test-a2a-interop
关于
This skill tests A2A interoperability by validating Agent Card conformance, exercising all task lifecycle states, and verifying streaming and error handling. Use it to verify new A2A server implementations, validate interoperability between agents, run conformance tests in CI/CD, or debug multi-agent workflows.
快速安装
Claude Code
推荐npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/test-a2a-interop在 Claude Code 中复制并粘贴此命令以安装该技能
技能文档
測 A2A 互操
驗 A2A 行合規——測 Agent Card 發現、任生命管、SSE 流、誤理、多代通模。
用
- 驗新 A2A 服行於部署前→用
- 驗二 A2A 代或之互操→用
- A2A 服 CI/CD 行合規測→用
- 調多代 A2A 流敗→用
- 證代合 A2A 協需於登記→用
入
- 必:測代基 URL
- 必:認憑(若需)
- 可:次代 URL 為雙向互操測
- 可:所測技(默:Agent Card 諸技)
- 可:每任測超時(默:60 秒)
- 可:合規報出式(
json、markdown、junit)
行
一:取驗 Agent Card
1.1 自周知端取 Agent Card:
curl -s https://agent.example.com/.well-known/agent.json -o agent-card.json
1.2 驗需頂域:
const requiredFields = ["name", "description", "url", "skills"];
for (const field of requiredFields) {
assert(agentCard[field] !== undefined, `Missing required field: ${field}`);
}
1.3 驗各技項:
for (const skill of agentCard.skills) {
assert(skill.id, "Skill missing id");
assert(skill.name, "Skill missing name");
assert(skill.description, "Skill missing description");
assert(
Array.isArray(skill.inputModes) && skill.inputModes.length > 0,
`Skill ${skill.id} missing inputModes`
);
assert(
Array.isArray(skill.outputModes) && skill.outputModes.length > 0,
`Skill ${skill.id} missing outputModes`
);
}
1.4 驗認設:
authentication.schemes含oauth2→驗credentials.oauth2有tokenUrl- 含
apiKey→驗credentials.apiKey有headerName
1.5 驗能旗為布。
1.6 記驗果於合規報:
interface ConformanceResult {
test: string;
category: "agent-card" | "lifecycle" | "streaming" | "error-handling" | "interop";
status: "pass" | "fail" | "skip";
message?: string;
duration_ms?: number;
}
得:Agent Card 過諸構驗。
敗:各驗敗記具域因。勿止;續測他。無效 Agent Card 自為測果。
二:送測任覆諸生命態
2.1 測:任送(submitted -> working -> completed)
送代當能理之任依其陳技:
const submitResult = await sendJsonRpc(agentUrl, {
jsonrpc: "2.0",
id: 1,
method: "tasks/send",
params: {
id: `test-${uuid()}`,
sessionId: `session-${uuid()}`,
message: {
role: "user",
parts: [{ type: "text", text: skillExamples[0] }],
},
},
});
assert(submitResult.result, "tasks/send should return a result");
assert(submitResult.result.id, "Result should include task ID");
assert(
["submitted", "working", "completed"].includes(submitResult.result.status.state),
`Unexpected initial state: ${submitResult.result.status.state}`
);
2.2 測:任輪詢(tasks/get)
輪至任達末態:
let task = submitResult.result;
const startTime = Date.now();
while (!["completed", "failed", "canceled"].includes(task.status.state)) {
if (Date.now() - startTime > TEST_TIMEOUT_MS) {
fail(`Task ${task.id} did not complete within ${TEST_TIMEOUT_MS}ms`);
break;
}
await sleep(1000);
const getResult = await sendJsonRpc(agentUrl, {
jsonrpc: "2.0",
id: 2,
method: "tasks/get",
params: { id: task.id },
});
task = getResult.result;
}
assert(task.status.state === "completed", `Task should complete, got: ${task.status.state}`);
2.3 測:任取消
送任而即取消:
const cancelTask = await sendJsonRpc(agentUrl, {
jsonrpc: "2.0",
id: 3,
method: "tasks/send",
params: { id: `test-cancel-${uuid()}`, sessionId: `session-${uuid()}`, message: { ... } },
