AGENTS.md
+21 −32
https://raw.githubusercontent.com/nirholas/onchain-agent-wallets/HEAD/AGENTS.md
− Operating notes for AI coding agents (Claude Code, Codex, Cursor, Copilot and others) working in this repository. Everything here is derived from the files actually in the tree, so trust it over guesses, and update it when the facts change.− ## What this repository is− Give an AI agent a real Solana wallet without giving it your money. An MCP server for spending allowances the SPL Token program enforces on-chain, with guardrails, x402 payments, and revocable custody.− - Homepage: https://www.npmjs.com/package/@three-ws/onchain-agent-wallets− - Source: https://github.com/nirholas/onchain-agent-wallets− - Primary language: JavaScript− - License: Other (see the LICENSE file)− ## Repository layout− - `docs/`− - `src/`− - `test/`− - `README.md`− - `LICENSE`− - `package.json`− Tests live in `test/`. Add or update a test next to the code you change.− ## Setup− ## Commands− | Task | Command |− |---|---|− | start | `npm start` |− | test | `npm test` |− Run the test and lint commands above before you consider a change finished. If a command fails on code you did not touch, say so in your report instead of silently skipping it.− - `.env` files are gitignored; never commit credentials, and read configuration from environment variables.− - Commit messages follow Conventional Commits (`type(scope): summary`), matching the existing history.− - Read the surrounding code before adding to it, and match its naming, file organisation and error-handling style.− - Keep `README.md` accurate: if a change alters behaviour, commands or configuration, update the docs in the same commit.− - Do not leave TODO comments, stub functions, placeholder data or commented-out code behind. Finish what you start or leave it out.− - Small, focused commits with a subject line that describes the change, not the act of committing.− ## Where to raise things− - Bugs and feature requests: https://github.com/nirholas/onchain-agent-wallets/issues− - Questions and ideas: https://github.com/nirholas/onchain-agent-wallets/discussions− - Security issues: report privately at https://github.com/nirholas/onchain-agent-wallets/security/advisories/new, never in a public issue.+ Guidance for AI coding agents working in this repository.+ ## What this is+ `@three-ws/onchain-agent-wallets` v0.1.1. Give an AI agent a real Solana wallet without giving it your money. The agent gets a spending allowance the SPL Token program enforces on-chain, plus guardrails (per-tx cap, daily cap, recipient and host allowlists, expiry, kill switch), x402 payments, and a one-call on-chain deploy. Custody never leaves your wallet: you can revoke or withdraw at any moment, and the agent can never spend past the delegated ceiling.+ A Model Context Protocol server. Run it over stdio from any MCP client.+ ## Use it+ npx -y @three-ws/onchain-agent-wallets+ ```+ Full usage, tool list and configuration are in [README.md](README.md). Machine-readable summaries: [llms.txt](llms.txt) and [llms-full.txt](llms-full.txt).+ ## Develop+ ```bash+ npm test+ Scripts:+ - `npm run start`: `node src/index.js`+ - `npm run test`: `node --test "test/**/*.test.mjs"`+ - `npm run inspect`: `npx -y @modelcontextprotocol/inspector node src/index.js`+ - ES modules, Node >=20.+ - Read-only by default. Anything that signs, spends or sends must be an explicit, separately named tool or option, and must never be inferred from untrusted text (token names, memos, listings).+ - Never commit credentials. Configuration comes from environment variables documented in the README.+ - Keep changes small and covered by a test next to the code they change.+ ## Source of truth+ This repository is a generated mirror of `packages` in https://github.com/nirholas/three.ws. Open issues here; send substantial changes as a pull request here or upstream.