什么是 Ask MCP - DeepWiki 研究助手？

ask-mcp 技能是一款复杂的技术研究工具，旨在提供基于存储库级上下文的 AI 驱动答案。通过利用 DeepWiki MCP 和专门的搜索工具，它允许开发人员以高精度查询复杂架构，如 Light Protocol 的 ZK 压缩和 Solana 智能合约。对于任何使用 Openclaw Skills 构建项目并需要直接从源代码中获取即时、准确的文档和实现细节的人来说，这项技能是必不可少的。

它超越了简单的文本匹配，通过使用结构化工作流来理解领域特定的查询、收集本地和远程上下文，并使用严格的技术精确性规则合成响应。无论您是在调试压缩账户还是探索 Claude Code 功能，此技能都能确保每个答案都由实际的存储库数据支撑，消除了通常与通用 AI 模型相关的猜测。

下载入口:https://github.com/openclaw/skills/tree/main/skills/tilo-14/ask-deep-wiki

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

Ask MCP - DeepWiki 研究助手 应用场景

• 研究 Light Protocol SDK 实现，包括压缩账户和状态树。
• 获取有关 Solana 程序开发和 CPI 操作的精确技术答案。
• 探索 Claude Code CLI 功能、配置和命令结构。
• 了解 Openclaw Skills 模式、子代理和开发最佳实践。
• 为 ZK 压缩证明验证和树协调生成代码示例。

1. 查询分析：代理识别用户问题的技术领域，并将其与特定的存储库范围进行匹配。
2. 上下文收集：它对本地文件使用 Glob 和 Grep 等本地工具，并向 DeepWiki MCP 查询深度存储库上下文。
3. 并行研究：对于复杂的查询，它会在需要时触发 Task 子代理在多个存储库中执行并行研究。
4. 技术合成：代理应用严格的精确性规则，将模糊的语言替换为确切的函数名称、具体的数据结构和特定的操作。
5. 响应格式化：以技术精确度、源引用（文件路径和行号）和带注释的代码块交付答案。

Ask MCP - DeepWiki 研究助手 配置指南

要将此研究能力添加到您的环境中，请使用以下命令从官方 Openclaw Skills 存储库进行安装：

npx skills add Lightprotocol/skills

访问公共存储库不需要额外的环境变量或 API 密钥，因为该技能已配置为通过 DeepWiki MCP 与公共 GitHub 存储库进行交互。

Ask MCP - DeepWiki 研究助手 数据架构与分类体系

该技能根据存储库领域和结构化 MCP 查询来组织数据，以确保高保真信息检索。参考以下分类：

name: ask-mcp
description: "For questions about Light Protocol's SDK, smart contracts and Solana development, Claude Code features, or agent skills. AI-powered answers grounded in repository context via DeepWiki MCP."
metadata:
  openclaw:
    requires:
      env: []
      bins: []
allowed-tools:
  - Read
  - Glob
  - Grep
  - Task
  - WebFetch(https://zkcompression.com/*)
  - WebFetch(https://github.com/Lightprotocol/*)
  - WebSearch
  - mcp__deepwiki__read_wiki_structure
  - mcp__deepwiki__read_wiki_contents
  - mcp__deepwiki__ask_question
  - mcp__zkcompression__SearchLightProtocol

DeepWiki Research

Query repositories via DeepWiki MCP to answer technical questions with precise, source-backed answers.

Workflow

1. Understand the question
   • Identify what the user is asking and which domain it falls into
   • If the question is ambiguous, state your understanding and ask for clarification
2. Gather context
   • Match question to the execution steps below
   • Use Glob, Grep, and Read to find relevant local files
   • Query DeepWiki MCP (mcp__deepwiki__ask_question) and mcp__zkcompression__SearchLightProtocol for repository-level context
   • Use Task subagents for parallel research across multiple repos when needed
3. Synthesize and respond
   • Apply precision rules to the answer
   • Format per format response

Execution Steps

1. Read required context in local repo

Use Glob and Grep to locate relevant files in the current repository. Use Read to pull in specific content needed to answer the question.

2. Identify question scope

Determine the domain:

• Programs, client SDKs, architecture, implementation details
• Specific components (LightAccount, ValidityProof, CPI, etc.)

