跳到正文
cccskills
免费在 GitHub 公开

context-prep

LLM Utility: Generate minimal project context map before starting any workflow. ALWAYS run this first at session start, before /prd, before /implement, or after context loss. Triggers on 'context prep', 'orient me', 'start', 'new session'.

中文简介正在准备中。当前显示原文说明。

OffchainLabs/arbitrum-docs星标 252(GitHub 252 + cccskills 0)

实测评分 未实测

查看安装方法

包含的文件(1)

  • SKILL.md4.4 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Purpose

Prepare LLM context before starting work. This is a utility that runs before any workflow phase.

When working on Arbitrum projects, use @arbitrum-best-practices for domain-specific patterns.

Project Context Overview

Generate a minimal, navigable context map using progressive revelation principles.

Philosophy

  1. Start shallow, drill deep only when necessary
  2. Provide pointers (file paths, line numbers), not full content
  3. Links over duplication - reference docs, don't copy them
  4. Minimize context window usage - treat it as expensive real estate

Instructions

Launch 4 parallel agents using the Task tool, then synthesize findings.

Agent 1: Documentation Survey

subagent_type: Explore (thoroughness: "quick")

PROGRESSIVE REVELATION: Build a documentation map with links, not content.

1. List the root documents (README.md, INTERNALS.md, CONTRIBUTE.md, STYLE-GUIDE.md) - document topics, don't read yet
2. Scan README.md first 50 lines - extract project purpose (1-2 sentences)
3. Check for todo.md, CHANGELOG.md, AGENTS.md - current status
4. List package.json locations - map monorepo structure

Deliver navigable map:
- Project purpose (1-2 sentences) → Link: README.md
- Current status → Link: todo.md or AGENTS.md
- Package structure → Links: packages/*/package.json
- Reference materials → Links: the root *.md files (filenames only)

Agent 2: Technology Stack

subagent_type: researcher

PROGRESSIVE REVELATION: Scout technology stack, link to deep dives.

Research via docs and MCP:
- What frameworks/libraries are used?
- What target platforms?
- Key integrations?

Deliver scout report:
- 1-sentence description of tech stack
- Top 3 constraints from architecture
- Critical dependencies (list, don't explain)
- Links to docs for deep dives

Agent 3: Git History & Progress

subagent_type: Explore (thoroughness: "quick")

PROGRESSIVE REVELATION: Surface recent activity, link to history.

1. Run: git log --oneline -10
2. Group commits by theme (1-2 words per group)
3. Check git status for uncommitted changes
4. Identify current branch

Deliver minimal summary:
- Last 5 commits (grouped by theme)
- Current branch and status (clean/dirty)
- Active work (from uncommitted files)
- Link: run `git log --oneline -50` if needed

Agent 4: Architecture Deep Dive

subagent_type: Explore (thoroughness: "quick")

PROGRESSIVE REVELATION: Map architecture with pointers, not full code.

Scan structure (file listings only):
- src/*.ts or src/**/*.ts - list main files
- contracts/*.sol - list contract files
- packages/*/src/ - list entry points

Deliver navigable map:
- Key packages (name + one-liner purpose)
- Data flow (5-step maximum)
- Entry points → Links to main files
- Critical contracts → Links to .sol files

Synthesis Format

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📋 PROJECT CONTEXT (Progressive Revelation)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

## What We're Building
[1-2 sentences]
→ Deep dive: README.md

## Current Status
- Branch: [branch]
- Progress: [summary]
- Active: [current work]
→ Full plan: todo.md

## Tech Stack
- Frameworks: [list]
- Targets: [list]
- Key integrations: [list]
→ Use MCP for details

## Architecture

**Key Packages:**
- [package]: [purpose] → path/to/main.ts

**Data Flow:**
[5-step flow]

## Reference Materials (Read IF Needed)

**Docs:**
- [list doc files]

**Code:**
- [list key source files]

## Recent Activity
[Last 5 commits grouped]
→ Full history: `git log --oneline -50`

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
💡 Context loaded. Traverse links for deeper detail.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Context Economy

  • This command: ~500 tokens (navigation map)
  • Reading one doc: ~2000 tokens (targeted detail)
  • Reading full codebase: ~50,000 tokens (avoid unless critical)

Progressive revelation saves 95%+ of context for actual work.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Arbitrum L2 development best practices. Use when implementing bridges, Orbit chains, SDK integrations, or debugging cross-chain transactions. Triggers on Arbitrum, L2, rollup, bridge, retryable, Orbit, Nitro, or cross-chain contexts.

中文简介正在准备中。当前显示原文说明。

OffchainLabs/arbitrum-docs2522026年10月24日 更新

Create or improve lightweight, brand-consistent SVG concept diagrams for the Arbitrum docs, using the official Google Drive infographic asset kit as the palette/style source. Use when asked to "make a diagram more efficient", replace a bloated haw-*.svg (draw.io PNG-in-SVG exports), replace an ASCII / box-drawing diagram in an MDX page with a real image, author a new concept diagram for content/docs/how-arbitrum-works or similar, or build a diagram from the Arbitrum brand assets. Hand-authored lean vector SVG — not Mermaid, not a full Excalidraw/draw.io pipeline.

中文简介正在准备中。当前显示原文说明。

OffchainLabs/arbitrum-docs2522026年10月24日 更新

Use when asked to rebrand, restyle, re-pace, resize, or re-render an existing explainer video, when a video's labels are baked-in raster pixels that cannot be edited, or when building a new animated explainer for the Arbitrum docs from a prompt or a doc section.

中文简介正在准备中。当前显示原文说明。

OffchainLabs/arbitrum-docs2522026年10月24日 更新

cast

免费

Foundry's cast CLI for Ethereum interactions. Use when querying blockchain data, calling contracts, sending transactions, encoding/decoding data, or debugging on-chain issues. Triggers on "cast", "get balance", "call contract", "decode calldata", "check tx", "get block".

中文简介正在准备中。当前显示原文说明。

OffchainLabs/arbitrum-docs2522026年10月24日 更新

Run full documentation quality audit, covering sidebar orphans, unused-file review, MDX structure, internal links, glossary references, contract addresses, variables, formatting, frontmatter, tests, and types. Triggers on "audit docs", "check docs quality", "find problems", "content audit".

中文简介正在准备中。当前显示原文说明。

OffchainLabs/arbitrum-docs2522026年10月24日 更新

Create new Arbitrum development rules following best practices format. Use when capturing a reusable pattern, common mistake, or architectural decision. Triggers on "create rule", "add rule", "new rule", "document this pattern".

中文简介正在准备中。当前显示原文说明。

OffchainLabs/arbitrum-docs2522026年10月24日 更新

查看 OffchainLabs 的全部技能

报告此技能的问题