菜单
A human-facing index of every skill and agent in this repo, grouped by category, with a one-line trigger for each. Use only when the user explicitly asks "what skills do I have", "what can you do", "is there a skill for X", or invokes `/catalog` directly — never fires on its own.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/yulonglin/dotfiles --skill catalog命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Spawn a detached tmux session running Claude Code in a chosen directory, seeded with an initial prompt, optionally reachable by Remote Control.
House plot styling — pastel matplotlib defaults + annotation helpers. Use when making publication-quality figures, when the user asks to "style my plots", "use house style", "use pastelplot/anthroplot", or when generating any matplotlib figure for reports, papers, or decks.
Drives a stalled session to a terminating end state - land the work, state the blocker, or take one step (dotfiles trial, manual-only)
正在显示 SKILL.md
| name | catalog |
| description | A human-facing index of every skill and agent in this repo, grouped by category, with a one-line trigger for each. Use only when the user explicitly asks "what skills do I have", "what can you do", "is there a skill for X", or invokes `/catalog` directly — never fires on its own. |
| disable-model-invocation | true |
One place to look up what already exists instead of re-discovering it mid-task. This skill only fires when asked for by name — it does not compete for context on other requests.
Most skills below are model-invoked: Claude fires them automatically when your request matches the trigger phrase, so you don't need to remember exact names. A few are user-invoked only (marked below) — Claude never reaches for these on its own, so the name is the only way in.
| Skill | Use when |
|---|---|
commit | "commit this", "commit these changes", "save my work" |
commit-push-sync | "commit and push", "sync changes", "update remote" |
merge-worktree | "merge this worktree", "merge my branch back", "finish this worktree" |
finishing-a-development-branch | Implementation is done, tests pass, deciding merge/PR/cleanup |
wrap-up | A session needs to reach a terminal state — land it as a draft PR, state the blocker, or say it's done (dotfiles trial) |
diagnosing-bugs | "diagnose"/"debug this", something broken/throwing/failing/slow |
mv-repo | Moving a repo to a new directory (venv, project state, tmux sessions) |
migrate-to-codex | Migrating instructions/skills/agents/MCP config into Codex |
| Skill | Use when |
|---|---|
agent-browser | CLI-driven browser automation — navigate, fill forms, screenshot, scrape |
browser-session | Before any agent-browser task — sets up the right Chrome profile |
playwright | Automating a real browser from the terminal for dev/testing |
chrome-devtools | DevTools-level access — evaluate JS, profiling, network inspection |
claude-in-chrome | Quick tasks in your actual live Chrome tabs |
| Skill | Use when |
|---|---|
anthropic-style | Any visual output that should match Anthropic's look (plots, TikZ, slides, web) |
check-bib-references | Verifying BibTeX citations aren't LLM-fabricated |
check-prose-claims | Fact-checking stats/comparatives/quotes in slides, reports, papers |
marp-deck | Building slide decks from Markdown with Marp |
pdf | Reading, creating, or reviewing PDFs where layout matters |
| Skill | Use when |
|---|---|
modal | Cloud GPU/serverless compute — training, batch jobs, serving |
jobs | Submitting experiments/agent jobs with resource caps, queue status |
server-storage-tiering | Root disk near full on a server/cloud box with an attached volume |
sweep-ai-safety | Sweeping recent AI safety research from curated sources |
spawn-session (user-invoked only) | Starting a detached Claude session in another directory with a seed prompt, optionally reachable by Remote Control |
| Skill | Use when |
|---|---|
bear | Reading/editing Bear notes (macOS) |
things3 | Reading/managing Things 3 tasks, projects, areas, tags |
use-spark | Searching/reading Spark email, calendar, contacts |
setup-channel | Setting up Telegram/iMessage/Things Cloud for a project |
reply | Stuck on replying to someone — scaffolds while you write |
context-fetcher (skill) | Gathering Gmail/Slack/Granola/Calendar context before drafting a reply |
log-gap | Recording a knowledge gap you just noticed ("wait, I didn't know that") |
recall-feedback | Spaced-repetition review of past feedback memories |
decide | Stuck overthinking/oscillating on a decision |
| Skill | Use when |
|---|---|
grilling | Stress-testing a plan or decision — "grill me", "poke holes in this" |
grill-me (user-invoked only) | Same as grilling, invoked directly by name |
chronicle | Resolving ambiguity via screen/recent-history context ("what was I just doing") |
writing-great-skills (user-invoked only) | Reference for writing/editing skills well — read before authoring a new one |
| Agent | Use when |
|---|---|
context-fetcher | Same as the skill above — gathers cross-source context for a named person |
llm-billing | "how much have I spent on OpenRouter/OpenAI/Anthropic", checking API credits |
When adding a new skill or agent, add one row under the closest-fitting category (or a new one). When removing a skill, remove its row. Keep each entry to a single trigger-focused line — the full behavior lives in the skill's own SKILL.md.
