中文 README · English

Turn real workflows into discoverable, verifiable, reusable Agent Skills.

Design systems · UI/UX · brand assets · Figma tools · Skill engineering · content workflows

Install in one command

npx skills@latest add AaronXu-Lab/AaronSkill

The installer discovers every Skill in the repository that contains a SKILL.md and lets you choose which ones to install. To list the available Skills without installing them:

npx skills@latest add AaronXu-Lab/AaronSkill --list

Install one Skill:

npx skills@latest add AaronXu-Lab/AaronSkill \
  --skill aw-find-and-port-ui-component

Install globally for Codex:

npx skills@latest add AaronXu-Lab/AaronSkill \
  --skill aw-find-and-port-ui-component \
  --agent codex \
  --global

The skills CLI installs to the current project by default. Add --global to use a Skill across projects. Interactive installation recommends symlinks; use --copy to create a standalone copy.

Skills

Install local repository symlinks

Run this from the repository to check all first-level Skill directories and create any missing symlinks in ~/.agents/skills. This includes the Working On and no-longer-maintained Skills.

On macOS, you can also double-click link-skills.command in the repository root, then press Return to close the results window when it finishes.

python3 link-skills.py

The script locates the repository relative to itself, so it can run from any working directory. Re-running it skips correct symlinks. It preserves and reports conflicting same-name files, directories, or symlinks that point elsewhere, and exits with code 1.

Add --prune to remove stale same-name symlinks whose resolved destinations are first-level directories in this repository without a SKILL.md. Other repositories' symlinks and regular files or directories are preserved.

Preview changes, clean stale links, or choose another global directory:

python3 link-skills.py --dry-run
python3 link-skills.py --prune --dry-run
python3 link-skills.py --prune
python3 link-skills.py --target ~/.codex/skills

The repository currently contains 17 Skills, grouped by primary purpose into meta Skills, tools, resource discovery, and design support. It also has a Working On group and a separate list of no-longer-maintained Skills.

Meta Skill

Capabilities for creating, upgrading, and standardizing other Skills.

Skill	Version	What it does	Key boundaries
aw-meta-skill	1.9.1	Creates or updates any Skill, serving as the unified entry point in place of skill-creator	Reads and validates according to the change scope; Markdown is the source of truth, and SVG details are loaded as needed

Tools

Reusable workflows for well-defined inputs and outputs, including asset processing, content handling, and focused engineering tasks.

Skill	Version	What it does	Key boundaries
aw-clash-network-doctor	1.12.0	Separates Claude cleanup, iterative macOS score optimization, and existing Android diagnostics	Selects the target Clash config; adds missing authorized app rules to the persistent overlay and updater, then verifies rule priority, loaded rules, and new app connections; tracks IPKK, Net.Coffee, and CLIS scores separately
aw-logo-asset-cook	1.3.1	Creates and validates cross-platform icon assets from SVGs or images that have been assessed and iteratively redrawn	Requires user-specified input; confirm low-fidelity conversions, and check compatibility and themes after a redraw is approved
aw-mail-read-later	1.1.1	Recommends, reads, summarizes, or translates one item from Outlook's Read Later folder	Handles one item per manual run; requires confirmation before archiving or removing an email
aw-tiered-task-dispatch	1.9.0	Handles read-only research in the main session and dispatches implementation work in Codex or Claude based on remaining uncertainty, impact, and coordination complexity	Uses the lower tier for settled, local work and the middle tier for substantive design or coordination; clarifies key plans first and treats history only as a weighting signal; resolves the latest available model by series, with user choice taking priority; execution sessions validate independently and do not communicate with each other; each tier uses an independent branch/worktree and preview port, then rebases and integrates the complete request once locally under a repository lock; the main session ends after dispatch, while execution sessions validate, integrate and report
rewrite-like-aaron	1.1.1	Rewrites AI-generated Chinese drafts in Aaron's current blog voice	Preserves facts and positions; limits catchphrases, rhetorical questions, and surface-level imitation through Chinese-English mixing

Resource discovery

Finds, verifies, filters, and organizes useful resources from external sources.

Skill	Version	What it does	Key boundaries
aw-comic-dossier-packer	1.1.0	Collects comic covers, source information, Xiaohongshu covers, and a final dossier	Confirm costs before upscaling; use original visuals rather than reproducing covers in social media graphics
aw-logo-finder	1.1.0	Finds, compares, and exports brand or product logos from official sites, logo directories, and app stores	Confirm candidates and output dimensions before generating lossless WebP files

Design support

Infrastructure, standards, and engineering support for design systems, component implementation, and design delivery.

