fix(chv-muse-r12fix-4): detect_checkout 接进 CDP broker(P10)
R12 真机:detect_checkout 在 chv-guest/、host.py、 executor.py 里零引用,结账页无任何拦截。按 P10 gap §4.1 接线:broker 加载 cloud/chv/checkout_detect.py(同目录 优先,仓库路径兜底;缺失时降级关闭检测、写警告,绝不 crash-loop);snapshot() 每次重算 verdict(aria 行拼伪节点, 与 _form_risk 同形),open() 导航前后两道门——落点是结账 页时返回 form_submit ask 并暂停,上一页是结账页时下一次 导航先审批;ask 批过后同 URL 不再重复 ask(批准即放行, 与 form_risk 同口径);verdict 进 snapshot()/status() 回包。 build.sh:checkout_detect.py 条件拷入 /usr/local/bin/(0644, P3/P6 同款只警告口径,M4 不硬要求 cloud/), extra_contents 与两处 lockfile 登记补齐。钱包三腿不在本轮 范围,未动。
This commit is contained in:
1 parent
1da0d86159
commit
0ebcdfb2a7
7 files changed
+258
-8
No files matched your search
@@ -232,6 +232,16 @@ if [ -f "$REPO_DIR/cloud/chv/privsep.py" ] && [ -f "$REPO_DIR/cloud/chv/safety.p
|
||||
else
|
||||
echo "警告:Muse P6 云端叶子模块不在仓库里(cloud/chv/privsep.py 与 safety.py),镜像暂不含 worker 叶子模块" >&2
|
||||
fi
|
||||
# P10 云端检测模块(checkout_detect.py)只警告、不失败——
|
||||
# 与 P3/P6 同款口径(M4 既定 build.sh 不硬要求 cloud/);缺失时
|
||||
# broker 的结账检测降级关闭(写警告、不 crash-loop),但完整
|
||||
# 镜像必须含它(extra_contents 与 lockfile 均有登记)。
|
||||
P10_CHECKOUT_PRESENT=0
|
||||
if [ -f "$REPO_DIR/cloud/chv/checkout_detect.py" ]; then
|
||||
P10_CHECKOUT_PRESENT=1
|
||||
else
|
||||
echo "警告:Muse P10 结账检测模块不在仓库里(cloud/chv/checkout_detect.py),镜像暂不含 checkout_detect" >&2
|
||||
fi
|
||||
|
||||
if [ "$DRY_RUN" = "0" ]; then
|
||||
if [ ! -e /dev/kvm ]; then
|
||||
@@ -412,6 +422,15 @@ if [ "$P6_LIBS_PRESENT" = "1" ]; then
|
||||
--copy-in "$SCRIPT_DIR/authd_init_stub.py:/usr/local/lib/chv-gitea/cloud/chv/"
|
||||
--run-command 'mv /usr/local/lib/chv-gitea/cloud/chv/authd_init_stub.py /usr/local/lib/chv-gitea/cloud/chv/__init__.py')
|
||||
fi
|
||||
# P10 交接(Muse P10 的结账检测模块,round-12 修复轮项 4):
|
||||
# 文件在仓库里存在时拷入 /usr/local/bin/ 供 chv-cdp-broker 加载
|
||||
#(broker 先找同目录,再找仓库 cloud/chv/;0644 即可,broker
|
||||
# 以 chv-browser 身份读);不存在时只警告、不失败(与 P3/P6
|
||||
# 同款口径,broker 降级关闭检测、不 crash-loop)。
|
||||
if [ "$P10_CHECKOUT_PRESENT" = "1" ]; then
|
||||
VC_ARGS+=(--copy-in "$REPO_DIR/cloud/chv/checkout_detect.py:/usr/local/bin/"
|
||||
--run-command 'chmod 0644 /usr/local/bin/checkout_detect.py')
|
||||
fi
|
||||
# P1 交接(Muse P1 的 vsock_notify.py):文件在仓库里存在时拷入
|
||||
# 镜像并建 vsock-notify.service(自动启动,round-11 真机修复:
|
||||
# 此前只拷文件、不建服务,vsock 通道从未自动启动;服务形状按
|
||||
@@ -488,6 +507,7 @@ else
|
||||
EXTRA_CONTENTS=""
|
||||
[ "$P3_LIBS_PRESENT" = "1" ] && EXTRA_CONTENTS="cloud/chv/authd.py,cloud/chv/sentinel.py,cloud/chv/errors.py,cloud/chv/__init__.py"
|
||||
[ "$P6_LIBS_PRESENT" = "1" ] && EXTRA_CONTENTS="${EXTRA_CONTENTS:+$EXTRA_CONTENTS,}cloud/chv/privsep.py,cloud/chv/safety.py"
|
||||
[ "$P10_CHECKOUT_PRESENT" = "1" ] && EXTRA_CONTENTS="${EXTRA_CONTENTS:+$EXTRA_CONTENTS,}cloud/chv/checkout_detect.py"
|
||||
[ "$VSOCK_NOTIFY_PRESENT" = "1" ] && EXTRA_CONTENTS="${EXTRA_CONTENTS:+$EXTRA_CONTENTS,}vsock_notify"
|
||||
python3 "$SCRIPT_DIR/buildinfo.py" "$OUT" "$TAG" "$UBUNTU_VERSION" "$CODEX_VERSION" "$REPO_COMMIT" "$UBUNTU_IMAGE_SHA256" "${PLAYWRIGHT:-1}" "${CHROMIUM_HEADLESS_SHELL_REV:-}" "$EXTRA_CONTENTS"
