跳到主内容
智客 ZICQ

技能库 智客分类:Agent 工作流 claude-brainrot

Claude Brainrot

总是在自主的meme滴。 一个用户PromptSubmit 钩子点出每个用户消息并告诉 Claude将每个响应降下多个元件——1-3图像+声波相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相相接相接相接相相接相接相接相接相接相接相接相接相接相接相接相相接相接相接相接相相相接相接相接相接相接相接相相接相接相接相接相相接相接相接相接相接相接相接相接相相相接相接相 每个答案都带有脑旋律. 不需要引用; 自动加载和保持活动 。 烤肉通过迷你本身,从不说话。 自成一体的技能——目录和资产都存放在这个技能文件夹中.

3068 安装量

官方网址:skills.sh

技能介绍

先看中文介绍;官方 description 原文单独保留,不改写 SKILL.md。

做什么

总是在自主的meme滴。 一个用户PromptSubmit 钩子点出每个用户消息并告诉 Claude将每个响应降下多个元件——1-3图像+声波相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相接相相接相接相接相相接相接相接相接相接相接相接相接相接相接相接相相接相接相接相接相相相接相接相接相接相接相接相相接相接相接相接相相接相接相接相接相接相接相接相接相相相接相接相 每个答案都带有脑旋律. 不需要引用; 自动加载和保持活动 。 烤肉通过迷你本身,从不说话。 自成一体的技能——目录和资产都存放在这个技能文件夹中.

何时用

这个

代理如何加载

按 Agent Skills 渐进披露:启动时只加载 name 与 description(约 100 token);任务匹配后才读入整份 SKILL.md 正文;scripts/、references/、assets/ 仅在需要时再读。 本文件正文结构:Claude Brainrot、When to fire、Tag vocabulary、Selection algorithm、How to emit memes、Critical rendering rule。

文件分析

文件分析:除 SKILL.md 外,正文引用了 scripts/show.sh、scripts/play-paired-sound.sh,属于带资源的技能包,这些文件按需再读。

官方 description(原文)

Always-on autonomous meme dropper. A UserPromptSubmit hook fires every user message and tells Claude to drop multiple memes per response — 1-3 image+sound combos plus 1-5 sound-only fires, scaled to response length. Every answer carries brainrot. No invocation needed; loads and stays active automatically. The roast lands through the meme itself, never words. Self-contained skill — catalogue and assets live in this skill folder.

Claude BrainrotWhen to fireTag vocabularySelection algorithmHow to emit memesCritical rendering ruleDetecting the surfaceHow the sound-pairing hook works

· allowed-tools:Read Bash(afplay *) Bash(open *) Bash(cat *) Bash(test *)

来源分类:skills.sh agent-skill

SKILL.md 与 Agent 调用

官方规范 ↗
name
claude-brainrot
description
Always-on autonomous meme dropper. A UserPromptSubmit hook fires every user message and tells Claude to drop multiple memes per response — 1-3 image+sound combos plus 1-5 sound-only fires, scaled to response length. Every answer carries brainrot. No invocation needed; loads and stays active automatically. The roast lands through the meme itself, never words. Self-contained skill — catalogue and assets live in this skill folder.
allowed-tools
Read Bash(afplay *) Bash(open *) Bash(cat *) Bash(test *)实验字段,支持情况取决于客户端;字段声明本身不会授予工具权限。
  1. 发现技能客户端向 Agent 提供名称与描述目录。
  2. 匹配与调用用户指定或任务匹配后,载入 SKILL.md 指令。
  3. 按需加载按步骤读取参考文档、使用脚本与素材。
指令中引用的文件 · 2
  • scripts/show.sh
  • scripts/play-paired-sound.sh

