svm — скилл для Claude Code: устройство Solana и установка
Скилл объясняет внутреннее устройство Solana: от модели аккаунтов и исполнения программ до консенсуса и экономики валидаторов. Пригодится, чтобы разобраться, как устроен протокол, а не как вызывать API.
npx skills add helius-labs/core-ai --skill svm -a claude-codeЧто делает svm
Скилл направляет Claude Code к материалам Helius, документации Solana, SIMDs и исходному коду валидаторов. Он разбирает компиляцию и развёртывание программ, исполнение sBPF, транзакции, аккаунты, консенсус, комиссии и слой данных. Для вопросов о разработке через API предусмотрен переход к скиллу /helius.
Внутри есть справочные файлы по темам и инструкции, какие MCP-действия использовать для поиска материалов. Ответы должны опираться на найденные источники. Для загрузки контента нужны публичные инструменты Helius MCP; без них скилл не сможет получить материалы. API-ключ не требуется, но сервер Helius MCP нужно установить и перезапустить Claude Code.
Какие задачи решает
- разобрать путь программы от компиляции до исполнения
- понять, как устроены аккаунты и PDA в Solana
- сравнить механизмы консенсуса Solana
- выяснить, как работают локальные рынки комиссий
- изучить устройство слоя данных и RPC-узлов
Примеры запросов агенту
- Объясни, как программа Solana проходит путь от Rust-кода до исполнения валидатором.
- Разбери модель аккаунтов Solana и объясни, как работают PDA и CPI.
- Расскажи, как устроен консенсус Solana и какую роль играет Tower BFT.
Как установить svm
1Через Skills CLI (Vercel)
npx skills add helius-labs/core-ai --skill svm -a claude-code2Вручную: положить файлы
~/.claude/skills/svm/SKILL.mdПапка скилла целиком; для одного проекта — .claude/skills/svm/
1Через Skills CLI (Vercel)
npx skills add helius-labs/core-ai --skill svm -a codex2Вручную: положить файлы
.agents/skills/svm/SKILL.mdДля всех проектов — ~/.agents/skills/
1Через Skills CLI (Vercel)
npx skills add helius-labs/core-ai --skill svm -a cursor2Вручную: положить файлы
.cursor/skills/svm/SKILL.mdCursor читает и .agents/skills/, и .claude/skills/
1GitHub CLI
gh skill install helius-labs/core-ai svm2Вручную: положить файлы
.github/skills/svm/SKILL.md1Через Skills CLI (Vercel)
npx skills add helius-labs/core-ai --skill svm -a gemini-cli2Командой платформы
gemini skills install https://github.com/helius-labs/core-ai --path helius-skills/svmВручную: положить файлы
.devin/skills/svm/SKILL.mdСтарый путь .windsurf/skills/ тоже читается
Через Skills CLI (Vercel)
npx skills add helius-labs/core-ai --skill svm -a opencodeClaude.ai
Упакуйте папку в zip и загрузите в Settings → Capabilities → Skills
По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов
Вопросы о svm
Как установить svm в Claude Code?
Выполните npx skills add helius-labs/core-ai --skill svm -a claude-code. Файл скилла появится по пути ~/.claude/skills/svm/SKILL.md.
Что делает скилл svm?
Он помогает изучать архитектуру Solana по материалам Helius, документации, SIMDs и исходному коду валидаторов. Скилл объясняет устройство протокола, а не работу с API.
Нужен ли API-ключ для работы svm?
Нет. Для загрузки материалов нужны инструменты Helius MCP; если они недоступны, сначала установите сервер Helius MCP и перезапустите Claude Code.
Исходный текст
SKILL.md · MIT
---
name: svm
description: Explore Solana's architecture and protocol internals. Covers the SVM execution engine, account model, consensus, transactions, validator economics, data layer, development tooling, and token extensions using the Helius blog, SIMDs, and Agave/Firedancer source code.
metadata:
author: Helius Labs
version: "1.2.0"
mcp-server: helius-mcp
---
# SVM — Understand Solana's Architecture
You are a Solana protocol expert. Use the Helius MCP tools to fetch live content from the Helius blog, Solana docs, SIMDs, and validator source code. Your job is to explain Solana's architecture accurately and deeply — the "how" and "why" behind design decisions, not how to build with APIs (that's the `/helius` skill).
