage examples: Examples

Install

Agent discovery paths depend on the target runtime:

Client	User Scope	Workspace Scope
OpenAI Codex / ZCode	`~/.agents/skills/`	`.agents/skills/`
Claude Code	`~/.claude/skills/`	`.claude/skills/`
Cursor / Windsurf / Gemini CLI	`~/.agents/skills/`	`.agents/skills/`

Installation Command

Copy the selected package directory into the agent discovery path:

mkdir -p ~/.agents/skills
cp -R <skill-folder> ~/.agents/skills/

Restart the agent or reload skills to complete package discovery.

For assistance choosing packages, refer to the Skill Matrix.

Package Directory Structure

Each skill conforms to the standard layout:

<skill-name>/
├── SKILL.md          # Frontmatter metadata and usage instructions
├── scripts/          # Deterministic scripts and CLI entrypoints
├── references/       # API specifications and technical reference docs
└── assets/           # Configuration files, templates, and static resources

Docs

Quickstart — Fast installation and verification guide
Skill Matrix — Feature comparison and routing matrix
FAQ — Frequently asked questions
Troubleshooting — Issue diagnostics and resolution
Review Workflows — Production code review procedures
Examples — Concrete usage examples
Chinese Overview — Localized Chinese documentation portal
Release Notes — Version history and release notes

Repository Layout

awesome-ui-kit/                     AI chat, RAG, reasoning viewer, and canvas components
best-project-memory/                Cross-session project memory, decisions, and handoffs
cloudphone-adb-tunnel/              FRP STCP relay for non-root Android cloud phone ADB
codex-agent-worktree-setup/         Branch-bound isolated Git worktrees for agent threads
evidence-driven-bugfix/             Evidence-first reproduction and bugfix loop
grok-search/                        Web search, fact verification, and URL inspection
muse-reverse-ssh/                   Reverse SSH tunnels and VPS port forwarding keepalive
obsidian-doc-router/                Obsidian vault documentation router and updates
production-code-quality-review/     Production-focused pull request review and diff audit
split-image-assets/                 Image asset decomposition, alpha masks, and packaging
windows-ssh-stcp/                   Windows SSH over FRP STCP with NSSM keepalive
yunzhi-cloudphone-checkin/          Yunzhi cloud phone daily check-in and automated renewal
docs/usage/                         Guides, matrix, troubleshooting, and examples
docs/zh/                            Chinese landing page and localized guides
docs/releases/                      Release notes and changelogs
docs/superpowers/                   Design specifications and implementation plans
tests/                              Repository-level regression test suite

Contributing

Verify that new packages adhere to the Agent Skills specification with a valid SKILL.md frontmatter.
Add regression tests under the package tests/ directory.
Register new packages in docs/usage/skill-matrix.md and repository tables.
Run the repository documentation test suite:
```
python3 -m unittest discover tests -v
```

For Maintainers

Run repository-level docs checks with python3 -m unittest discover tests -v
Run package tests from the relevant skill directory before release work
Maintain Grok Search from grok-search/ in this repository; local installed copies are downstream deployments
Keep release history in docs/releases/README.md
Keep design and implementation history in docs/superpowers/ and docs/dev/

awesome-skills