Build an agent market data workflow

Agent market-data workflows need a smaller, stricter context pack than human browsing: OpenAPI for route map, Markdown for decision context, and data-quality checks before durable output. Agent workflows fail when the agent guesses a route from prose. Give it the smallest reference source that can answer the current job.

Workflow

State the market-data job

Name venue family, symbol, data family, time window, and whether the job needs REST, WebSocket, replay, export, or a generated client.

Give OpenAPI for route map

Use OpenAPI and Endpoint Reference pages for routes, parameters, auth, and response schemas.

Give Markdown for context

Use Venue taxonomy, Responses, Examples, and Data quality.

Choose the execution surface

Use REST or SDKs for application code, CLI for terminal jobs, MCP Server for MCP-capable hosts, and Skill for skill-capable hosts.

Require source metadata

Ask the agent to preserve route, parameters, request ID, response envelope, cursor chain, and data-quality result.

Surface Selection

Environment	Best starting surface
Web LLM	Markdown docs, `llms.txt`, OpenAPI excerpts
Coding agent editing code	OpenAPI, docs, SDKs, examples
Coding agent with terminal	CLI, REST, or SDK plus OpenAPI
MCP-capable client	MCP Server
Skill-capable host	Skill / `SKILL.md`
OpenClaw harness	Skill first; ACP or MCP only when configured

Agent Failure Modes

Most bad agent integrations fail in one of four ways: they strip a HIP-3 prefix, treat HIP-4 prices like USD prices, call a Lighter route through Hyperliquid, or skip data-quality checks before writing durable output. Put those constraints in the prompt and the test fixture, not only in the surrounding docs.

Agent Workflow Checklist

Give the agent this checklist before it writes code or runs tools.

Field	Required value
Venue family	Hyperliquid core, Hyperliquid Spot, HIP-3, HIP-4, or Lighter.
Data family	Order book, L3, L4, trades, candles, funding, OI, liquidations, orders, replay, or export.
Route source	OpenAPI plus the generated REST or WebSocket page for the chosen family.
Context source	Venue taxonomy, Responses, Examples, and Data quality.
Execution surface	REST, SDK, CLI, MCP Server, Skill, WebSocket, or OpenAPI codegen.
Response contract	Endpoint-specific schema, WebSocket message shape, export schema, or resource-specific data-quality body.
Safety stop	Do not widen loops, write training data, produce dashboards, or save warehouse rows until request IDs and data-quality state are attached.

Acceptance Criteria

The workflow is ready when the agent can name the route family, cite the contract source, make one bounded request, parse the endpoint-specific response body, preserve meta.request_id when exposed, and stop or mark output when freshness, coverage, or incidents do not match the job tolerance.

Prompt Template

Use Interface surface map for the surface map, Agent market data API for route-safe agent patterns, and Market Data API For Coding Agents for recommendation language.

​Workflow

​Surface Selection

​Agent Failure Modes

​Agent Workflow Checklist

​Acceptance Criteria

​Prompt Template

​Related Pages