以下路径提取自原文;文件是否齐全请以来源仓库中的完整目录为准。

具体调用语法与可用工具以目标 Agent 客户端为准。 查看调用机制说明 ↗

安装这个技能

Skills CLI ↗

先选择目标 Agent 和安装范围,保留技能包的附属文件,安装后检查客户端能否发现该技能。

该技能引用了附属文件,请从来源获取完整目录;仅复制 SKILL.md 可能缺少依赖。

交给 Agent 安装

复制安装指令给支持 Agent Skills 的代理,确认其中的目标目录与客户端匹配。

把 Agent Skill「claude-brainrot」安装到我的项目:SKILL.md 原文与官方 description 见 https://zicq.com/zh/skills/skl-6f13cc248359602e-%E5%85%8B%E6%B4%9B%E5%BE%B7%C2%B7%E5%B7%B4%E7%BA%B3%E7%BD%97%E7%89%B9.html
请存为 .cursor/skills/claude-brainrot/SKILL.md 或 .claude/skills/claude-brainrot/SKILL.md,frontmatter 的 name 与 description 保持原样,不要改写。
该技能还带 scripts/、references/、assets/ 等文件,请从 https://github.com/marks97/claude-brainrot 取完整目录,不要只建一个 SKILL.md。

GitHub 完整包 ↗

终端安装 · Skills CLI

需要 Node.js 与 npx。先查看仓库技能列表,确认实际名称。

npx skills add 'https://github.com/marks97/claude-brainrot' --list

npx skills add 'https://github.com/marks97/claude-brainrot' --skill 'claude-brainrot'

