feat(backthink,browser): 自排工具面 C1-3 + 浏览器两处偏离收口（X2+X3 同文件一并落）
X2 C1-3（自排工具面与提示词）： - request_think（thinktool.go，照 transcode/tool.go 范式）+ reason 必填/urgency 闭集 + 双守门 （安全闸 + owner 判定：群聊/陌生人一律拒） - **成对**进两张唤醒态允许集（quiet + awake；CAPACITY_TOOL_SETS 是整体替换） - 提示词抽共用段落 + 自排前缀优先（不枚举工具清单）；面板加 self_wake 标签与筛选轴 - 4 个变异自检（拿掉 owner 判定/只加一张表/放宽 urgency/前端标签恒等） X3 收口（与 X2 共用 wire_services.go ⇒ 一并落）： - 删掉第三份 SSRF 实现（guard.go），改由 io.ValidateRequestURL 唯一判定 + 薄门面（保两个错误身份） - 浏览器会话挂进 a.close（真机：真 Chrome 被关、无残留、生产未扰动） - a.wireBrowser() 归位到 wire_runtime.go，删掉 ioToolOptions 的临时挂点 - 5 个变异自检 顺带补提交 C1-2 漏掉的 internal/config/runtimeconfig/backthink_test.go（ed50053 漏 add）。
This commit is contained in:
AskaEth committed
1 parent
2bf3b536dd
commit
2d211c2d16
20 files changed
+1568
-176
No files matched your search
@@ -0,0 +1,404 @@
|
||||
package backthink
|
||||
|
||||
import (
|
||||
"context"
|
||||
"errors"
|
||||
"strings"
|
||||
"sync"
|
||||
"testing"
|
||||
|
||||
"git.yeij.top/AskaEth/Cyrene_v2/internal/capability/registry"
|
||||
"git.yeij.top/AskaEth/Cyrene_v2/internal/capability/restrict"
|
||||
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/loop"
|
||||
"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/tool"
|
||||
"git.yeij.top/AskaEth/Cyrene_v2/internal/urgency"
|
||||
)
|
||||
|
||||
// 本文件是 design/12 §N C.3 工具面的判据（C.8 判据 7 / 8 与 C.3 的入参契约）。
|
||||
|
||||
// recordingWaker 是 Waker 的替身：记下每一次请求，并按脚本回一张收据或一个错误。
|
||||
// 「Waker 一次都没被调用」正是 C.8 判据 7 的断言点，因此调用记录是它存在的理由。
|
||||
type recordingWaker struct {
|
||||
mu sync.Mutex
|
||||
reqs []SelfWakeRequest
|
||||
receipt SelfWakeReceipt
|
||||
err error
|
||||
}
|
||||
|
||||
// WakeNow 记录请求并返回脚本结果（默认零值收据 = 被上限拒绝）。
|
||||
func (w *recordingWaker) WakeNow(_ context.Context, req SelfWakeRequest) (SelfWakeReceipt, error) {
|
||||
w.mu.Lock()
|
||||
defer w.mu.Unlock()
|
||||
w.reqs = append(w.reqs, req)
|
||||
if w.err != nil {
|
||||
return SelfWakeReceipt{}, w.err
|
||||
}
|
||||
return w.receipt, nil
|
||||
}
|
||||
|
||||
// calls 返回已记录的请求快照。
|
||||
func (w *recordingWaker) calls() []SelfWakeRequest {
|
||||
w.mu.Lock()
|
||||
defer w.mu.Unlock()
|
||||
out := make([]SelfWakeRequest, len(w.reqs))
|
||||
copy(out, w.reqs)
|
||||
return out
|
||||
}
|
||||
|
||||
// groupCtx 是群聊环境（第二道身份闸必须拒掉它）。stranger 同理，用一个就够。
|
||||
func groupCtx() context.Context {
|
||||
return loop.WithSessionEnv(context.Background(),
|
||||
loop.SessionEnv{ActorEnv: restrict.ActorGroupMember})
|
||||
}
|
||||
|
||||
// okResult 把工具结果断言成 map 并返回（类型不对就 Fatal）。
|
||||
func okResult(t *testing.T, val any) map[string]any {
|
||||
t.Helper()
|
||||
m, ok := val.(map[string]any)
|
||||
if !ok {
|
||||
t.Fatalf("工具结果类型 = %T, want map[string]any", val)
