菜单
Use this skill when the user asks about A2A (Agent-to-Agent) protocol communication, OASF record formats, AGNTCY directory operations, agent card parsing, or dirctl CLI usage. It provides comprehensive reference material for A2A integrations.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | a2a-protocol |
| description | Use this skill when the user asks about A2A (Agent-to-Agent) protocol communication, OASF record formats, AGNTCY directory operations, agent card parsing, or dirctl CLI usage. It provides comprehensive reference material for A2A integrations. |
This skill provides knowledge about the A2A (Agent-to-Agent) protocol, the AGNTCY directory, and the OASF (Open Agent Standard Format) — everything needed to discover, connect to, and communicate with remote agents.
The /recruit command discovers agents from the AGNTCY directory and connects them to Claude Code. Two creation modes are available:
Skills create a /slash-command that the parent model executes directly. This is the most reliable way to proxy requests because the parent model runs the a2a-send command itself — no intermediary model to refuse.
Workflow:
/recruit <query> — discover agents, pick which to create, choose "skill" mode/<agent-name> <message> — e.g., /brazil-coffee-farm What is the current yield?rm -rf .claude/skills/<agent-name>/Skills are available immediately after creation — no restart needed.
Sub-agents create a .claude/agents/<name>.md file. Claude Code spawns a separate model to handle requests. This is useful when you want the sub-agent to have its own model (haiku/sonnet/opus) and reasoning, but note that sub-agent models may refuse to forward requests they deem outside the agent's advertised capabilities.
Workflow:
/recruit <query> — discover agents, pick which to create, choose "subagent" moderm .claude/agents/coffee-farm.mdSub-agents require a restart to load — run /exit and relaunch claude.
A2A (Agent-to-Agent) is a JSON-RPC 2.0 protocol for inter-agent communication over HTTP. This plugin uses v0.3.0+ (message/send method). Always check the protocolVersion in an agent's card to confirm compatibility.
OASF records are the standard format for describing agents in the AGNTCY directory. Each record contains metadata (name, description, skills, domains) and modules — including the integration/a2a module that holds the Agent Card.
A directory where agents register their OASF records. The dirctl CLI is the primary interface:
dirctl search — find agents by skill, name, domain, or module → returns CIDsdirctl pull <CID> — fetch the full OASF record for a CIDThe directory server address is configured via DIRECTORY_CLIENT_SERVER_ADDRESS (default 0.0.0.0:8888).
The A2A Agent Card (found inside the integration/a2a module of an OASF record) describes an agent's communication interface: endpoint URL, supported protocol version, skills, capabilities, and transport options. It can also be fetched directly at <endpoint>/.well-known/agent.json.
A2A uses JSON-RPC 2.0 over HTTP POST with the message/send method (v0.3.0+). Key format details:
messageId field (UUID)"kind": "text"contextId in configurationresult.kind: "message" with result.parts[]curl -s <ENDPOINT_URL>/.well-known/agent.json
Check protocolVersion — must be "0.3.0" or higher.
For detailed information, see these reference files in the references/ directory:
| File | Contents |
|---|---|
oasf-structure.md | OASF record fields, Agent Card fields, HTTP endpoint resolution |
a2a-protocol-cheatsheet.md | Protocol version detection, method tables, curl templates, response formats |
error-handling.md | Timeouts, retry policy, structured error format |
dirctl-search.md | dirctl search and pull command examples |
Authors and maintains the human-facing Agentic Workflows API documentation for the lungo subproject at coffeeAGNTCY/coffee_agents/lungo/docs/workflow-instance_api.md, covering the catalog API, workflow-instance lifecycle/state, the internal event ingress, the SSE and NDJSON streaming formats, the canonical event_v1 JSON Schema, and the topology/use-case response shapes a frontend consumes. Use when documenting or updating the lungo workflow API; when an OpenAPI document under coffeeAGNTCY/coffee_agents/lungo/schema/openapi, the JSON Schema under coffeeAGNTCY/coffee_agents/lungo/schema/jsonschemas, or the router under coffeeAGNTCY/coffee_agents/lungo/api/agentic_workflows changes; or when a prompt references a ticket/issue, pull request, or file/folder that changes the lungo API contracts or specifications. DO NOT TRIGGER AUTOMATICALLY. ASK THE USER IF THE SKILL SHOULD BE USED.
Generates coffeeAgntcy CHANGELOG release notes and README Built With updates from PRs and dependency lockfiles since the previous release tag. Use when the user asks for release notes, a changelog entry, GitHub release text, or preparing a semver release.
Generates and validates Python FastAPI routers and DTOs from the OpenAPI schemas in coffeeAGNTCY/coffee_agents/lungo/schema/openapi for the lungo subproject. Use when adding, removing, renaming, or modifying endpoints in any lungo OpenAPI document; when generating or regenerating files under coffeeAGNTCY/coffee_agents/lungo/api/<tag>/ (router.py, dtos.py); when validating that hand-edited handlers or DTOs still match the OpenAPI spec; or when scaffolding the OpenAPI unit tests for a router.
Generates Pydantic v2 model classes from the JSON Schema documents under `coffeeAGNTCY/coffee_agents/lungo/schema/jsonschemas/` into `coffeeAGNTCY/coffee_agents/lungo/schema/types/*.py`, and a matching pytest module per types module under `coffeeAGNTCY/coffee_agents/lungo/tests/unit/schemas/types/test_<module>.py`. DO NOT TRIGGER AUTOMATICALLY. ASK THE USER IF THE SKILL SHOULD BE USED. Use when a `*_v*.json` schema under that folder is added or modified, when regenerating the lungo Pydantic types or their tests after a schema change, or when the user mentions regenerating schema types, JSON Schema → Pydantic, or fixing drift between schema and types.