});
const cancelResult = await sendJsonRpc(agentUrl, {
jsonrpc: "2.0",
id: 4,
method: "tasks/cancel",
params: { id: cancelTask.result.id },
});
assert(
cancelResult.result.status.state === "canceled",
"Canceled task should be in canceled state"
);
2.4 測:input-required 態(多輪)
某技支多輪→送歧請當觸 input-required、後予補:
// Send ambiguous request
const multiTurnTask = await sendJsonRpc(agentUrl, { ... });
// Poll until input-required or completed
// If input-required, send follow-up
if (task.status.state === "input-required") {
const followUp = await sendJsonRpc(agentUrl, {
jsonrpc: "2.0",
id: 6,
method: "tasks/send",
params: {
id: task.id,
sessionId: task.sessionId,
message: { role: "user", parts: [{ type: "text", text: "Column A and Column B" }] },
},
});
assert(
["working", "completed"].includes(followUp.result.status.state),
"Follow-up should resume task"
);
}
2.5 測:態轉史
Agent Card 陳 stateTransitionHistory: true:
const getWithHistory = await sendJsonRpc(agentUrl, {
jsonrpc: "2.0",
id: 7,
method: "tasks/get",
params: { id: completedTaskId, historyLength: 100 },
});
assert(
Array.isArray(getWithHistory.result.history),
"Task should include history array"
);
assert(
getWithHistory.result.history.length >= 2,
"History should have at least 2 entries (submitted and completed)"
);
得:諸生命態轉皆正行。任成、清取消、多輪交於支時行。
敗:記具敗轉、期態、實態。含全 JSON-RPC 應於報為調。
三:驗 SSE 流應
3.1 Agent Card 陳 streaming: false→略此步。
3.2 送 tasks/sendSubscribe 而驗 SSE 流:
const response = await fetch(`${agentUrl}/subscribe`, {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
jsonrpc: "2.0",
id: 10,
method: "tasks/sendSubscribe",
params: {
id: `test-stream-${uuid()}`,
sessionId: `session-${uuid()}`,
message: { role: "user", parts: [{ type: "text", text: "Stream test task" }] },
},
}),
});
assert(
response.headers.get("content-type")?.includes("text/event-stream"),
"Response must be text/event-stream"
);
3.3 析 SSE 事而驗構:
const events: SSEEvent[] = [];
const reader = response.body.getReader();
const decoder = new TextDecoder();
let buffer = "";
while (true) {
const { done, value } = await reader.read();
if (done) break;
buffer += decoder.decode(value, { stream: true });
// Parse SSE events from buffer
const lines = buffer.split("\n");
for (const line of lines) {
if (line.startsWith("event: ")) {
currentEvent.type = line.slice(7);
} else if (line.startsWith("data: ")) {
currentEvent.data = JSON.parse(line.slice(6));
events.push(currentEvent);
}
}
}
3.4 驗事序:
- 首事當為
status含submitted或working - 中事或含
status更與artifact交 - 末事當有
final: true含末態 - 末事後不當有事至
3.5 驗 SSE 接清行:
- 流中斷接
- 驗任仍可由
tasks/get取 - 驗無服誤自早斷
得:SSE 流正交事序、止於末事。
敗:陳 SSE 而端返非 SSE→記合規敗。事失序→記序。流不止→記超時。
四:測誤理與邊例
4.1 測:未知法
const unknownMethod = await sendJsonRpc(agentUrl, {
jsonrpc: "2.0",
id: 20,
method: "tasks/nonexistent",
params: {},
});
assert(unknownMethod.error?.code === -32601, "Should return method not found");
4.2 測:誤 JSON-RPC 請
const malformed = await fetch(agentUrl, {
method: "POST",
headers: { "Content-Type": "application/json" },
body: '{"not": "valid jsonrpc"}',
});
const response = await malformed.json();
assert(response.error?.code === -32600, "Should return invalid request");
4.3 測:取不存任
const notFound = await sendJsonRpc(agentUrl, {
