Skip to main content
The Voltr Agent Skill is an Agent Skill that teaches a coding agent — Claude Code, Codex, Cursor, and others — how to build on Voltr. Install it once and your agent can scaffold vaults, deposit/withdraw flows, custom adaptors, and CPI integrations — grounded in the official docs and the on-chain programs instead of guessing.
The Skill activates automatically when a task matches its triggers — Voltr vaults, @voltr/vault-sdk, sdk-scripts, adaptors, CPI into the vault program, and related terms — or explicitly in Claude Code with /voltr.

​
Install

The quickest path uses the open skills CLI:
Add -a claude-code -g to target a specific agent and install globally. Prefer no third-party CLI? Clone and run the bundled installer:
install.sh also targets other agents and formats — --codex, --cursor, --windsurf, --cline, --continue, --agents-md, plus --all, --project, and --path modifiers. Update an installed copy with git -C ~/.claude/skills/voltr pull.

​
What’s inside

The Skill uses progressive disclosure: the agent loads SKILL.md first for orientation and a track picker, then pulls in only the dense reference and runnable example it needs for the task at hand. A shared architecture.md reference underpins all four tracks: programs, fund flow, roles, accounting (LP / asset-per-LP / locked profit / high water mark / fees), PDAs, deployed addresses, and the instruction and error lists.

​
How it relates to these docs

The Skill is grounded in this documentation and the same tooling it describes — it is a way to build with an AI agent, not a replacement for the references here:

CLI & Scripts

The sdk-scripts CLI and examples the manager track builds on

SDK Reference

The @voltr/vault-sdk v2 surface the Skill targets

Adaptor Creation

The three-instruction adaptor interface for yield protocols

CPI Integration

Composing protocols that CPI into the vault program
The TypeScript surface targets v2 (@voltr/vault-sdk + @solana/kit); the legacy VoltrClient (v1) API is out of scope except as a migration source. Always preview transactions (print → simulate) before executing on mainnet, and verify program addresses and discriminators against the Skill’s architecture.md and cpi-integration.md.