|
||||
fi
|
||||
|
||||
@@ -54,6 +54,7 @@ DECLARED_SOURCES = {
|
||||
"chv-sync-worker-client": "repo_commit",
|
||||
"chv-gitea-worker.service": "repo_commit",
|
||||
"cloud/chv/privsep.py": "repo_commit",
|
||||
"cloud/chv/checkout_detect.py": "repo_commit",
|
||||
"cloud/chv/safety.py": "repo_commit",
|
||||
}
|
||||
|
||||
|
||||
+123
-6
@@ -8,7 +8,8 @@ socket 说话,绝不直连浏览器、绝不见 CDP 端点。
|
||||
|
||||
操作(Unix socket /run/chv/cdp-broker.sock,JSON 行):
|
||||
open {url} -> 导航并返回 {title, url, a11y, truncated,
|
||||
untrusted_input, injection_flags, form_risk}
|
||||
untrusted_input, injection_flags, form_risk,
|
||||
checkout}
|
||||
snapshot {} -> 当前页同上
|
||||
screenshot {} -> 截图落盘 /run/chv/shots/<uuid>.png,返回 {path}
|
||||
close {} -> 关当前页(浏览器进程继续跑)
|
||||
@@ -26,8 +27,11 @@ socket 说话,绝不直连浏览器、绝不见 CDP 端点。
|
||||
导航前 validate_url + 静态拦截名单;快照恒带
|
||||
untrusted_input=true(外部页面文本永远是数据、不是指令),
|
||||
命中注入启发式带 injection_flags;高风险表单在页时 open
|
||||
导航即返回 ask(入队、执行停止);观测到导航 POST 即自动
|
||||
暂停 + ask;下载命中可执行扩展名即取消 + 暂停 + ask。
|
||||
导航即返回 ask(入队、执行停止);结账页命中时 open 导航
|
||||
同样返回 ask 并暂停(round-12 修复轮项 4,P10 接线,
|
||||
detect_checkout 在快照时重算、导航前后两道门);观测到
|
||||
导航 POST 即自动暂停 + ask;下载命中可执行扩展名即取消
|
||||
+ 暂停 + ask。
|
||||
所有回包与日志过 sanitize_error;响应里绝无 ws 端点、
|
||||
DevTools 地址、原始 HTML。
|
||||
|
||||
@@ -56,6 +60,7 @@ MAX_REQUEST_BYTES = 65536
|
||||
NAV_TIMEOUT_MS = 30000
|
||||
|
||||
_GUARD = None
|
||||
_CHECKOUT = None
|
||||
|
||||
|
||||
def _load_guard():
|
||||
@@ -75,6 +80,40 @@ def _load_guard():
|
||||
return module
|
||||
|
||||
|
||||
def _load_checkout():
|
||||
"""加载结账检测模块 checkout_detect(纯函数模块)。
|
||||
|
||||
运行时安装在 /usr/local/bin/checkout_detect.py;测试时在
|
||||
仓库 cloud/chv/checkout_detect.py。找到即只加载一次;两处
|
||||
都找不到时返回 None(降级:结账检测关闭,broker 继续跑并
|
||||
写警告——不因缺一个检测模块把浏览器整路 crash-loop,但完
|
||||
整镜像必须含它,build.sh 有登记、BUILDINFO 有登记)。
|
||||
"""
|
||||
global _CHECKOUT
|
||||
if _CHECKOUT is not None:
|
||||
return _CHECKOUT
|
||||
here = os.path.dirname(os.path.abspath(__file__))
|
||||
candidates = [
|
||||
os.path.join(here, "checkout_detect.py"),
|
||||
os.path.join(os.path.dirname(here), "cloud", "chv",
|
||||
"checkout_detect.py"),
|
||||
]
|
||||
for path in candidates:
|
||||
if not os.path.exists(path):
|
||||
continue
|
||||
spec = importlib.util.spec_from_file_location(
|
||||
"checkout_detect", path)
|
||||
module = importlib.util.module_from_spec(spec)
|
||||
spec.loader.exec_module(module)
|
||||
_CHECKOUT = module
|
||||
return module
|
||||
sys.stderr.write(
|
||||
"chv-cdp-broker: 警告:checkout_detect 未找到(%s),"
|
||||
"结账检测关闭\n" % candidates)
|
||||
_CHECKOUT = None
|
||||
return None
|
||||
|
||||
|
||||
def _load_playwright():
|
||||
"""懒加载 playwright;缺失时报明错(不静默假装有浏览器)。
|
||||
返回调用后的上下文管理器(sync_playwright()),不是函数本身——
|
||||
@@ -102,6 +141,15 @@ class CdpBroker:
|
||||
blocklist_path=BLOCKLIST_PATH,
|
||||
ask_queue_dir=ASK_QUEUE_DIR, shots_dir=SHOTS_DIR):
|
||||