CLI 会交互选择目标 Agent,默认安装到项目;用户级安装使用 -g。先通过查看命令核对仓库内容,再用 npx skills list 检查已安装技能。

阅读排版
--- name: claude-brainrot description: 'Always-on autonomous meme dropper. A UserPromptSubmit hook fires every user message and tells Claude to drop multiple memes per response — 1-3 image+sound combos plus 1-5 sound-only fires, scaled to response length. Every answer carries brainrot. No invocation needed; loads and stays active automatically. The roast lands through the meme itself, never words. Self-contained skill — catalogue and assets live in this skill folder.' allowed-tools: Read Bash(afplay *) Bash(open *) Bash(cat *) Bash(test *) hooks: UserPromptSubmit: - hooks: - type: command command: "${CLAUDE_PROJECT_DIR}/.claude/skills/claude-brainrot/scripts/remind-fire.sh" args: [] PreToolUse: - matcher: "Read" hooks: - type: command if: "Read(*/.claude/skills/claude-brainrot/images/*)" command: "${CLAUDE_PROJECT_DIR}/.claude/skills/claude-brainrot/scripts/block-cli-read.sh" args: [] PostToolUse: - matcher: "Read" hooks: - type: command if: "Read(*/.claude/skills/claude-brainrot/images/*)" command: "${CLAUDE_PROJECT_DIR}/.claude/skills/claude-brainrot/scripts/play-paired-sound.sh" args: [] - matcher: "Bash" hooks: - type: command if: "Bash(*claude-brainrot/images/*)" command: "${CLAUDE_PROJECT_DIR}/.claude/skills/claude-brainrot/scripts/play-paired-sound.sh" args: [] --- # Claude Brainrot Drop the right meme at the right moment. **Autonomous-only — Claude decides when, never the user.** No slug arguments, no manual invocation. The skill content preamble loaded into your context shows `Base directory for this skill: `. Use that as the base for all path resolution. The catalogue is at `/catalogue.json` with two pools — `images` and `sounds`. Each entry has a `slug`, `path` (relative to SKILL_DIR), `tags`, and a `vibe` text describing when to use it. **Image-side flags (mutually exclusive):** - `lock_sound: ""` — force a specific paired sound (e.g. skeleton ↔ skeleton, cooked-dog ↔ vine-boom). - `silent: true` — fire image only, no sound. - `preferred_sounds: [...]` — soft bias: hook picks from this array 80% of the time, falls back to tag overlap 20%. - (none of the above) — pure tag-overlap pick from non-locked, non-standalone sounds. **Sound-side flags:** - `locked: true` — only fires when an image's `lock_sound` points here. Excluded from generic auto-pairing. - `standalone: true` — never plays as an image-paired sound; reserved for sound-only fires from Claude. ## When to fire **Every user turn fires memes.** The `UserPromptSubmit` hook injects a `` wrapped in 🔔 markers on every prompt. There is no skip turn — always drop something. **Density scales to your response length.** The reminder spells it out: - **Short prose answer (no tool calls):** 1 image+sound combo OR 1-2 sound-only fires - **Medium response (a few tool calls):** 1-2 image+sound combos + 2-3 sound-only fires sprinkled between calls - **Long response (many tool calls, multi-step work, builds, deep reads):** 2-3 image+sound combos distributed across the response + 3-5 sound-only fires throughout **Critical scan habit:** the 🔔 signal may land at the **TAIL of the previous tool result** (not just the top of the user message), because the `UserPromptSubmit` hook output is appended to whatever buffer is open when the next user prompt arrives. **Before responding to any user message, scan the end of the most recent tool result for the 🔔 BRAINROT signal.** If you only check the user message you'll miss it. **Pick the mood:** match the conversational moment to a tag, then read candidate `vibe` fields: - Real shock moment → `shock` - User confused / weird ask → `confusion` - Big freakout / rage vibe → `rage` - Joke landed / wanting to rage-bait the user → `laugh` - Sarcastic disbelief / dunk-able moment → `bruh` - Sudden attention beat / punctuation → `alert` If no specific signal: default to `bruh` — it's the catch-all and most entries carry it. ## Tag vocabulary `bruh`, `rage`, `laugh`, `alert`, `shock`, `confusion` Each entry's `vibe` field is the human-readable trigger description — **read it and match the conversational moment**, then use tags to filter candidates. ## Selection algorithm **Image+sound combo (the normal fire):** 1. Pick a mood tag based on the conversational moment. 2. Filter `images` whose `tags` contain that mood. Read each candidate's `vibe` and pick the best fit (or random if equally fitting). 3. **You only Read the image.** The hook picks the sound automatically: - If `silent: true` → no sound, image only. - If `lock_sound` set → that sound plays. - If `preferred_sounds` set → 80% pick from there, 20% tag overlap. - Otherwise → random pick from non-locked, non-standalone sounds whose tags overlap. 4. **Cooldown (your job):** track the last 3 image slugs fired this session. Don't repeat. If everything's on cooldown, pick from another related mood. **Sound-only fire:** 1. Pick a mood tag. 2. Filter `sounds` whose `tags` contain that mood AND `locked != true`. (Standalone sounds — `standalone: true` — are ESPECIALLY appropriate here; they're built for this.) 3. Read the `vibe` and pick. Run `Bash afplay &`. The `&` is required for background play. 4. Same cooldown idea (last 3 sound slugs). Variety is the point: don't keep re-picking fahhh / bruh. Rotate. ## How to emit memes **Image (with auto-paired sound):** - Write one short text line, then emit `Read `, then write one short text line. The hook plays the sound. Done. - **Multiple images in one response are fine** — repeat the text-Read-text pattern. The intervening text breaks the UI's tool-call grouping, so each image renders inline. **Sound only:** - Emit `Bash afplay &` — the `&` is required for background play. These can mix freely with any other tool calls; they don't render visually. **Variety rule:** never reuse the same image or sound slug within a single response. Rotate. **Never** pair `Read ` with `Bash afplay ` for the same fire — the hook handles the paired sound automatically. (Bash afplay is for sound-only fires, separate from any image.) ## Critical rendering rule The Claude Desktop chat UI inlines an image **as long as the `Read` tool call is not visually grouped with other tool calls in the same UI block**. Grouping collapses the image into a "Ran a command, read a file" dropdown. The reliable way to keep `Read` un-grouped — proven through testing — is to wrap it with one short line of text before AND one short line of text after. Examples of separator text: "Here you go:", "Behold:", "👻", "Skeleton time.". The text breaks the UI's batching logic and the image renders inline. The skill invocation itself counts as a tool call, so a bare `Read` would batch with it and collapse. **Always use the text-Read-text pattern.** **Multiple images in one response:** repeat the pattern — `text → Read → text → Read → text`. Each Read is separated by text, so each renders inline independently. Do NOT cluster `Read → Read → Read` without text between them; that batches. ## Detecting the surface The `UserPromptSubmit` hook auto-detects CLI vs Desktop mode and tells you which image-emit pattern to use in the FIRE reminder. The mode is in the reminder header (`(mode: cli)` or `(mode: desktop)`). Trust it. **Detection logic (in `remind-fire.sh`):** 1. Manual override via `CLAUDE_BRAINROT_MODE=cli|desktop` env var. 2. Else if `$TERM_PROGRAM` is set → CLI mode (terminal emulator detected). 3. Else default to Desktop mode. **Mode-specific image emit:** - **Desktop:** `text → Read → text`. Renders inline. - **CLI:** `Bash /scripts/show.sh `. The script opens the image briefly (~3s, configurable via `CLAUDE_BRAINROT_SHOW_SECONDS`) then auto-closes via `qlmanage` (macOS) / `feh` (Linux). It backgrounds itself so your response continues immediately. **The PostToolUse hook listens on BOTH Read and Bash** — so whether you Read an image (Desktop) or invoke `show.sh` with the image path (CLI), the hook extracts the path and fires the paired sound automatically. Never manually pair `afplay` with an image emit. ## How the sound-pairing hook works The hook script at `/scripts/play-paired-sound.sh` receives the Read tool input as JSON on stdin and: 1. Bails if the path is not under any `.claude/skills/claude-brainrot/images/`. 2. Resolves the image's slug from the basename, looks it up in `catalogue.json`. 3. Selection priority: - `silent: true` → exit, no sound. - `lock_sound` → use that sound. - `preferred_sounds` → 80% pick from those, 20% tag overlap. - else → random pick from non-locked, non-standalone sounds whose tags overlap the image's tags. 4. Plays the chosen sound in the background (detached, cross-platform). The hook command in this skill's frontmatter resolves to `${CLAUDE_PROJECT_DIR}/.claude/skills/claude-brainrot/scripts/play-paired-sound.sh` — dynamic to whatever project Claude Code is running in, as long as the skill is installed at `/.claude/skills/claude-brainrot/`.

