Agent Requirements
Safe Agent Flow
1
Name the market job
The prompt should include venue family, symbol, data family, time window, and output goal whenever possible.
2
Cite the route before calling it
Ask the agent to name the REST route, WebSocket channel, or generated OpenAPI operation before execution.
3
Run one small probe
Use one symbol and one small window to confirm auth, response shape, freshness, and request-ID logging.
4
Widen only after gates pass
Add pagination, batching, backoff, and data-quality decisions before broad loops or exports.
Agent Execution Checklist
Before an agent calls a route, generated client, CLI command, MCP tool, or Skill recipe, require a checklist with the fields below.
If the checklist is incomplete, keep the agent in route-selection mode. Do not let it guess a venue family, broaden a symbol list, or switch from docs context to live execution without the missing fields.