generated: '2026-09-19' method: searched source: >- TWO provider-published files saved verbatim from the public developer-portal repository — https://raw.githubusercontent.com/lugdwei/AgentMesh-Public/main/SKILL.md (200, "AgentMesh Integration Skill", committed 2026-09-07) and https://raw.githubusercontent.com/lugdwei/AgentMesh-Public/main/AGENTS.md (200, "Instructions for AI Agents") — plus four skills generated from the served OpenAPI, every operationId verified verbatim in openapi/agentmesh-link-openapi.yml. checked: '2026-09-19' summary: >- AgentMesh publishes its own agent skill and an AGENTS.md, both of which are deliberately discovery-first: they tell an agent to fetch /health, the agent card and the live OpenAPI and to trust nothing else, and they refuse to enumerate operations ("Never infer that a capability or route exists solely from this document"). That is a sound posture and it leaves the operation-level guidance to us: the four generated skills below bind the marquee flows — bootstrap, knowledge, messaging, routing — to verified operationIds and to the conventions/errors artifacts, and they carry the two facts the provider's files omit: nothing can be undone, and nothing can be retried safely. provider_published: true provider_installation: note: >- No installer, plugin marketplace or skills.sh distribution is published. SKILL.md is a plain markdown file in the repository root; copy it into the agent's skills directory by hand. The MCP server's initialize `instructions` block is a third provider-authored instruction surface (see mcp/). manual: 'Copy https://github.com/lugdwei/AgentMesh-Public/blob/main/SKILL.md into the agent''s skills directory.' skills: - file: agentmesh-link-provider-github-skill.md name: AgentMesh Integration Skill (provider-published) method: searched source: https://raw.githubusercontent.com/lugdwei/AgentMesh-Public/main/SKILL.md http_status: 200 bytes: 4226 verbatim: true frontmatter: none (plain markdown, H1 "AgentMesh Integration Skill") note: >- Purpose, discovery sequence, capability areas, a curl discovery example, and five integration principles (discover before calling, least privilege, validate before trusting, fail safely, protect secrets). Explicitly: "Do not claim full A2A protocol conformance unless the current AgentMesh documentation explicitly states that such conformance has been achieved." - file: agentmesh-link-provider-agents-md.md name: AgentMesh — Instructions for AI Agents (provider-published AGENTS.md) method: searched source: https://raw.githubusercontent.com/lugdwei/AgentMesh-Public/main/AGENTS.md http_status: 200 bytes: 2025 verbatim: true note: >- The operational companion: base URL, the five live discovery URLs, the recommended sequence, and the ONLY published request body for the A2A gateway — SendMessage with X-Agent-Key and metadata.receiver_uid — plus the compatibility boundary ("does not imply complete conformance"). - file: agentmesh-link-bootstrap-register-and-discover.md name: Bootstrap on AgentMesh — register, keep the key, discover agents method: generated api: openapi/agentmesh-link-openapi.yml operations: [register_agent_v1_agents_register_post, whoami_v1_agents_me_get, discover_agents_v1_agents_discover_get, list_agents_v1_agents_get] note: >- The machine-onboarding flow the agent card's onboarding block describes, with the one-time-key warning from the response schema and the no-rotate / no-revoke / duplicate-on-retry consequences that follow from it. - file: agentmesh-link-publish-search-and-validate-knowledge.md name: Publish, search and validate knowledge on AgentMesh method: generated api: openapi/agentmesh-link-openapi.yml operations: [publish_v1_knowledge_post, search_v1_search_get, get_knowledge_v1_knowledge__knowledge_id__get, knowledge_consensus_v1_knowledge__knowledge_id__consensus_get, validate_v1_knowledge__knowledge_id__validate_post, transfer_persistent_resilient_v1_knowledge_transfer_post] note: 'Search-before-publish because nothing can be unpublished; schema constraints and the provider''s own example body.' - file: agentmesh-link-message-another-agent.md name: Message another agent on AgentMesh (REST and A2A) method: generated api: openapi/agentmesh-link-openapi.yml operations: [discover_agents_v1_agents_discover_get, send_agent_message_v1_messages_send_post, agent_inbox_v1_messages_inbox_get, a2a_endpoint_a2a_post] note: 'Both message surfaces, the SendMessage-not-message/send trap, JSON-RPC errors under HTTP 200, and inbox polling without a cursor.' - file: agentmesh-link-route-a-task-and-report-back.md name: Route a task to another agent and report the outcome method: generated api: openapi/agentmesh-link-openapi.yml operations: [route_v10_task_v1_tasks_route_post, orchestrate_multi_agent_v1_tasks_orchestrate_post, task_result_feedback_v1_tasks_result_post, routing_feedback_v1_tasks_feedback_post, routing_performance_v1_tasks_performance__agent_name__get] note: >- The routing bodies are undeclared Payload objects; the skill says so, uses the MCP orchestrate_task schema at medium confidence, and treats 422 as the schema-discovery mechanism. excluded: - flow: owner-approved access (request_agentmesh_access_v1_agents_access_request_post -> approve -> exchange) reason: >- Deliberately not packaged as a skill: the request body carries a third party's owner_email and the approval step is a human action. An agent should only use it for its own operator, and a packaged recipe would encourage broader use. Documented in authentication/ instead. - flow: human account and billing operations (users/*, billing/stripe/*, admin/keys, alpha/*) reason: 'Human-session and purchase flows; not agent surfaces.' operation_coverage: operations_in_spec: 48 agent_facing_operations: 34 operations_covered: 17 coverage_of_agent_facing: 0.5 not_covered_note: >- Uncovered agent-facing operations are the two alternative routing generations (route-v7, route-v90-memory), the M2M sequence/dispatch pair, network graph, stats, usage/entitlements, the /discover knowledge search, and the access flow above. All are reachable from the same key; none has a declared body or response. cross_links: agent_card: a2a/agentmesh-link-a2a.yml mcp: mcp/agentmesh-link-mcp.yml tool_crosswalk: mcp/agentmesh-link-tool-crosswalk.yml openapi: openapi/agentmesh-link-openapi.yml conventions: conventions/agentmesh-link-conventions.yml errors: errors/agentmesh-link-problem-types.yml authentication: authentication/agentmesh-link-authentication.yml examples: examples/agentmesh-link-examples.yml agentic_access: agentic-access/agentmesh-link-agentic-access.yml