Skill	Version	What it does	Key boundaries
aw-design-md-author	1.7.0	Creates, reviews, and maintains complete DESIGN.md visual contracts following Google Labs conventions	Protects ownership and comment-only boundaries; reports when official validation is unavailable; does not replace code or Figma
aw-design-system-gallery	4.13.1	Creates, reviews, or improves Gallery default examples, real design axes including perceptible animation behavior, and state comparisons	Reviews source first and uses browser checks only when needed; routes other pages and Components separately, with nine component checks and a numbered findings matrix; default instances, stable Anatomy parts, and caller-owned effects are checked semantically; Text and descriptive Description placeholders are allowed; Anatomy requires explanatory visual parts owned by the component; reuses the local standard structure and documentation layout for component panels; Internal pages may contain only one Playground when the project specifies it; Playground and Properties share axis order, values, and dependency-based disabling; Reset includes theme overrides; section links use path routing and in-page anchors; robustness-only checks do not enter the production Gallery by default; composite examples do not replace child matrices; captions show only real public axes; boundary hints use the existing toggle and both states are verified; project configuration stays in the target repository
aw-design-fake	1.8.1	Unifies fake data, demo source code, and placeholder interactions in prototype projects, and initializes or syncs bundles	Prioritizes real contract data and states; demo scenarios must be explicitly escapable and isolate writes; source code is reused verbatim and displayed without execution; does not touch unit-test mocks
aw-design-token-consistency-auditor	0.9.0	Compares Figma Variables, DESIGN.md, and CSS/Less tokens	Produces audit evidence only and does not rewrite tokens automatically
aw-find-and-port-ui-component	2.2.1	Finds and compares specific components from Base UI and other React sources, including Lobe UI, then adapts them for Base UI and CSS Modules	Handles component-level requests only; Find verifies the exact source and license, then waits for a selection; Port confirms details lost or dependencies added in advance

Working On

Skills under active development, each kept in its own directory for optional installation and use.

Skill	Version	What it does	Key boundaries
temp-local-service-doctor	1.1.1	Starts local multi-service projects and diagnoses port, API, and page-loading failures	Prefers existing entry points; confirms process ownership; verifies through call paths and the target page
temp-prd-verifier	1.1.1	Verifies requirements against a project PRD and, when needed, compares them with the interface or implementation	Separates explicit requirements, inferences, and gaps; performs read-only checks or fixes within existing authorization
temp-small-improves	1.2.0	Sequentially checks 66 small UI improvements across interface details, motion, typography, colors, accessibility, layout, and wording	Runs only when explicitly named; merges overlapping checks and makes the smallest evidence-based changes

No longer maintained

These Skills remain in the repository for existing users, but are no longer actively developed or considered for new capabilities.

Skill	Version	Status
aw-figma-component-governance	0.10.0	No longer maintained

How these Skills work

Real request
   │
   ├─ Read project, source, and environment constraints
   │
   ├─ Follow a narrow, traceable workflow
   │
   ├─ Pause for confirmation before high-risk or costly actions
   │
   └─ Verify with linting, structured reports, or read-back

Skills in this repository put fragile, repetitive steps in scripts/, keep rules and schemas in references/, and use SKILL.md as a clear entry point for execution.

Repository structure

<skill-name>/
├── SKILL.md              # Trigger guidance and complete workflow
├── scripts/              # Repeatable, deterministic tools (optional)
├── references/           # Schemas, standards, and runbooks (optional)
└── fixtures / graders    # Evaluation assets (optional)

The skills CLI recursively discovers SKILL.md files, so every first-level directory can be installed as an independent Skill.

Check dependencies before use

The eight design agents are maintained in Design-Agent and installed separately. Its workflow entry is aw-design-orchestrator; aw-component-advisor provides guidance/audit using the manually started AaronUI-Web knowledge MCP. Only component/dialog is covered in this iteration; Gallery displays more components. The old aw-component-checker is a deprecated redirect with no component-rule fallback. Design support and aw-meta-skill remain in this repository.

Dependencies vary by Skill. Read the relevant SKILL.md before invoking one:

• Workflows that use Figma or other external tools require the relevant app, permissions, or authorization.
• Image upscaling requires a Gemini API key and may incur API charges.
• Web research, GitHub source verification, and remote publishing require network access.
• If an optional Skill is unavailable, use a fallback instead of claiming capabilities you do not have.

Development and validation

After modifying a Skill, at minimum check its frontmatter and directory structure:

python /path/to/skill-creator/scripts/quick_validate.py ./<skill-name>

If the Skill includes scripts, tests, or graders, run the relevant checks too. Successfully loading SKILL.md does not mean its workflow has been validated.

Updates

After installing with the skills CLI, update all Skills or a specific Skill:

npx skills@latest update
npx skills@latest update aw-find-and-port-ui-component

See the skills CLI for more installation options.