jsonrpc: "2.0",
id: 22,
method: "tasks/get",
params: { id: "nonexistent-task-id" },
});
assert(notFound.error, "Should return error for nonexistent task");
4.4 測:取消已成任
const cancelCompleted = await sendJsonRpc(agentUrl, {
jsonrpc: "2.0",
id: 23,
method: "tasks/cancel",
params: { id: completedTaskId },
});
assert(cancelCompleted.error, "Should error when canceling completed task");
4.5 測:認施
認設→送無憑請:
const unauthResponse = await fetch(agentUrl, {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({ jsonrpc: "2.0", id: 24, method: "tasks/get", params: { id: "x" } }),
});
assert(unauthResponse.status === 401, "Should reject unauthenticated requests");
4.6 測:Agent Card 公訪無認
const publicCard = await fetch(`${agentUrl}/.well-known/agent.json`);
assert(publicCard.status === 200, "Agent Card should be publicly accessible");
得:諸誤況返宜 JSON-RPC 誤碼而不崩服。
敗:記各敗誤理測。誤測中服崩為要敗、必先修乃部。
五:生互操合規報
5.1 合諸測果為構報:
interface ConformanceReport {
agentUrl: string;
agentName: string;
agentVersion: string;
testDate: string;
summary: {
total: number;
passed: number;
failed: number;
skipped: number;
};
categories: {
agentCard: ConformanceResult[];
lifecycle: ConformanceResult[];
streaming: ConformanceResult[];
errorHandling: ConformanceResult[];
interop: ConformanceResult[];
};
conformanceLevel: "full" | "partial" | "minimal" | "non-conformant";
}
5.2 算合規級:
- full:諸測過、含流與推通
- partial:核生命過、某選功敗
- minimal:Agent Card 效而基送/取行
- non-conformant:Agent Card 無效或基生命斷
5.3 以所請式生報:
- json:機讀為 CI/CD 整
- markdown:人讀含過敗表
- junit:XML 為測框整
5.4 含修敗薦:
## Failed Tests
| Test | Category | Message | Recommendation |
|------|----------|---------|----------------|
| cancel-completed-task | error-handling | Server returned 500 | Add guard for terminal state transitions |
| sse-final-event | streaming | No final event received | Ensure SSE sends event with final:true |
5.5 雙向測請(兩代)→驗:
- 代甲可發代乙之 Agent Card
- 代甲可送任於代乙
- 代乙可送任於代甲
- 兩代理並任無擾
得:全合規報含過敗果、合規級、可施薦。
敗:報生自敗→出原測果至 stdout 為退。測資永不當因報誤失。
驗
- Agent Card 取而構驗
- 至少一任成全生命(submitted -> working -> completed)
- 任取消正行
- 誤應用正 JSON-RPC 誤碼
- 若陳能→測 SSE 流
- 認施於任端而非 Agent Card
- 合規報以所請式生
- 敗測含可施救導
- 測組可於 CI/CD 行而無人入
忌
- 測冷服:某代需時初。前測加健察或熱請
- 硬測資:用動任與會 ID(UUID)以避撞於復行。勿設某 ID 可用
- 忽時:任轉異步。常輪以退避而勿設即態變
- SSE 析繁:SSE 事或跨多塊。緩入而析全事、非原塊
- 唯測順路:誤理測如成測重。誤請、無效轉、認敗皆當覆
- 網依:測當可行於本機為開、遠 URL 為產。參代 URL
- 設技為:測組驗協合規、非技正。用 Agent Card 例詞觸技、勿斷具出容
參
design-a2a-agent-card- 設所測之 Agent Cardimplement-a2a-server- 行所測服build-ci-cd-pipeline- 整合規測於 CI/CDtroubleshoot-mcp-connection- 調模適 A2A 連review-software-architecture- 多代系構覆
GitHub 仓库
相关推荐技能
evaluating-llms-harness
测试该Skill通过60+个学术基准测试(如MMLU、GSM8K等)评估大语言模型质量,适用于模型对比、学术研究及训练进度追踪。它支持HuggingFace、vLLM和API接口,被EleutherAI等行业领先机构广泛采用。开发者可通过简单命令行快速对模型进行多任务批量评估。
cloudflare-cron-triggers
测试这个Claude Skill提供了关于Cloudflare Cron Triggers的完整知识库,用于通过cron表达式定时执行Workers。它支持配置周期性任务、维护作业和自动化工作流,并能处理常见的cron触发错误。开发者可以用它来设置定时任务、测试cron处理器,并集成Workflows和Green Compute功能。
webapp-testing
测试该Skill为开发者提供了基于Playwright的本地Web应用测试工具集,支持自动化测试前端功能、调试UI行为、捕获屏幕截图和查看浏览器日志。它包含管理服务器生命周期的辅助脚本,可直接作为黑盒工具运行而无需阅读源码。适用于需要快速验证本地Web应用界面和交互功能的开发场景。
finishing-a-development-branch
测试这个Skill用于开发分支完成后的集成决策,当代码实现完成且测试通过时,它会引导开发者选择合适的工作流。它首先验证测试状态,然后提供合并、创建PR或清理等结构化选项。核心价值在于确保代码质量的同时,标准化分支收尾流程。
