{"name":"Zesia Consultant","description":"Independent expert advice from supplied context. Multi-turn messages share contextId; the caller resubmits the reviewed context snapshot each round. No server-side history. Does not execute work or retain conversation history.","supportedInterfaces":[{"url":"https://api.zesia.dev/a2a","protocolBinding":"JSONRPC","protocolVersion":"1.0"}],"version":"0.2.0","capabilities":{"streaming":false,"pushNotifications":false,"extendedAgentCard":false},"securitySchemes":{"bearer":{"httpAuthSecurityScheme":{"scheme":"bearer"}}},"securityRequirements":[{"schemes":{"bearer":{}}}],"defaultInputModes":["text/plain","application/vnd.zesia.consult+json"],"defaultOutputModes":["text/plain","application/vnd.zesia.consult+json"],"skills":[{"id":"consult","name":"Expert consultation","description":"Send a text question with message.metadata.zesia containing requester and project, or one structured data Part containing the Zesia Consult request (requester, project, brief, optional context). Context may contain goal, originalReply, facts, assumptions, attempts, userFeedback, constraints, attributed statements, previousAdvice and materials. Results disclose input attribution limits, actual expert matching, action feasibility checks and stage traces. Follow-up requests distinguish evidence, user decisions and clarification. Reuse the returned contextId on subsequent SendMessage calls with a complete reviewed snapshot. No tracked Tasks or file fetching. Advice only. GET /v1/consult/limits publishes current limits; POST /v1/consult/preflight validates the same input with zero model calls.","tags":["consult","second-opinion"],"examples":["Review a proposed recovery plan using the supplied facts and failed attempts."]}]}