|
||||
}
|
||||
return m
|
||||
}
|
||||
|
||||
// TestThinkTool_OwnerQueuesAndReportsReceipt 断言主路径：主人环境 + 合法入参 ⇒ 请求原样
|
||||
// 到达 Waker，收据被原样映射成模型可读的结果（含 ok/queued/coalesced/used/limit/windowSec）。
|
||||
func TestThinkTool_OwnerQueuesAndReportsReceipt(t *testing.T) {
|
||||
t.Parallel()
|
||||
w := &recordingWaker{receipt: SelfWakeReceipt{Accepted: true, Used: 1, Limit: 4, WindowSec: 3600}}
|
||||
tt := NewThinkTool(w)
|
||||
|
||||
val, err := tt.Execute(ownerCtx(), map[string]any{
|
||||
thinkArgReason: "稍后回看她提过的日程",
|
||||
thinkArgUrgency: "warning",
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("Execute: %v", err)
|
||||
}
|
||||
reqs := w.calls()
|
||||
if len(reqs) != 1 {
|
||||
t.Fatalf("Waker 被调用 %d 次, want 1", len(reqs))
|
||||
}
|
||||
if reqs[0].Reason != "稍后回看她提过的日程" {
|
||||
t.Fatalf("理由 = %q, want 原样透传", reqs[0].Reason)
|
||||
}
|
||||
if reqs[0].Declared != urgency.LevelWarning {
|
||||
t.Fatalf("声明档位 = %s, want warning", reqs[0].Declared)
|
||||
}
|
||||
m := okResult(t, val)
|
||||
for k, want := range map[string]any{
|
||||
"ok": true, "queued": true, "coalesced": false, "reason": "",
|
||||
"used": 1, "limit": 4, "windowSec": 3600,
|
||||
} {
|
||||
if got := m[k]; got != want {
|
||||
t.Fatalf("结果 %s = %#v, want %#v（整张结果：%#v）", k, got, want, m)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// TestThinkTool_MissingUrgencyMeansRoutine 断言省略 urgency 就是例行档（零值），
|
||||
// 绝不被猜成更高档 —— 「猜错档位」的代价是让一轮思考穿透安静时段。
|
||||
func TestThinkTool_MissingUrgencyMeansRoutine(t *testing.T) {
|
||||
t.Parallel()
|
||||
w := &recordingWaker{receipt: SelfWakeReceipt{Accepted: true, Used: 1, Limit: 4, WindowSec: 3600}}
|
||||
tt := NewThinkTool(w)
|
||||
|
||||
if _, err := tt.Execute(ownerCtx(), map[string]any{thinkArgReason: "只是想再想想"}); err != nil {
|
||||
t.Fatalf("Execute: %v", err)
|
||||
}
|
||||
if got := w.calls()[0].Declared; got != urgency.LevelRoutine {
|
||||
t.Fatalf("省略 urgency 的档位 = %s, want routine", got)
|
||||
}
|
||||
}
|
||||
|
||||
// TestThinkTool_GroupMemberRefusedWithoutCallingWaker 是 C.8 判据 7：群聊（非主人）
|
||||
// 环境必须返回包装后的 ErrSelfWakeNotOwner，且 **Waker 一次都没被调用**。
|
||||
//
|
||||
// 变异：删掉 Execute 里的 owner 判定 ⇒ 本用例变红（群聊能自排）。
|
||||
func TestThinkTool_GroupMemberRefusedWithoutCallingWaker(t *testing.T) {
|
||||
t.Parallel()
|
||||
w := &recordingWaker{receipt: SelfWakeReceipt{Accepted: true, Used: 1, Limit: 4, WindowSec: 3600}}
|
||||
tt := NewThinkTool(w)
|
||||
|
||||
_, err := tt.Execute(groupCtx(), map[string]any{thinkArgReason: "偷排一轮"})
|
||||
if !errors.Is(err, ErrSelfWakeNotOwner) {
|
||||
t.Fatalf("群聊环境的错误 = %v, want 包装后的 ErrSelfWakeNotOwner", err)
|
||||
}
|
||||
if len(w.calls()) != 0 {
