Market Evidence Check

Purpose

Read-only pre-execution evidence for a stated OKX instrument, side, and hypothetical USDT notional. It reads a public OKX order-book snapshot and returns spread, estimated price impact, estimated one-way cost against a declared bps budget, quote age, and displayed-depth checks. Use when the user wants to check market cost/freshness before deciding anything themselves.

GO only means the declared checks passed on that public snapshot. It is not a buy/sell recommendation, trading signal, profit prediction, fill guarantee, or permission to trade.

Tooling

All commands go through bin/market_evidence_check.py (Python standard library only):

bin/market_evidence_check.py health
bin/market_evidence_check.py check --instrument BTC-USDT --side buy --notional-usdt 1000 --max-cost-bps 30

The check prints one JSON object: the API response plus http_status. Valid inputs return a report with decision, reason_codes, market_evidence, execution_cost, policy_checks, policy_version, and evidence_hash.

Auth

• Provider id: market-evidence-check
• Collection: none. Never ask for an API key, exchange credential, account login, wallet, password, seed phrase, or private key. If any step appears to require credentials, stop and report that as a blocker.
• Required scopes: none
• Allowed hosts: devday.167-179-82-72.nip.io
• Status check: bin/market_evidence_check.py health (expect HTTP 200 and status: "ok")
• Public API source: OpenAPI at https://raw.githubusercontent.com/yijigao/market-evidence-check-devday/main/connectors/muse/openapi.json (checked 2026-10-06)

Operating Rules

1. Read-only only: this connector reads public OKX order-book data. It never places, modifies, or cancels an order; never accesses an account or wallet; never signs anything.
2. Fail closed: if the API returns WATCH, REJECT, HTTP 400, HTTP 502, invalid JSON, or a timeout, report it as shown. Never convert a failure into GO and never suggest placing an order.
3. HTTP 200 alone is not GO; always inspect decision and report_status.
4. Separate observed snapshot values from caller-supplied cost assumptions (taker_fee_bps, slippage_buffer_bps; policy defaults 10 bps and 2 bps). Label assumptions as assumptions, never as the user's actual account fees.
5. Do not change the user's stated max_cost_bps limit without telling them first.
6. This is market evidence, not financial advice.

Files

• SKILL.md
• bin/market_evidence_check.py

Maturity

🧪 Draft: On 2026-10-08 an external user installed a reusable read-only skill from our public muse.md and openapi.json and passed three smoke checks (health HTTP 200; BTC-USDT sample GO; invalid input REJECT). This covers the brief/OpenAPI installation route only. The catalog one-paste install path and the CLI in this PR remain externally unverified.