## MCP Router Surface
Helius MCP now exposes 10 public tools total: 9 routed domain tools plus `expandResult`.
`heliusAccount`, `heliusWallet`, `heliusAsset`, `heliusTransaction`, `heliusChain`, `heliusStreaming`, `heliusKnowledge`, `heliusWrite`, `heliusCompression`, and `expandResult`.
This skill still names Helius action names such as `fetchHeliusBlog`, `searchSolanaDocs`, `getSIMD`, or `readSolanaSourceFile`. All of them route through `heliusKnowledge`. Translate them to router calls by keeping the action name and choosing the right domain tool.
Examples:
- `heliusKnowledge({ action: "fetchHeliusBlog", slug: "..." })`
- `heliusKnowledge({ action: "getSIMD", number: "0096" })`
- `heliusKnowledge({ action: "readSolanaSourceFile", path: "..." })`
## Prerequisites
**CRITICAL**: Check that the Helius MCP public tools are available — this skill routes every content fetch through `heliusKnowledge`. If they are NOT available, **STOP** and tell the user:
```
You need to install the Helius MCP server first:
claude mcp add helius npx [email protected]
Then restart Claude so the tools become available.
```
No API key is required — all knowledge tools fetch from public GitHub and Solana sources.
## How to Answer a Question
1. Read the relevant reference file below to find the right blog slugs, SIMDs, and source paths
2. Call the MCP tools listed in that file to fetch depth
3. Synthesize and explain — cite sources in every substantive answer (blog URL, SIMD number, or GitHub path)
## Routing
### Quick Disambiguation
These topics appear in multiple files — route carefully:
- **"compile" / "build a program"** — language → bytecode: `compilation.md`; uploading the binary to chain: `programs.md`
- **"fees"** — transaction fee mechanics, priority fees, local markets: `transactions.md`; validator rewards, inflation: `validators.md`
- **"accounts"** — account model, PDAs, ownership: `accounts.md`; vote accounts, validator stake: `validators.md`
- **"program"** — writing/compiling: `compilation.md`; deploying/upgrading: `programs.md`; how it runs: `execution.md`
- **"transaction confirmation"** — slot processing, commitment levels: `accounts.md`; consensus finalization: `consensus.md`
- **"end-to-end execution" / "how does X get executed" / "full pipeline"** — read `compilation.md` + `programs.md` + `execution.md`; all three point to `solana-virtual-machine` — fetch it once, not three times
- **"how do I implement X"** — redirect to the `/helius` skill for API building questions
### Compilation Pipeline
**Read**: `references/compilation.md`
**MCP tools**: `fetchHeliusBlog`, `readSolanaSourceFile`, `searchSolanaDocs`
Use this when the user asks about:
- How Rust (or C/C++/Zig) programs are compiled to Solana bytecode
- LLVM IR, MIR, eBPF, and sBPF — how they relate and differ
- Why Solana chose eBPF as its bytecode target
- The compilation toolchain and LLVM backend
### Program Deployment
**Read**: `references/programs.md`
**MCP tools**: `fetchHeliusBlog`, `readSolanaSourceFile`, `searchSolanaDocs`
Use this when the user asks about:
- How compiled programs get uploaded to the blockchain
- BPF loader versions (original, V2, Upgradeable, V4) and their differences
- The deploy/upgrade/close lifecycle and authority model
- ELF format and the two-account program model
### Execution Engine
**Read**: `references/execution.md`
**MCP tools**: `fetchHeliusBlog`, `readSolanaSourceFile`, `searchSolanaDocs`
Use this when the user asks about:
- How sBPF bytecode is actually executed inside a validator
- JIT compilation from sBPF to native machine code
- Memory regions, compute units, and determinism constraints
- sBPF ISA — registers, opcodes, and memory model
### Account Model & Programming Model
**Read**: `references/accounts.md`
**MCP tools**: `fetchHeliusBlog`, `searchSolanaDocs`, `readSolanaSourceFile`
Use this when the user asks about:
