Treasures Finance Agent Skills

Agent Skills for building AI agents on the Treasures finance APIs.

A skill is a folder of plain-Markdown instructions (SKILL.md) that a coding agent loads on demand. The skills here teach an agent to call the Treasures finance APIs correctly — discover tokenized stocks, quote and execute trades, bridge USDC across chains, operate a delegated wallet, and read portfolios — including the signing details and footguns that are easy to get wrong.

Skill catalog

⚠️ Read this before any trade attempt (verified live 2026-07-27)

Two things fail silently if you don't know them. Both were confirmed with live calls, not docs.

1. Base URL needs a path prefix

https://api.treasures.io alone 404s on every path. Two planes off the same host:

2. Trade paths are geo-fenced at the load balancer

Reads are open worldwide; writes are not.

• The block sits at the AWS load balancer (server: awselb/2.0), matched on path + real TCP source IP, before auth and before the app. Unknown paths return 404 (not 451), and GET/OPTIONS on a trade path also 451 → the path is fenced, not the write method. Both planes are covered — /api/v1/quote/sell 451s identically to /public/v1/quote/sell.
• Spoofed geo headers do not work (X-Forwarded-For, CF-IPCountry, X-Real-IP, True-Client-IP all still 451).
• Blocked: US (default Starchild egress is San Jose, CA) and GB. Clear: DE, NL, SE, CH, SG, HK, JP — tested via the sc-vpn skill with a per-request proxy (proxies= / curl -x). Never set a global HTTP_PROXY; it breaks sc-proxy billing.

A 451 is never a wallet, signing, or delegation problem. Don't debug signatures until you're past it.

Error ladder on /quote/sell — tells you which layer rejected you

451 unavailable_for_legal_reasons  → geo edge (change exit region)
400 invalid_request                → schema (fix the body)
401 ownership_proof_eth_invalid    → signature (re-sign the challenge)
422 holdings_insufficient          → auth OK, wallet just lacks the asset

Reaching 422 proves geo and signing both work — it's the healthiest failure you can get.

Eligibility — not optional

The 451 is an eligibility rule working as designed, not an outage. Ondo/xStocks tokenized equities are not offered to US persons, and Ondo additionally enforces KYC/transfer whitelists at the token contract, so an ineligible wallet is rejected on-chain no matter what the exit IP says. Confirm the user is eligible in a permitted jurisdiction before routing around the fence. Clearing a geo-block for a user who is genuinely non-US is fine; using it to evade a securities restriction is not.

How to route trade calls through sc-vpn

Prerequisite: the sc-vpn skill (official). Install if missing — npx skills add Starchild-ai-agent/official-skills --skill sc-vpn --agent openclaw. No credentials; internal network only. Health check: curl http://sc-vpn.internal:8081/health → {"status": "ok"}.

Proxy URL format is http://<region>:[email protected]:8080 — the username is the region code. Omit it and you get 400 No region specified.

Rule: reads go direct, only the fenced trade paths go through the VPN. Sending everything through the tunnel is slower and pointless — /stocks, /portfolio and /trades already work from anywhere.

import requests, time
from core.skill_tools import wallet   # run from a dir where `core` is importable (e.g. /tmp or /data/workspace)

BASE = "https://api.treasures.io/public/v1"
VPN  = "http://de:[email protected]:8080"          # de | nl | se | ch | sg | hk | jp
PROX = {"http": VPN, "https": VPN}                  # per-request ONLY — never os.environ

ETH = "0x...".lower()                               # wallet address, lowercase

# READ — direct, no proxy
pf = requests.get(f"{BASE}/portfolio", params={"eth_wallet": ETH}, timeout=30).json()

# WRITE — through the VPN, with a signed ownership proof
issued    = int(time.time())
challenge = f"treasures-finance-quote-v1\n{issued}\n\n{ETH}"     # verified format; note the blank line
sig       = wallet.wallet_sign(challenge)["signature"]

body = {
    "ticker": "GOOGL", "chain": "eth",
    "amount_shares": "0.01", "max_slippage_bps": 100,
    "eth_wallet": ETH,
    "ownership_proof": {"issued_at": issued, "eth_signature": sig},
}
r = requests.post(f"{BASE}/quote/sell", json=body, proxies=PROX, timeout=60)
print(r.status_code, r.text)      # 422 holdings_insufficient = geo + signing both OK

Same thing from bash:

curl -x "http://de:[email protected]:8080" \
     -X POST https://api.treasures.io/public/v1/quote/sell \
     -H 'content-type: application/json' -d @body.json

Confirm the exit before trusting a result: curl -x "http://de:[email protected]:8080"; https://ipinfo.io/json.

Footguns

Keep the whole trade sequence (/quote/* → /trade/submit, and /bridge/quote → bridge submit) on the same exit region. Mixing exits mid-flow risks the backend seeing an inconsistent origin between legs.

Install

# Install one sub-skill (recommended — pulls just what you need)
npx skills add treasures-io/treasures-finance-agent-skills --skill treasures-b2b-api
npx skills add treasures-io/treasures-finance-agent-skills --skill treasures-wallet

# Or install everything (auto-detects your environment and installs accordingly)
npx skills add treasures-io/treasures-finance-agent-skills

npx skills installs SKILL.md files into the right place for 70+ coding agents (Claude Code, Codex, Cursor, GitHub Copilot, Windsurf, Cline, OpenCode, …) and auto-detects which ones you have. It reads this repo's skills/<name>/SKILL.md layout directly, so no extra setup is required.

Target specific agents with -a:

npx skills add treasures-io/treasures-finance-agent-skills -a claude-code -a codex -a cursor

License

[MIT](LICENSE)