PlugWiki

solana-dev skill: build Solana dApps, Anchor programs and tests with your agent

The Solana Foundation's skill that gives your agent current Solana know-how for dApps, Anchor and Pinocchio programs, testing and security reviews.

Installs
105.3K
on skills.sh
Stars
582
on GitHub
Files
35
besides SKILL.md
Updated
Oct 2026
last repo commit

solana-dev is a large skill that tells your agent which Solana tools to pick and how to use them. Its defaults are @solana/kit 8 clients with v1 transactions, Anchor 1.2.x for programs, Surfpool for integration tests and LiteSVM or Mollusk for unit tests. It loads when you ask to build a Solana dapp, write an Anchor program, create a token, debug Solana errors or test a program.

It bundles no scripts but 35 reference guides that the agent reads only when needed. SKILL.md also tells the agent to install the Solana Developer MCP server if it is missing, and to run CLI tools such as anchor and surfpool with NO_DNA=1. You need Node.js 20.18+, a Rust toolchain, the Solana CLI and the Anchor CLI for most tasks.

Good for

  • Developers building Solana dApps with React or Next.js and wallet sign-in
  • Program authors writing in Anchor or Pinocchio who need tests and an audit-style review
  • Teams fixing toolchain errors or upgrading Anchor and web3.js versions
  • Backend builders adding sponsored fees, KMS signing or token issuance

Install solana-dev

Claude Code

  1. 1

    Copy this prompt.

    Install the solana-dev skill in Claude Code. Run this command:
    
    npx skills add solana-foundation/solana-dev-skill
    
    Then check that it loaded and tell me how to use it.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

Most Solana tasks also need Node.js 20.18+, a Rust toolchain, the Solana CLI and the Anchor CLI.

From the project README. Source.

What's in the skill

When it loads

Use when user asks to "build a Solana dapp", "write an Anchor program", "create a token", "debug Solana errors", "set up wallet connection", "test my Solana program", "fuzz my Solana program", "deploy to devnet", "send a v1 transaction", "support larger transactions", "fix maxSupportedTransactionVersion", or "explain Solana concepts" (rent, accounts, PDAs, CPIs). Also for program architecture (state layout, compute units, throughput, naming), quick on-chain lookups via public RPC + curl, gasless or sponsored-fee transactions (Kora), server-side signing with KMS/Turnkey/Privy/Fireblocks (Solana Keychain), and issuing stablecoins or allowlisted tokens (Mosaic). Covers wallets, Anchor/Pinocchio, Codama clients, Surfpool/LiteSVM/Mollusk testing, security review, and v1 transactions (SIMD-0385). Prefers @solana/kit 8 plugin clients building transaction v1 by default, @solana/kit-plugin-wallet + @solana/react for wallets, web3.js v3 as the legacy migration target, and Surfpool for local networks.

Sections in SKILL.md

  1. What this Skill is for
  2. Default stack decisions (opinionated)
  3. Agent safety guardrails
  4. Transaction review (W009)
  5. Untrusted data handling (W011)
  6. Agent-friendly CLI usage (NO_DNA)
  7. Operating procedure (how to execute tasks)
  8. 1. Classify the task layer
  9. 2. Pick the right building blocks
  10. 3. Implement with Solana-specific correctness
  11. 4. Add tests
  12. 5. Deliverables expectations
  13. Solana MCP server (live docs + expert assistance)
  14. Auto-install
  15. Available MCP tools
  16. When to reach for MCP tools

Key facts

Installs on skills.sh
105,283 as of 2026-11-02
GitHub stars
582 (2026-11-02)
Last repo commit
2026-10-27

What it includes

  • @solana/kit 8, Kit wallet plugin, Anchor 1.2.x, Surfpool

    Sets the defaults your agent uses: Kit plugin clients with transaction v1, wallet connection through the Kit wallet plugin and @solana/react, web3.js v3 only as a migration target, Anchor for programs, Pinocchio for performance.

    Try: “Help me set up a Next.js app with Solana wallet connection.”

    Read · Default stack decisions

  • Never sign without approval; devnet first; simulate first

    Tells the agent not to sign or send without your approval, never to ask for or store private keys or seed phrases, to default to devnet or localnet, and to simulate before sending. On-chain data is treated as untrusted.

    Read · Agent safety guardrails

  • Runs surfpool and anchor commands without prompts

    Tells the agent to prefix CLI calls such as `anchor build` and `surfpool start` with NO_DNA=1 so tools skip interactive prompts.

    Runs code · NO_DNA=1 prefix

  • Classify layer, pick blocks, be explicit, add tests, report risks

    Five steps: classify the task layer, pick building blocks, spell out cluster, fee payer, compute budget and transaction version, add tests, then list changed files and risk notes for signing, fees and token transfers.

    Read · Operating procedure

  • One-shot RPC reads with curl against public endpoints

    For a single balance, transaction or token-account read, the agent calls a public Solana RPC with curl instead of scaffolding a project. It defaults to mainnet-beta for real wallets and says so.

    Try: “What is the SOL balance of this wallet address?”

    Runs code · Quick on-chain lookups

  • Adds the Solana Developer MCP server if it is missing

    SKILL.md tells the agent to check for the server and add it with the host's MCP command (claude mcp add, gemini mcp add or codex mcp add) if absent. It gives three tools: an expert helper, docs search and an Anchor expert.

    Runs code · Solana MCP server auto-install

  • Program writing, architecture, CU limits and account lifecycle

    Try: “Create an Anchor program for a simple escrow.”

    Files · references/programs/ (anchor, pinocchio, design-patterns)

  • Surfpool, LiteSVM and Mollusk tests; 26 surfnet cheatcodes

    Covers unit tests with LiteSVM or Mollusk and integration tests on Surfpool, a local network that forks mainnet. The guide shows installing Surfpool with a curl-to-bash script.

