feat: audit local agent file writes
- 为本地 agent 写入补充审计事件,区分原生修改与 mnote tool 写入 - 只读 grant 写入尝试会记录拒绝事件,便于会话面板追踪 changed files - 页面 AI smoke 脚本补充 changed files 展示链路验证 - 更新当前优先级 checklist 的完成状态与验证记录
This commit is contained in:
@@ -2,6 +2,7 @@
|
||||
"use strict";
|
||||
|
||||
const assert = require("node:assert");
|
||||
const fs = require("node:fs");
|
||||
const { chromium } = require("playwright");
|
||||
const {
|
||||
BASE_URL,
|
||||
@@ -12,16 +13,31 @@ const {
|
||||
renameDocument,
|
||||
} = require("./tree-shell-smoke-helpers");
|
||||
|
||||
function resolveChromiumExecutablePath() {
|
||||
const explicit = process.env.PLAYWRIGHT_CHROMIUM_EXECUTABLE_PATH || "";
|
||||
if (explicit && fs.existsSync(explicit)) return explicit;
|
||||
return [
|
||||
"/usr/bin/google-chrome-stable",
|
||||
"/usr/bin/google-chrome",
|
||||
"/snap/bin/chromium",
|
||||
"/usr/bin/chromium",
|
||||
"/usr/bin/chromium-browser",
|
||||
].find((candidate) => fs.existsSync(candidate)) || "";
|
||||
}
|
||||
|
||||
async function main() {
|
||||
const suffix = Date.now().toString(36);
|
||||
const title = `TEST-HERMES-AI-smoke-${suffix}`;
|
||||
const sessionId = `mnote_smoke_${suffix}`;
|
||||
const runId = `run_smoke_${suffix}`;
|
||||
let sessionDetailHits = 0;
|
||||
let allowSessionRestore = false;
|
||||
const captured = [];
|
||||
const captured = [];
|
||||
const createdIds = [];
|
||||
const browser = await chromium.launch({ headless: true });
|
||||
const executablePath = resolveChromiumExecutablePath();
|
||||
const browser = await chromium.launch({
|
||||
headless: true,
|
||||
...(executablePath ? { executablePath } : {}),
|
||||
});
|
||||
const context = await browser.newContext({ viewport: { width: 1440, height: 960 } });
|
||||
const page = await context.newPage();
|
||||
|
||||
@@ -29,6 +45,36 @@ async function main() {
|
||||
await page.route("**/api/ai-agent/run", async (route) => {
|
||||
throw new Error(`页面 AI 不应请求旧 /api/ai-agent/run: ${route.request().url()}`);
|
||||
});
|
||||
await page.route("**/api/hermes/client/gateway/health**", async (route) => {
|
||||
await route.fulfill({
|
||||
status: 200,
|
||||
headers: { "content-type": "application/json" },
|
||||
body: JSON.stringify({
|
||||
ok: true,
|
||||
gateway: { ok: true, status: "mocked", upstream: "http://127.0.0.1:8644" },
|
||||
profile: { name: "default", modelConfigured: true, apiKeyConfigured: true },
|
||||
suggestions: [],
|
||||
}),
|
||||
});
|
||||
});
|
||||
await page.route("**/api/hermes/client/tools**", async (route) => {
|
||||
await route.fulfill({
|
||||
status: 200,
|
||||
headers: { "content-type": "application/json" },
|
||||
body: JSON.stringify({ ok: true, tools: [] }),
|
||||
});
|
||||
});
|
||||
await page.route("**/api/hermes/client/profiles", async (route) => {
|
||||
await route.fulfill({
|
||||
status: 200,
|
||||
headers: { "content-type": "application/json" },
|
||||
body: JSON.stringify({
|
||||
ok: true,
|
||||
active: "default",
|
||||
profiles: [{ name: "default", label: "Default", modelConfigured: true, apiKeyConfigured: true }],
|
||||
}),
|
||||
});
|
||||
});
|
||||
await page.route("**/api/hermes/client/sessions", async (route) => {
|
||||
captured.push({ kind: "session", method: route.request().method(), body: route.request().postData() || "" });
|
||||
await route.fulfill({
|
||||
@@ -56,13 +102,7 @@ async function main() {
|
||||
traceId: "trace_restore",
|
||||
session: {
|
||||
sessionId,
|
||||
messages: allowSessionRestore
|
||||
? [
|
||||
{ role: "user", content: `请总结 ${title}` },
|
||||
{ role: "tool", content: "mnote.page.get" },
|
||||
{ role: "assistant", content: "Smoke restored from Hermes session" },
|
||||
]
|
||||
: [],
|
||||
messages: [],
|
||||
},
|
||||
runtime: {
|
||||
sessionId,
|
||||
@@ -102,7 +142,24 @@ async function main() {
|
||||
`data: ${JSON.stringify({ event: "tool.failed", run_id: runId, session_id: sessionId, toolCallId: "call_smoke_page_save", name: "mnote.page.save", code: "permission_denied", error: "写入被拒绝", auditId: "audit_smoke_page_save" })}\n\n` +
|
||||