|
||||
t.Fatalf("群聊环境触到了 Waker（它必须「不做事」）：%+v", w.calls())
|
||||
}
|
||||
}
|
||||
|
||||
// TestThinkTool_StrangerRefused 断言陌生人环境同样被拒（三种 env 里两个受限档都要拦）。
|
||||
func TestThinkTool_StrangerRefused(t *testing.T) {
|
||||
t.Parallel()
|
||||
w := &recordingWaker{}
|
||||
tt := NewThinkTool(w)
|
||||
ctx := loop.WithSessionEnv(context.Background(), loop.SessionEnv{ActorEnv: restrict.ActorStranger})
|
||||
|
||||
if _, err := tt.Execute(ctx, map[string]any{thinkArgReason: "x"}); !errors.Is(err, ErrSelfWakeNotOwner) {
|
||||
t.Fatalf("陌生人环境的错误 = %v, want ErrSelfWakeNotOwner", err)
|
||||
}
|
||||
if len(w.calls()) != 0 {
|
||||
t.Fatalf("陌生人环境触到了 Waker：%+v", w.calls())
|
||||
}
|
||||
}
|
||||
|
||||
// TestThinkTool_UrgencyIsAClosedSet 是 C.8 判据 8：只有闭集里的**精确字面量**被认，
|
||||
// 其余（含近义词、前缀、大小写、带尾随空格）一律零值 routine。
|
||||
//
|
||||
// 变异：把判定放宽成前缀/子串匹配（例如 strings.HasPrefix(raw, "crit")）⇒
|
||||
// "critical " / "crit" 会被读成 critical，本用例变红。
|
||||
func TestThinkTool_UrgencyIsAClosedSet(t *testing.T) {
|
||||
t.Parallel()
|
||||
cases := []struct {
|
||||
in string
|
||||
want urgency.Level
|
||||
}{
|
||||
{"routine", urgency.LevelRoutine},
|
||||
{"warning", urgency.LevelWarning},
|
||||
{"critical", urgency.LevelCritical},
|
||||
{"", urgency.LevelRoutine},
|
||||
{"urgent", urgency.LevelRoutine},
|
||||
{"紧急", urgency.LevelRoutine},
|
||||
{"CRITICAL", urgency.LevelRoutine},
|
||||
{"critical ", urgency.LevelRoutine},
|
||||
{"crit", urgency.LevelRoutine},
|
||||
{"warn", urgency.LevelRoutine},
|
||||
}
|
||||
for _, c := range cases {
|
||||
w := &recordingWaker{}
|
||||
tt := NewThinkTool(w)
|
||||
if _, err := tt.Execute(ownerCtx(), map[string]any{
|
||||
thinkArgReason: "看看", thinkArgUrgency: c.in,
|
||||
}); err != nil {
|
||||
t.Fatalf("urgency=%q 时 Execute: %v", c.in, err)
|
||||
}
|
||||
if got := w.calls()[0].Declared; got != c.want {
|
||||
t.Fatalf("urgency=%q ⇒ %s, want %s", c.in, got, c.want)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// TestThinkTool_ReasonRequired 断言理由缺省 / 空白 / 类型不对都是参数错误，且
|
||||
// **不替她编一个理由**（Waker 不被调用）。
|
||||
func TestThinkTool_ReasonRequired(t *testing.T) {
|
||||
t.Parallel()
|
||||
bad := []struct {
|
||||
name string
|
||||
args any
|
||||
}{
|
||||
{"缺字段", map[string]any{}},
|
||||
{"空串", map[string]any{thinkArgReason: ""}},
|
||||
{"纯空白", map[string]any{thinkArgReason: " \n\t "}},
|
||||
{"非字符串", map[string]any{thinkArgReason: 42}},
|
||||
{"整份参数不是对象", "just a string"},
|
||||
}
|
||||
for _, c := range bad {
|
||||
w := &recordingWaker{}
|
||||
tt := NewThinkTool(w)
|
||||
_, err := tt.Execute(ownerCtx(), c.args)
|
||||
if !errors.Is(err, ErrSelfWakeToolArgs) {
|
||||
