solana-dev-skill packs the 2026 Solana stack into your coding agent

A skill that encodes the current Solana stack

Agent skills live or die by how current their knowledge is, and this one leads with its date: the Solana Foundation's solana-dev-skill claims July 2026 best practices, packed as a standard SKILL.md bundle any coding agent can read. The README lists first-class support across Claude Code, OpenAI Codex and ChatGPT desktop, GitHub Copilot in CLI and coding-agent forms, Gemini CLI, Cursor, Windsurf, Cline and OpenCode, which is to say anything that reads agent skill files.

The scope is the whole dApp and program surface: the modern SDK, wallet connection, program development in Anchor and Pinocchio, testing, client codegen, security patterns and toolchain troubleshooting. For a working Solana developer the pitch is simple: your agent stops answering Solana questions with 2024 folklore.

What the skill actually knows

The references directory is the real content, and it is organized like a small docs site. Kit documentation covers the plugin system, React integration, codecs, accounts and Codama client generation. Testing material spans Surfpool for integration tests, with mainnet forking and cheatcodes, plus LiteSVM and Mollusk for unit tests. A security reference collects vulnerability patterns and prevention, and a compatibility matrix maps Anchor, Solana CLI, Rust and GLIBC versions against each other.

The long tail is where the value hides: common-errors.md mapping error messages to solutions, a migration guide for Anchor v0.32 to v1.1, transaction v1 material for the SIMD-0385 larger-transaction world, confidential transfers under Token-2022, and a Surfpool cheatcodes reference listing twenty-six surfnet methods. That is curated institutional knowledge, the kind agents rarely reconstruct correctly from stale training data.

Installing it into one agent or all of them

The quickest route rides the skills CLI, which detects your installed agents and installs or symlinks the skill for each:

npx skills add solana-foundation/solana-dev-skill

The repository's own installer covers more deliberate setups:

git clone https://github.com/solana-foundation/solana-dev-skill
cd solana-dev-skill
./install.sh            # user-level: ~/.agents/skills + ~/.claude/skills
./install.sh --project  # project-level: .agents/skills + .claude/skills
./install.sh --link     # symlink instead of copy (auto-updates with git pull)

Manual installation is a copy or symlink of skills/solana-dev/ into your agent's skills directory, and the README's table maps each supported agent to its project and personal paths, with the cross-agent .agents/skills/ convention plus .claude/skills/ covering most combinations.

Opinionated defaults, stated as a table

Rather than cataloguing every option, the skill takes positions and prints them. Client SDK work defaults to @solana/kit v7 with plugin clients, with web3.js v3 RC positioned as the migration target for legacy v1 codebases rather than a first choice. Wallet and UI work goes through the Kit wallet plugin and @solana/react. Program development defaults to Anchor 1.1.x, with Pinocchio 0.11 and later reserved for compute-unit-sensitive code.

Testing splits by level: LiteSVM or Mollusk for unit tests, Surfpool for integration, explicitly preferred over the older solana-test-validator. Codama is the default for IDL and client generation. Whether you agree with each call matters less than the fact that the skill will answer consistently with them, which is exactly what you want from a shared team configuration: one opinion, versioned, instead of every session re-litigating the stack.

Where a skill beats a docs search

The README's example prompts show the intended daily use: scaffold a Next.js app with wallet connection, write an escrow program in Anchor, convert it to Pinocchio for compute efficiency, migrate a web3.js v1 script to v3, write Surfpool integration tests for a token transfer, or review a program for security issues. Then the unglamorous ones that actually save evenings: a GLIBC_2.39 error from anchor, version compatibility between Solana CLI and Anchor 1.1, an Anchor upgrade path from 0.32.

Those toolchain questions are where general web search fails softly, because answers are scattered across versions and GitHub threads. A curated compatibility matrix and error map inside the agent's context turns them into one-turn answers. The skill triggers automatically when the conversation enters its territory, so there is no ceremony to remember beyond having it installed.

Limits to keep in mind

A skill is knowledge, not a runtime. Installing it does not install Rust, the Solana CLI or Anchor; it makes the agent better at using whatever toolchain exists, and the compatibility reference will tell the agent what versions to ask you for. The knowledge is frozen at authoring, labeled July 2026 best practices, so a fast-moving ecosystem will outpace specific version numbers; the git repository, not the skill text, is the update channel.

Provenance is respectable but worth reading: content draws on the Blueshift learning platform, official Solana docs, Kit documentation and the web3.js v3 tree, among others. The repo is MIT licensed, carries no releases or tags, and its last push was on 2026-09-09. Treat the version labels inside the skill as claims to verify against your project, the same way you would treat any tutorial's versions.

Against docs, courses and hand-rolled prompts

The alternatives are familiar. Official Solana documentation is authoritative and current but lives outside the agent, so every answer costs a context switch and a copy-paste. Blueshift's courses, which this skill credits as a source, teach humans systematically but do nothing for a coding session. Hand-rolled prompt files are what most teams use today: a CLAUDE.md full of Solana notes that rots privately, differs per developer and cites nothing.

This skill's bet is that the knowledge should be shared, versioned, installable across every major agent, and maintained by the foundation whose stack it describes. If your team writes Solana code with agents more than occasionally, that bet looks sound; if you touch the ecosystem once a quarter, the official docs and a bookmark folder remain perfectly adequate.

Editorial conclusion

solana-dev-skill is worth one npx command for any team or solo developer writing Solana programs or dApps with coding agents, especially if Anchor versions, testing setup or toolchain errors currently cost you real time. Skip it if you never touch Solana, or if your agents lack Agent Skills support entirely. Verify its fit in one session: install it, ask your agent the version-compatibility question that last bit you, and check the answer against the compatibility matrix in references before you trust the skill on anything load-bearing.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. Project website
4. README
5. solana-foundation/solana-dev-skill on GitHub