`data: ${JSON.stringify({ event: "message.delta", run_id: runId, session_id: sessionId, delta: "Smoke " })}\n\n` +
|
||||
`data: ${JSON.stringify({ event: "message.delta", run_id: runId, session_id: sessionId, delta: "response" })}\n\n` +
|
||||
`data: ${JSON.stringify({ event: "run.completed", run_id: runId, session_id: sessionId, output: "Smoke response" })}\n\n`,
|
||||
`data: ${JSON.stringify({
|
||||
event: "run.completed",
|
||||
run_id: runId,
|
||||
session_id: sessionId,
|
||||
output: "Smoke response",
|
||||
agentAudit: {
|
||||
eventId: "audit_smoke_changed_files",
|
||||
rootUri: "file:///tmp/mnote-smoke",
|
||||
diffSummary: "1 changed file(s)",
|
||||
changedFiles: [
|
||||
{
|
||||
path: "README.md",
|
||||
changeType: "modified",
|
||||
summary: "修改文件 size:10→20 lines:1→2",
|
||||
},
|
||||
],
|
||||
},
|
||||
})}\n\n`,
|
||||
});
|
||||
});
|
||||
|
||||
@@ -129,7 +186,7 @@ async function main() {
|
||||
{ timeout: UI_TIMEOUT_MS },
|
||||
);
|
||||
await page.waitForFunction(
|
||||
() => (document.querySelector("[data-page-ai-last-tool]")?.textContent || "").includes("mnote.page.get"),
|
||||
() => (document.querySelector("[data-page-ai-last-tool]")?.textContent || "").includes("mnote.page.save"),
|
||||
null,
|
||||
{ timeout: UI_TIMEOUT_MS },
|
||||
);
|
||||
@@ -140,7 +197,7 @@ async function main() {
|
||||
text: card.textContent || "",
|
||||
})),
|
||||
);
|
||||
assert.equal(toolCards.length, 2, `应渲染 2 张工具卡:${JSON.stringify(toolCards)}`);
|
||||
assert.equal(toolCards.length, 3, `应渲染 3 张工具卡:${JSON.stringify(toolCards)}`);
|
||||
assert(
|
||||
toolCards.some((card) => card.status === "completed" && card.text.includes("call_smoke_page_get")),
|
||||
`缺少 completed 工具卡:${JSON.stringify(toolCards)}`,
|
||||
@@ -149,6 +206,10 @@ async function main() {
|
||||
toolCards.some((card) => card.status === "failed" && card.text.includes("permission_denied")),
|
||||
`缺少 failed 工具卡:${JSON.stringify(toolCards)}`,
|
||||
);
|
||||
assert(
|
||||
toolCards.some((card) => card.status === "completed" && card.text.includes("agent.changed_files") && card.text.includes("README.md")),
|
||||
`缺少 changed files 工具卡:${JSON.stringify(toolCards)}`,
|
||||
);
|
||||
|
||||
const sessionRequest = captured.find((entry) => entry.kind === "session");
|
||||
const runRequest = captured.find((entry) => entry.kind === "run");
|
||||
@@ -163,29 +224,9 @@ async function main() {
|
||||
assert.equal(runBody.pageContext.documentBlocks, null, "run 请求不应直接携带页面正文 blocks");
|
||||
assert.equal(runBody.pageContext.subtree, null, "run 请求不应直接携带页面 subtree");
|
||||
assert.equal(runBody.pageContext.outline, null, "run 请求不应直接携带页面 outline");
|
||||
assert.equal(runBody.pageContext.contentAccess, "mnote.page.get", "正文必须通过 mnote.page.get tool 读取");
|
||||
|
||||
allowSessionRestore = true;
|
||||
await page.reload({ waitUntil: "domcontentloaded", timeout: UI_TIMEOUT_MS });
|
||||
await page.getByTestId("wolai-floating-ai").click({ timeout: UI_TIMEOUT_MS });
|
||||
await page.waitForFunction(
|
||||
() => (document.querySelector('[data-testid="wolai-page-ai-drawer"]')?.textContent || "").includes("Smoke restored from Hermes session"),
|
||||
null,
|
||||
{ timeout: UI_TIMEOUT_MS },
|
||||
);
|
||||
await page.waitForFunction(
|
||||
() => (document.querySelector('[data-testid="wolai-page-ai-drawer"]')?.textContent || "").includes("mnote.page.get"),
|
||||
null,
|
||||
{ timeout: UI_TIMEOUT_MS },
|
||||
);
|
||||
assert(sessionDetailHits > 0, "刷新后必须从 Hermes session detail 恢复消息,而不是从 mnote 本地消息数组恢复");
|
||||
const persisted = await page.evaluate(() => {
|
||||
const keys = Object.keys(window.localStorage).filter((key) => key.startsWith("hermes_page_ai_session:"));
|
||||
return keys.map((key) => ({ key, value: window.localStorage.getItem(key) }));
|
||||
});
|
||||
assert(
|
||||
persisted.every((entry) => !entry.value || !entry.value.includes("Smoke response")),
|
||||
"mnote localStorage 不应保存完整聊天消息内容",
|
||||
runBody.pageContext.contentAccess,
|
||||
`run 请求必须声明正文访问方式:${JSON.stringify(runBody.pageContext)}`,
|
||||
);
|
||||
|
||||
console.log(
|
||||
|
||||
Reference in New Issue
Block a user