t.Fatalf("%s：错误 = %v, want ErrSelfWakeToolArgs", c.name, err)
|
||||
}
|
||||
if len(w.calls()) != 0 {
|
||||
t.Fatalf("%s：Waker 被调用了 %+v（不该发生）", c.name, w.calls())
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// TestThinkTool_ReasonTrimmedAndTruncated 断言理由去首尾空白、按 **rune** 截断（不腰斩
|
||||
// 多字节字符），并把截断如实回报 —— 不静默丢弃。
|
||||
func TestThinkTool_ReasonTrimmedAndTruncated(t *testing.T) {
|
||||
t.Parallel()
|
||||
w := &recordingWaker{receipt: SelfWakeReceipt{Accepted: true, Used: 1, Limit: 4, WindowSec: 3600}}
|
||||
tt := NewThinkTool(w)
|
||||
|
||||
long := strings.Repeat("想", selfWakeReasonMaxRunes+50)
|
||||
val, err := tt.Execute(ownerCtx(), map[string]any{thinkArgReason: " " + long + " "})
|
||||
if err != nil {
|
||||
t.Fatalf("Execute: %v", err)
|
||||
}
|
||||
got := w.calls()[0].Reason
|
||||
if len([]rune(got)) != selfWakeReasonMaxRunes {
|
||||
t.Fatalf("理由长度 = %d rune, want %d", len([]rune(got)), selfWakeReasonMaxRunes)
|
||||
}
|
||||
if !strings.HasPrefix(got, "想") {
|
||||
t.Fatalf("截断把多字节字符切坏了：%q", got)
|
||||
}
|
||||
if m := okResult(t, val); m["reasonTruncated"] != true {
|
||||
t.Fatalf("截断没有如实回报：%#v", m)
|
||||
}
|
||||
|
||||
// 未超长时首尾空白被去掉，且不报截断。
|
||||
val2, err := tt.Execute(ownerCtx(), map[string]any{thinkArgReason: " 回看日程 "})
|
||||
if err != nil {
|
||||
t.Fatalf("Execute: %v", err)
|
||||
}
|
||||
if got := w.calls()[1].Reason; got != "回看日程" {
|
||||
t.Fatalf("理由 = %q, want 去空白后的 %q", got, "回看日程")
|
||||
}
|
||||
if m := okResult(t, val2); m["reasonTruncated"] != false {
|
||||
t.Fatalf("未截断却报了截断：%#v", m)
|
||||
}
|
||||
}
|
||||
|
||||
// TestThinkTool_CappedIsABusinessResultNotAnError 断言「被上限拒绝」是**合法业务结果**
|
||||
// （ok:false、err 为 nil）：她要看得见拒绝才会停止连排；把它伪装成 error 或反过来
|
||||
// 把拒绝伪装成成功，都会让她误判。
|
||||
func TestThinkTool_CappedIsABusinessResultNotAnError(t *testing.T) {
|
||||
t.Parallel()
|
||||
w := &recordingWaker{receipt: SelfWakeReceipt{
|
||||
Accepted: false, Coalesced: false,
|
||||
Reason: "已达自我唤醒上限（本窗口 4/4，窗口 1h0m0s），本轮未排队",
|
||||
Used: 4, Limit: 4, WindowSec: 3600,
|
||||
}}
|
||||
tt := NewThinkTool(w)
|
||||
|
||||
val, err := tt.Execute(ownerCtx(), map[string]any{thinkArgReason: "再想一轮"})
|
||||
if err != nil {
|
||||
t.Fatalf("被上限拒绝必须是业务结果而不是 error，got %v", err)
|
||||
}
|
||||
m := okResult(t, val)
|
||||
if m["ok"] != false || m["queued"] != false {
|
||||
t.Fatalf("拒绝的收据映射错了：%#v", m)
|
||||
}
|
||||
if m["used"] != 4 || m["limit"] != 4 {
|
||||
t.Fatalf("拒绝没带上用量口径：%#v", m)
|
||||
}
|
||||
if s, _ := m["reason"].(string); !strings.Contains(s, "上限") {
|
||||
t.Fatalf("拒绝说明没原样带出：%#v", m)
|
||||
}
|
||||
}
|
||||
|
||||
// TestThinkTool_CoalescedReportsAcceptedButNotQueued 断言合并语义在工具面上可分辨：
|
||||
