WHAT NEW IN X11.03 TUE

X 上的 AI 合集档案 · 22 个合集 · 66,489 条

今天新增jevVIDEO

Claude Code 插件化配置与 JEV 微决策分层

把 skills、agents、hooks、MCP 打包成一个插件一键安装,并让 JEV 在毫秒内处理无需大模型的分支。

delost
@thedelost1 小时前收录原推发布于 2026.11.03
delost @thedelost原推发布于 2026.11.03
在 X 上打开原推

THIS IS A PURE F*CKING GOLDMINE Anthropic just shared one of the best tricks for Claude Code. The idea: a plugin is one directory of skills, subagents, hooks and MCP servers that Claude Code installs and loads as a single unit. Your whole setup stops living in one laptop's ~/.claude and becomes something a teammate installs with one command. You just run this once in Claude Code: /plugin marketplace add org/repo here's the split: → skills/ ships the instructions, and each one runs as /plugin-name:skill-name → agents/ ships the subagents, so your reviewers travel with the repo → hooks/ ships the rules that fire on lifecycle events, like after every edit → .mcp.json ships the servers, so the whole team connects to the same systems A plugin earns its place at three moments: → a new machine, where one command rebuilds the entire setup → a new teammate, who installs your workflow instead of reading your docs → a version bump, where everyone moves at once Skills instruct. Agents delegate. Hooks enforce. JEV engineering is the same move one layer down: the forks that need no thinker (which file, which tool, retry or stop) go to JEV in milliseconds, and the big models only see the ones that split the full tree > a plugin is a directory with .claude-plugin/plugin.json and its components beside it > scope decides who gets it: user, project through committed settings, or local > an enabled plugin is in every session, so each name and description costs context every turn > plugin skills are namespaced, so they never collide with the ones you wrote > edits to hooks, agents and .mcp.json need /reload-plugins before they take effect > JEV routes micro-forks in milliseconds paste the tree and this prompt into Claude Code ↓ "Turn my setup into a plugin from this tree: 1. Create .claude-plugin/plugin.json and move my reusable skills, subagents and hooks into skills/, agents/ and hooks/. 2. Put the MCP servers the team shares into .mcp.json so everyone connects to the same ones. 3. Install it at project scope through the committed .claude/settings.json, not at user scope. 4. List what each component costs in context every turn, and cut anything I have not used this month. 5. Route fast deterministic forks (tool choice, path selection, retries) through JEV. Show me every change as a diff first. No edits until I say go." ↳ http://code.claude.com/docs/en/plugins save it, it'll come in handy.

这条是引用着另一条写的:x.com/hanakoxbt/status/2110067711363457244

x.com/thedelost/status/2117361587975492093采集于 2026.11.03 10:06