pursekeeper skill

An OpenClaw / ClawHub skill: earn and spend Nano (XNO) with other agents. The skill text is SKILL.md; scripts/ holds a no-node Nano wallet and an x402 exact client for nano:mainnet; references/ holds the seller directory snapshot and the recipes.

Install from ClawHub once published: clawhub install pursekeeper. Until then:

git clone https://github.com/pursekeeper/skill ~/.openclaw/skills/pursekeeper   # or your skills dir
cd ~/.openclaw/skills/pursekeeper/scripts && npm install

Everything here is MIT-0, like every ClawHub skill. Source of the recipes and the endpoints: https://github.com/pursekeeper/api. Site: https://pursekeeper.dev.

Tests

With Node.js 18 or newer, install the root development dependency once, then run:

npm install --include=dev --ignore-scripts
npm test

(--include=dev: on a box with NODE_ENV=production npm omits development dependencies and the tests then stop at “Cannot find module ‘nanocurrency’”.)

npm test uses node --test test/ (test/index.js is the directory entry point). It needs no Internet access, API key, wallet, or funds after dependency installation. Each test starts an ephemeral 127.0.0.1 HTTP server and runs the real scripts/no-node.js CLI in a child process with a new throwaway seed. The child gets only explicit test environment variables, and its fetch is restricted to the mock’s exact origin with redirects disabled. No real proof of work is generated.

The mock implements account_info, receivable, work, process, and verify. It accepts and hashes the first block, updates the actual frontier and balance, then returns a 502/Fork error instead of the success reply (the node-like stand from issue #3). A valid second send is accepted too, so the mock exposes a double payment rather than hiding it. Both tests require one process call, one accepted block, one balance change, and a CLI receipt containing the original block hash. The receive case also removes the pending source when pocketed, rather than leaving a stale receivable.

To reproduce the regression against the historical scripts without modifying production code or fetching anything during the tests:

snapshots=$(mktemp -d)
git show fe43bbd:scripts/no-node.js > "$snapshots/0.1.3.cjs"
git show 8fe3ac7:scripts/no-node.js > "$snapshots/0.1.4.cjs"
NO_NODE_SCRIPT="$snapshots/0.1.3.cjs" npm test # expected failure
NO_NODE_SCRIPT="$snapshots/0.1.4.cjs" npm test # expected success
npm test                                    # current scripts
rm -r "$snapshots"

On 0.1.3 the send test reports two process calls and accepted blocks instead of one; the receive test also fails because the CLI skips the pocketed source without returning its accepted hash. Both pass on 0.1.4 and current main. These tests model that specific Fork-shaped lost reply; they do not claim to