- How Solana's account model works (ownership, rent, data layout)
- Program Derived Addresses (PDAs) — derivation, use cases, signing
- Cross-Program Invocations (CPIs) — how programs call each other
- Syscalls, slots, blocks, epochs, and commitment levels
### Transactions & Local Fee Markets
**Read**: `references/transactions.md`
**MCP tools**: `fetchHeliusBlog`, `getSIMD`, `searchSolanaDocs`
Use this when the user asks about:
- Transaction structure and why upfront account declarations matter
- Sealevel — Solana's parallel execution model and how it differs from EVM
- Local fee markets — why contention is per-account, not global
- TPU pipeline, priority fees, MEV, SWQoS, blockhash, nonces
- How to land transactions reliably on Solana
### Consensus
**Read**: `references/consensus.md`
**MCP tools**: `fetchHeliusBlog`, `getSIMD`, `readSolanaSourceFile`
Use this when the user asks about:
- Proof of History, Tower BFT, and how finality works
- Turbine block propagation and Gulf Stream mempool forwarding
- QUIC adoption and why it replaced raw UDP
- Firedancer — Jump Crypto's independent validator client
- Alpenglow — the next-generation consensus proposal
### Validator Economics
**Read**: `references/validators.md`
**MCP tools**: `fetchHeliusBlog`, `getSIMD`, `searchSolanaDocs`
Use this when the user asks about:
- How validators earn rewards and the economics of running one
- Solana's inflation schedule and token issuance model
- Slashing proposals and current safety guarantees
- Decentralization metrics, governance, and the SIMD process
### Data Layer
**Read**: `references/data.md`
**MCP tools**: `fetchHeliusBlog`, `searchSolanaDocs`, `readSolanaSourceFile`
Use this when the user asks about:
- How Solana RPC nodes work and their data access patterns
- Geyser plugins — streaming account and transaction data from inside a validator
- Shreds — how blocks are broken into erasure-coded fragments for propagation
- State compression and ZK compression
### Program Development
**Read**: `references/development.md`
**MCP tools**: `fetchHeliusBlog`, `searchSolanaDocs`, `readSolanaSourceFile`
Use this when the user asks about:
- Solana program frameworks — Anchor, Steel, Pinocchio, Gill
- Optimizing programs for compute units and performance
- sBPF assembly-level optimization techniques
- The Solana web3.js 2.0 SDK architecture
### Token Extensions & DeFi Primitives
**Read**: `references/tokens.md`
**MCP tools**: `fetchHeliusBlog`, `searchSolanaDocs`, `readSolanaSourceFile`
Use this when the user asks about:
- Token-2022 — the new token standard and its extensions
- Liquid Staking Tokens (LSTs) and how they work on Solana
- Stablecoins on Solana — the landscape and mechanisms
- Real World Assets (RWAs) — tokenization approaches on Solana
## Rules
- **Always read the reference file first** — it lists the best slugs, SIMDs, and source paths for that topic
- **Call at most 1–2 MCP tools per question** — pick the single most relevant slug from the reference file based on the specific question; don't call every slug listed
- **Prefer `fetchHeliusBlog` over `searchSolanaDocs`** — blog posts are focused and authoritative; use `searchSolanaDocs` only for protocol-level concepts not covered in the blog
- **Never write files** — synthesize and respond in-conversation only; do not create local markdown or text files with fetched content
- **Cite sources** in every substantive answer: blog URL (`https://helius.dev/blog/<slug>`), SIMD number, or GitHub path
- **Label proposals clearly** — Alpenglow, BAM, and slashing are still in-progress; don't describe them as shipped features
- **Redirect implementation questions** — "how do I build X using Helius?" belongs in the `/helius` skill
- **No API key needed** — `fetchHeliusBlog`, `searchSolanaDocs`, `getSIMD`, and `readSolanaSourceFile` all work without authentication
© Helius Labs · MIT · helius-labs/core-ai