    Try: “Write Surfpool integration tests for my token transfer flow.”

    Runs code · references/testing.md and surfpool/

  • Vulnerability patterns and an audit-style review checklist

    Try: “Review this program for security issues.”

    Read · references/security.md

  • Error-to-fix map and Anchor/Solana/Rust/GLIBC versions

    Fixes for version mismatches, GLIBC errors and dependency conflicts. Some fixes delete cache and build folders, such as ~/.cache/solana and target/.

    Try: “I'm getting GLIBC_2.39 not found when running anchor.”

    Runs code · references/common-errors.md and compatibility-matrix.md

  • Upgrade Anchor 0.32 programs to 1.x

    Runs code · references/anchor/migrating-v0.32-to-v1.md

  • Kit clients, plugins, React bindings, codecs, web3.js v3 routing

    Try: “Migrate this web3.js v1 script to web3.js v3.”

    Read · references/kit/ and kit-web3-interop.md

From the official docs

Permissions

Medium risk

Your agent can run installs, delete cache folders and, if you approve, sign Solana transactions that move real funds. Stay on devnet, keep keys out of the chat and read each command before you approve it.

What it can do

  • Read your data
  • Create and change things
  • Delete things
  • Read-only mode

Watch out for

  • Mainnet transactions move real money

    The skill tells the agent to use devnet by default and ask before signing, but that is an instruction. Check cluster, recipient and amount yourself before you approve any signature. github.com

  • It tells your agent to add an MCP server

    SKILL.md says that before any Solana task the agent should check for the Solana Developer MCP server and, if it is missing, add it with claude mcp add, gemini mcp add or codex mcp add. It names no confirmation step, only 'ask the user' when the agent cannot edit the MCP config. The Codex command runs npx -y mcp-remote. Once connected, the agent is told to use it for conceptual and debugging questions, so those queries go to mcp.solana.com. Decline the install if you do not want that. github.com

  • Guides run curl-to-bash and cargo installs

    The Surfpool guide (also testing.md and common-errors.md) installs with curl -sL https://run.surfpool.run/ | bash, and common-errors.md installs the Solana CLI with sh -c "$(curl -sSfL https://release.anza.xyz/stable/install)". Others run cargo install for Anchor (from git repositories) and Kora. Read each command before approving. github.com

  • Some fixes delete folders

    common-errors.md shows rm -rf on ~/.cache/solana, old Solana releases, ~/.cargo/registry/cache, target/ and test-ledger/. Make sure each path is what you expect. github.com

Technical details
Talks tohttps://mcp.solana.com/mcp when the MCP server is installed. Public Solana RPC endpoints such as api.mainnet-beta.solana.com for quick lookups. run.surfpool.run, release.anza.xyz, cargo, git and npm downloads when the guides install tools
Sign-in and keysThe guides say never to request, generate or store private keys or seed phrases and to use wallet signing flows. Server signing guides read KMS key IDs and API keys from environment variables.
Scopes
  • Your agent can write program and app code and run anchor, surfpool, cargo, npm and curl commands

  • It is told to add the Solana Developer MCP server to your agent's config if it is missing

  • Guides show deleting cache and build folders to fix errors

Questions

What does the solana-dev skill do?

It gives your agent the Solana Foundation's current defaults and guides for Solana development. That covers dApp UI and wallets, Anchor and Pinocchio programs, testing, security review and toolchain fixes.

How do I install it in Claude Code, the Claude app or Codex?

Use the install box at the top of this page. The README says that for Claude app uploads you should use the solana-dev.zip from the latest release, not GitHub's "Download ZIP", because that archive nests SKILL.md too deep.

Does it need an API key?

The skill itself does not ask for one. Some guides do: the Kora guide mentions KORA_API_KEY and JUPITER_API_KEY, and the Keychain guide reads KMS_KEY_ID and KMS_PUBLIC_KEY from the environment. Those apply only if you build those features.

Is it safe to use with real funds?

It tells the agent to default to devnet, simulate first and never sign or send without your approval. Those are instructions, not enforced limits, so keep keypairs and seed phrases out of the chat and check any mainnet transaction yourself. The guides also run cargo, npm and curl-to-bash installs (Surfpool and the Solana CLI), and some fixes delete cache folders.

Does it install the Solana MCP server for me?

SKILL.md tells your agent to check for the Solana Developer MCP server before any Solana task and add it if it is missing, using your host's MCP command (claude mcp add, gemini mcp add or codex mcp add). It names no confirmation step, so watch for the command in your approval prompt. If you decline it, the agent has no MCP docs search to follow the skill's "use it first" advice.

Which tools does it expect me to have?

SKILL.md lists Node.js 20.18+, a Rust toolchain, the Solana CLI and the Anchor CLI. Surfpool is its default for local networks.

Does it work in Cursor, Copilot or Gemini CLI?

The README says it works with Claude Code, Codex / ChatGPT desktop, GitHub Copilot, Gemini CLI, Cursor, Windsurf, Cline and OpenCode, and lists the skill folder for each.

Information

updated Nov 2, 2026

Folder
skills/solana-dev

solana-foundation/solana-dev-skillskills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 10
  1. solana-dev-skill README · github.com
  2. solana-dev SKILL.md · github.com
  3. skills.sh listing · skills.sh
  4. LICENSE · github.com
  5. solana-dev testing.md · github.com
  6. common-errors.md · github.com
  7. kora.md · github.com
  8. keychain.md · github.com
  9. surfpool/overview.md · github.com
  10. solana-dev rpc-quick-lookups.md · github.com

Updated Nov 2, 2026.