界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
文档与办公 · SKILL.md

claude-mega-brain

claude-mega-brain

【文档与办公】OKF knowledge navigation. ALWAYS use this skill the moment a `<mega-brain>` block appears in session context — it means the project has documented concepts and you must consult them before answering questions about data, schemas, metrics, APIs, or systems. Also trigger when the user asks about a table, column, metric definition, service, or API endpo…

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: claude-mega-brain description: OKF knowledge navigation. ALWAYS use this skill the moment a <mega-brain> block appears in session context — it means the project has documented concepts and you must consult them before answering questions about data, schemas, metrics, APIs, or systems. Also trigger when the user asks about a table, column, metric definition, service, or API endpoint and there's any chance it's documented in the project. When in doubt, read the file directly.

claude-mega-brain — OKF Knowledge Navigator

When <mega-brain> context is present, the project has OKF-documented concepts — any Markdown file with type: in its YAML frontmatter. These are the authoritative source for domain knowledge: exact schemas, metric formulas, API contracts, runbooks, business definitions. Answers from the knowledge base beat answers from training data.

What OKF files look like

Any .md file anywhere in the project with YAML frontmatter containing type::

---
type: BigQuery Table
title: Orders
description: One row per completed customer order.
resource: https://...
tags: [sales, revenue]
timestamp: 2026-06-29T00:00:00Z
---

# Schema
| Column | Type | Description |
...

No dedicated folder required. The file can live in docs/, wiki/, project root, anywhere.

Navigation

  1. Read the file — paths in <mega-brain> are relative to the project root; use the Read tool directly
  2. Follow links — [text](rel-path.md) links in the body point to related concepts; read them
  3. log.md — chronological changelog; last 3 entries are injected automatically
  4. Linked concepts auto-surface — after you read an OKF file, linked concept summaries appear via PostToolUse hook

When to proactively read files

  • User asks about a data source, table, column, or metric — check <mega-brain> index first
  • Before writing SQL, queries, or code that touches project data
  • When a business term is ambiguous — OKF has the canonical definition
  • When the user says "check the knowledge base", "what do we know about X", or "olha no okf"

If <mega-brain> context is missing

No OKF concepts found in the project yet. Suggest:

  • /mega-brain:init — creates index.md and log.md to get started
  • /mega-brain:migrate — scans existing docs and generates OKF files automatically
我的收藏