muki-ai-plugins
繁體中文版
A Claude Code plugin marketplace — visual regression testing, test review, project knowledge capture, AI session review, and Jev-powered web search.
Plugins
| Plugin | Description |
|---|
| figma-visual-reviewer | Visual regression testing — compare Figma designs against live web pages |
| review-tests | Test review doctor — diagnose a test file for blind spots, output a self-contained HTML report |
| review-code | Code review — standards, spec, and bugs in parallel sub-agents, important findings confirmed by independent votes |
| lore | Project lore — scaffold, consult, capture, guard, maintain, health-check, eval, and language-align the implicit knowledge your codebase can't show |
| session-review | AI session review — turn every Claude Code and Codex session of a project into an interactive report on how you work with AI, and which routines to make into skills |
| jev-search | Web search through Jev Search: answers the built-in WebSearch with Jev-ranked results and falls back to the built-in tool when Jev cannot answer |
| hyday-pet | Virtual pet above your prompt: the work Claude does decides its species, tool errors make it sick, plus mini games and a shop |
| world-news | News under your prompt: reads the Taiwan edition of Google News for the categories you pick in the background and pins the top headline |
Install
# Add the marketplace
/plugin marketplace add mukiwu/muki-ai-plugins
# Install individual plugins
/plugin install figma-visual-reviewer
/plugin install review-tests
/plugin install review-code
/plugin install lore
/plugin install session-review
/plugin install jev-search
/plugin install hyday-pet
/plugin install world-news
Codex, Cursor, Gemini CLI, and other agents
npx skills add mukiwu/muki-ai-plugins
The installer asks which skills to install, which agents to install them to, and whether to install for the current project or for every project. In Codex, invoke a skill with $name (for example $session-review) instead of /name.
This installs the skill folders only. Plugin extras such as lore's session-start reminder and pre-commit check are not included, so Claude Code users should prefer /plugin install. jev-search lives in its own repository and is not part of this list; hyday-pet and world-news are Claude Code mods with no skill to install.
Plugin Overview
figma-visual-reviewer
Pixel-level visual comparison between Figma designs and live web implementations.
/review — Interactive visual review- Figma API export → Playwright screenshot → pixel diff → AI judgment
- Generates HTML diff reports with side-by-side comparison
- Supports RWD multi-viewport checks
Read more →
review-tests
Reads an existing test file, finds blind spots, and produces a self-contained HTML report — read-only, never touches your code.
- Checks assertion validity, behavior gaps, mock health, and test structure
- Backlinks every finding to the key lines of the source under test
- Outputs a single inline-everything HTML report to
.review-tests/
- Diagnose only — fixes are left to TDD
Read more →
review-code
Reviews the diff since a fixed point along three axes, each in its own sub-agent, then confirms important findings with independent votes — read-only, never touches your code.
- Standards, spec, and bugs reviewed in parallel, reported separately
- Findings labelled Important, Nit, or Pre-existing
- Two fresh sub-agents vote on each important finding; a split gets a third
- Re-reviews only look at the new diff and stop after three rounds
Read more →
lore
Scaffold, consult, capture, guard, maintain, health-check, and language-align project lore — the implicit knowledge your codebase carries but can't show on its own.
lore-init / lore-consult / lore-capture / lore-guard / lore-check / lore-eval / lore-maintain / lore-ul — eight skills covering the full lifecycle- Stores business rules, pitfalls, API maps, a shared glossary, and the why behind decisions under
docs/lore/
- Consult before planning or bug-fixing; capture what you learn as you go; guard your diff against recorded lore before commit
- Mark over delete — outdated-but-once-true knowledge keeps its lesson
Read more →
session-review
Reads every AI session you had in one or more projects (Claude Code and Codex) and turns it into one interactive HTML report.
/session-review [path ...] — defaults to the current project; several paths are merged into one report- Skill ranking split by who invoked it (you vs. the model), your real workflow drawn as diagrams, and verbatim quotes on how you build and debug with AI
- Suggests custom skills for routines that keep repeating, each with a
SKILL.md draft and a user/project scope call
- Flags installed skills you never used and never needed in these sessions, with the command to remove each
- Skips automated SDK sessions and de-duplicates rewound messages; falls back to built-in diagram rules without
diagram-design
Read more →
jev-search
Answers Claude Code's built-in WebSearch with Jev Search: Jev reads the request in plain language, picks the sources and time window, and ranks every result by relevance. When Jev cannot answer, the built-in search runs as before. The code lives in its own repository because it is also an npm package.
- Function hook on
WebSearch, nothing changes in how you or the model search
jev_search MCP tool for explicit sources / window filters, plus a skill on when to reach for it- Also usable outside Claude Code as an MCP server or CLI via
npx jev-search-mcp
- Function hooks are early access: set
CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 in your settings env
Read more →
hyday-pet
A virtual pet above your Claude Code prompt that feeds on the work you and Claude do together
- Every tool call and finished turn gives it XP, from egg to adult
- Whether Claude mostly edits files, runs commands or looks things up decides if it grows into a mole, otter, owl or chimera, and it branches again as an adult
- Tool errors give it germs; green tests make it well
- The side pane has an animated ASCII pet, rock paper scissors and guess the number, a coin shop for premium food and color styles, and a detail page
- Function hooks are early access: set
CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 in the env of your settings
Read more →
world-news
Taiwan's top news pinned under your Claude Code prompt
- Each message you send checks whether the cooldown has passed (30 minutes by default), then reads the Taiwan edition of Google News for your categories, in the background
- Eight sections plus seven international regions, picked like
/world-news 14ace
- Only stories from the last day that you have not seen, in Google's order
- Nothing is added to the model's prompt, so no tokens and no cache misses
- Function hooks are early access: set
CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 in the env of your settings
Read more →
License
MIT
Sponsor
If this project helps you, consider sponsoring to support ongoing development