self._guard = _load_guard()
|
||||
# round-12 修复轮(项 4,P10):结账检测接线。模块缺失
|
||||
# 时 _checkout_mod 为 None,检测降级关闭(见 _load_checkout)。
|
||||
self._checkout_mod = _load_checkout()
|
||||
self._checkout = {"is_checkout": False, "confidence": "low",
|
||||
"reasons": [], "form_risk": None,
|
||||
"amount_text": None}
|
||||
# 本次会话里已批过 ask 的结账页 URL:批过之后同一页
|
||||
# 再 open 不再重复 ask(与 form_risk“批准即放行”同口径)。
|
||||
self._checkout_approved = set()
|
||||
self._factory = playwright_factory or _load_playwright
|
||||
self._blocklist_path = blocklist_path
|
||||
self._ask_queue_dir = ask_queue_dir
|
||||
@@ -176,6 +224,39 @@ class CdpBroker:
|
||||
return {"ok": False, "ask": True, "ask_id": ask_id,
|
||||
"error": "需审批:%s" % detail}
|
||||
|
||||
def _checkout_verdict(self, url, aria):
|
||||
"""结账检测:URL 信号 + ARIA 快照行拼成的伪节点。
|
||||
|
||||
broker 只有 aria_snapshot() 字符串,没有节点树;每行
|
||||
快照文本作一个节点(name=行文本),与 detect_checkout
|
||||
的节点口径一致(_node_text 拼 role/name/label/type)。
|
||||
模块缺失时返回“非结账页”(降级,见 _load_checkout)。
|
||||
"""
|
||||
if self._checkout_mod is None:
|
||||
return {"is_checkout": False, "confidence": "low",
|
||||
"reasons": [], "form_risk": None,
|
||||
"amount_text": None}
|
||||
nodes = [{"role": "", "name": line, "label": "", "type": ""}
|
||||
for line in (aria or "").splitlines()
|
||||
if line.strip()]
|
||||
verdict = self._checkout_mod.detect_checkout(url, nodes)
|
||||
return {"is_checkout": verdict.is_checkout,
|
||||
"confidence": verdict.confidence,
|
||||
"reasons": list(verdict.reasons),
|
||||
"form_risk": verdict.form_risk,
|
||||
"amount_text": verdict.amount_text}
|
||||
|
||||
def _checkout_ask(self, url, pause):
|
||||
"""结账页 ask:与 P4 的 form_submit 同种(P10 gap §4.1
|
||||
对齐),入队并按需暂停,返回 ask 响应。"""
|
||||
reasons = ",".join(self._checkout.get("reasons") or [])
|
||||
ask_id = self._guard.enqueue_ask(
|
||||
self._ask_queue_dir, "form_submit", url,
|
||||
"结账页命中(%s),购买前须审批" % reasons)
|
||||
if pause:
|
||||
self._paused = True
|
||||
return self._ask_resp(ask_id, "结账页命中,购买前须审批")
|
||||
|
||||
def open(self, url):
|
||||
if self._paused:
|
||||
return self._paused_resp()
|
||||
@@ -193,6 +274,11 @@ class CdpBroker:
|
||||
"高风险表单在页(%s),导航须审批"
|
||||
% self._form_risk["level"])
|
||||
return self._ask_resp(ask_id, "高风险表单在页,导航须审批")
|
||||
# round-12 修复轮(项 4,P10):导航前门——上一页已是
|
||||
# 结账页时,下一次导航须先审批(与 _form_risk 门同形)。
|
||||
if (self._checkout.get("is_checkout")
|
||||
and url not in self._checkout_approved):
|
||||
return self._checkout_ask(url, pause=False)
|
||||
try:
|
||||
self._ensure_browser()
|
||||
self._new_page()
|
||||
@@ -202,7 +288,13 @@ class CdpBroker:
|
||||
except Exception as e:
|
||||
return {"ok": False,
|
||||
"error": self._guard.sanitize_error(e)}
|
||||
return self.snapshot()
|
||||
result = self.snapshot()
|
||||
# 导航后门——落点本身是结账页时,返回 ask 并暂停
|
||||
#(P10 gap §4.1:open 导航返回 ask 并暂停)。
|
||||
if (result.get("ok") and self._checkout.get("is_checkout")
|
||||
and url not in self._checkout_approved):
|
||||
return self._checkout_ask(url, pause=True)