3. Fetch repository context

Select the appropriate repository based on question scope:

Light Protocol (compressed accounts, state trees, ZK compression, Light SDK)

mcp__deepwiki__read_wiki_structure("Lightprotocol/light-protocol")
mcp__deepwiki__read_wiki_contents("Lightprotocol/light-protocol")
mcp__deepwiki__ask_question("Lightprotocol/light-protocol", "your question")

Solana Development (programs, accounts, general Solana)

mcp__deepwiki__read_wiki_structure("blueshift-gg/blueshift-dashboard")
mcp__deepwiki__read_wiki_contents("blueshift-gg/blueshift-dashboard")
mcp__deepwiki__ask_question("blueshift-gg/blueshift-dashboard", "your question")

Claude Code (CLI features, commands, configuration)

mcp__deepwiki__read_wiki_structure("anthropics/claude-code")
mcp__deepwiki__read_wiki_contents("anthropics/claude-code")
mcp__deepwiki__ask_question("anthropics/claude-code", "your question")

Agent Skills (skills, subagents, patterns)

mcp__deepwiki__read_wiki_structure("anthropics/skills")
mcp__deepwiki__read_wiki_contents("anthropics/skills")
mcp__deepwiki__ask_question("anthropics/skills", "your question")

4. Apply precision rules

AVOID:

• Vague verbs: "handles", "manages", "processes", "enables", "provides"
• Abstract nouns: "operations", "management", "coordination"
• -ing endings when a precise verb exists
• Marketing language: "comprehensive", "rich", "seamless", "enables", "provides capability"
• ZK terminology in user-facing docs: "inclusion proof", "non-inclusion proof" → Instead describe what it does: "prove the address doesn't exist", "prove the account hash exists"
• Generic descriptions: "account metadata" instead of "CompressedAccountMeta"

USE:

• Exact function/method names
• Concrete data structures
• Specific operations: "sets discriminator", "creates hash", "nullifies account"
• File:line references
• "reference" when describing u8 indices that reference accounts in the accounts array (standard Solana terminology)
• "CPI" as a verb (e.g., "CPIs the Account Compression Program")
• Technical method/function names used as verbs when contextually clear

Describe exactly what happens.

5. Format response

Structure answers with:

• Technical precision
• Code examples with inline comments
• Source references (file paths, line numbers)

Example Output

Question: What do you include in instruction data for compressed account creation?

Answer:

Compressed Account Creation Instruction Data

Instruction data contains three components for proof verification and tree coordination.

Components:

pub struct InstructionData {
    proof: ValidityProof,
    address_tree_info: PackedAddressTreeInfo,
    output_state_tree_index: u8,
    // + custom account data
}

1. Non-inclusion proof

• ValidityProof proves the derived address does not exist in the specified address tree
• Client generates via getValidityProof() RPC call

2. Tree location parameters

• PackedAddressTreeInfo: Specifies which address tree registers the address (u8 index, not pubkey)
• output_state_tree_index: Specifies which state tree stores the compressed account hash

3. Custom account data

• Program-specific fields (e.g., message: String, user data)

Execution flow:

1. Address tree proves uniqueness, stores address
2. State tree stores account hash
3. ValidityProof cryptographically guarantees address is unused

Why separate trees:

Compressed accounts require client-generated cryptographic proof that address doesn't exist (unlike regular Solana where runtime checks PDA existence). Address trees enforce uniqueness; state trees store account hashes.

Packed structs use u8 indices to reference accounts in remaining_accounts, reducing transaction size.

Security

This skill does not pull, store, or transmit external secrets. It provides code patterns, documentation references, and development guidance only.

• No credentials consumed. The skill requires no API keys, private keys, or signing secrets. env: [] is declared explicitly.
• DeepWiki MCP accesses public repositories only. All mcp__deepwiki__* calls query public GitHub repositories (Lightprotocol/light-protocol, anthropics/claude-code, anthropics/skills). No authentication tokens are required or transmitted. DeepWiki does not access private repositories unless explicitly configured with a token — this skill does not configure one.
• User-provided configuration. RPC endpoints, wallet keypairs, and authentication tokens (Privy, wallet adapters) are configured in the user's own application code — the skill only demonstrates how to use them.
• Tool boundary enforced. The allowed-tools list restricts this skill to read-only operations (Read, Glob, Grep), research subagents (Task), web fetches to Light Protocol domains, and MCP queries. It cannot load other skills, write files, or execute shell commands. Verify the allowed-tools list in the frontmatter above matches these constraints.
• Install source. npx skills add Lightprotocol/skills installs from the public GitHub repository (Lightprotocol/skills). Verify the source before running.
• Audited protocol. Light Protocol smart contracts are independently audited. Reports are published at github.com/Lightprotocol/light-protocol/tree/main/audits.