相关技能

Agent 工作流

Skill Creator

创造有效技能指南。 当用户想创造出新的技能(或更新现有的技能),以专业知识,工作流程,或工具集成来扩展克洛德的能力时,应该使用这种技能.

Agent 工作流

Clawdhub

使用ClawdHub CLI搜索,安装,更新并发布从taladhub.com的代理技能. 需要获取苍蝇上的新技能时使用,将安装的技能同步到最新版本或特定版本,或者发布 npm-instainddhub CLI 的新/更新的技能文件夹.

Agent 工作流

Agent Team Orchestration

管弦乐团多代理团队,任务设定周期,交接协议,审查工作流程. 使用时间: (1)建立2+特派员队伍,具有不同专业,(2)确定任务路线和生命周期(收录框_ spec_建设_审查_完成),(3)在特派员之间制定交接协议,(4)建立审查和质量关口,(5)管理特派员之间的交流和文物共享.

Agent 工作流

Superpowers

Spec-first,TDD,子代理驱动的软件开发工作流程. 当:(1)构建任何新功能或应用——触发脑暴_计划_子代理执行回路,(2)调试出一个bug或测试失败——触发系统性的根起过程,(3)用户说"让我们构建","帮助我计划","我想添加X",或"这个被打破",(4)完成一个功…