|
||||
return result
|
||||
|
||||
def snapshot(self):
|
||||
if self._paused:
|
||||
@@ -226,11 +318,15 @@ class CdpBroker:
|
||||
aria, truncated = self._guard.truncate_aria(aria or "")
|
||||
flags = self._guard.detect_injection(aria)
|
||||
self._form_risk = self._guard.detect_form_risk(aria)
|
||||
# round-12 修复轮(项 4,P10):每次快照重算结账 verdict,
|
||||
# 与 _form_risk 同形(快照缓存,open 的两道门读它)。
|
||||
self._checkout = self._checkout_verdict(url, aria)
|
||||
self._current_url = url
|
||||
return {"ok": True, "title": title, "url": url,
|
||||
"a11y": aria, "truncated": truncated,
|
||||
"untrusted_input": True, "injection_flags": flags,
|
||||
"form_risk": self._form_risk}
|
||||
"form_risk": self._form_risk,
|
||||
"checkout": self._checkout}
|
||||
|
||||
def screenshot(self):
|
||||
if self._paused:
|
||||
@@ -255,6 +351,9 @@ class CdpBroker:
|
||||
pass
|
||||
self._page = None
|
||||
self._form_risk = {"level": "none", "fields": []}
|
||||
self._checkout = {"is_checkout": False, "confidence": "low",
|
||||
"reasons": [], "form_risk": None,
|
||||
"amount_text": None}
|
||||
self._current_url = ""
|
||||
return {"ok": True, "closed": True}
|
||||
|
||||
@@ -272,13 +371,22 @@ class CdpBroker:
|
||||
def status(self):
|
||||
return {"ok": True, "paused": self._paused,
|
||||
"url": self._current_url,
|
||||
"form_risk": self._form_risk}
|
||||
"form_risk": self._form_risk,
|
||||
"checkout": self._checkout}
|
||||
|
||||
def list_asks(self):
|
||||
return {"ok": True,
|
||||
"asks": self._guard.list_asks(self._ask_queue_dir)}
|
||||
|
||||
def resolve_ask(self, ask_id, decision):
|
||||
# 批之前先读回 ask 记录:批后它不再是 pending,
|
||||
# list_asks 读不到——结账 ask 批过后要把 URL 记进
|
||||
# _checkout_approved(同页再 open 不再重复 ask)。
|
||||
record = None
|
||||
for ask in self._guard.list_asks(self._ask_queue_dir):
|
||||
if ask.get("ask_id") == ask_id:
|
||||
record = ask
|
||||
break
|
||||
ok, err = self._guard.resolve_ask(
|
||||
self._ask_queue_dir, ask_id, decision)
|
||||
if not ok:
|
||||
@@ -288,6 +396,15 @@ class CdpBroker:
|
||||
# 批准即放行这次操作:清掉快照缓存的表单风险(下一次
|
||||
# snapshot 会按新页面重算,不会丢检测)。
|
||||
self._form_risk = {"level": "none", "fields": []}
|
||||
# round-12 修复轮(项 4,P10):结账 ask 批过后,
|
||||
# 同 URL 再 open 不重复 ask;_checkout 本身也清掉,
|
||||
# 下一次 snapshot 按新页面重算。
|
||||
if record is not None and str(
|
||||
record.get("detail", "")).startswith("结账页命中"):
|
||||
self._checkout_approved.add(record.get("url"))
|
||||
self._checkout = {"is_checkout": False,
|
||||
"confidence": "low", "reasons": [],
|
||||
"form_risk": None, "amount_text": None}
|
||||
else:
|
||||
self.close()
|
||||
self._paused = False
|
||||
|
||||
@@ -76,6 +76,7 @@ DECLARED_SOURCES = {
|
||||
"chv-sync-worker-client": "repo_commit",
|
||||
"chv-gitea-worker.service": "repo_commit",
|
||||
"cloud/chv/privsep.py": "repo_commit",
|
||||
"cloud/chv/checkout_detect.py": "repo_commit",
|
||||
"cloud/chv/safety.py": "repo_commit",
|
||||
}
|
||||
|
||||
|
||||
Reference in new issue
Block a user