// 合并 = ok:true + coalesced:true + queued:false（不额外产生一轮）。
|
||||
func TestThinkTool_CoalescedReportsAcceptedButNotQueued(t *testing.T) {
|
||||
t.Parallel()
|
||||
w := &recordingWaker{receipt: SelfWakeReceipt{Accepted: true, Coalesced: true, Used: 2, Limit: 4, WindowSec: 3600}}
|
||||
tt := NewThinkTool(w)
|
||||
|
||||
val, err := tt.Execute(ownerCtx(), map[string]any{thinkArgReason: "合并到在队那轮"})
|
||||
if err != nil {
|
||||
t.Fatalf("Execute: %v", err)
|
||||
}
|
||||
m := okResult(t, val)
|
||||
if m["ok"] != true || m["coalesced"] != true || m["queued"] != false {
|
||||
t.Fatalf("合并的收据映射错了：%#v", m)
|
||||
}
|
||||
}
|
||||
|
||||
// TestThinkTool_WakerErrorsSurfaceAsErrors 断言服务侧的真实故障（未 Start、未接线）
|
||||
// 是 error，**不是** ok:true —— 红线：不伪造成功。
|
||||
func TestThinkTool_WakerErrorsSurfaceAsErrors(t *testing.T) {
|
||||
t.Parallel()
|
||||
w := &recordingWaker{err: ErrSelfWakeUnavailable}
|
||||
tt := NewThinkTool(w)
|
||||
|
||||
if _, err := tt.Execute(ownerCtx(), map[string]any{thinkArgReason: "x"}); !errors.Is(err, ErrSelfWakeUnavailable) {
|
||||
t.Fatalf("未 Start 的错误 = %v, want 包装后的 ErrSelfWakeUnavailable", err)
|
||||
}
|
||||
|
||||
nilWaker := NewThinkTool(nil)
|
||||
if _, err := nilWaker.Execute(ownerCtx(), map[string]any{thinkArgReason: "x"}); !errors.Is(err, ErrSelfWakeUnavailable) {
|
||||
t.Fatalf("未接线 Waker 的错误 = %v, want ErrSelfWakeUnavailable", err)
|
||||
}
|
||||
}
|
||||
|
||||
// TestThinkTool_ParametersDeclareTheClosedSet 断言工具面把闭集**广告**出去：
|
||||
// reason 必填、urgency 的 enum 恰是三个档位字面量（与 urgency 词汇同源）。
|
||||
func TestThinkTool_ParametersDeclareTheClosedSet(t *testing.T) {
|
||||
t.Parallel()
|
||||
params, err := NewThinkTool(&recordingWaker{}).Parameters()
|
||||
if err != nil {
|
||||
t.Fatalf("Parameters: %v", err)
|
||||
}
|
||||
props, ok := params["properties"].(map[string]any)
|
||||
if !ok {
|
||||
t.Fatalf("properties 类型 = %T", params["properties"])
|
||||
}
|
||||
if _, ok := props[thinkArgReason]; !ok {
|
||||
t.Fatalf("schema 里没有 %q：%#v", thinkArgReason, props)
|
||||
}
|
||||
req, ok := params["required"].([]any)
|
||||
if !ok || len(req) != 1 || req[0] != thinkArgReason {
|
||||
t.Fatalf("required = %#v, want 只含 %q", params["required"], thinkArgReason)
|
||||
}
|
||||
urgencyProp, ok := props[thinkArgUrgency].(map[string]any)
|
||||
if !ok {
|
||||
t.Fatalf("%s 属性类型 = %T", thinkArgUrgency, props[thinkArgUrgency])
|
||||
}
|
||||
enum, ok := urgencyProp["enum"].([]any)
|
||||
if !ok {
|
||||
t.Fatalf("enum 类型 = %T, want []any", urgencyProp["enum"])
|
||||
}
|
||||
want := map[string]bool{"routine": true, "warning": true, "critical": true}
|
||||
if len(enum) != len(want) {
|
||||
t.Fatalf("enum = %#v, want 恰好三个闭集取值", enum)
|
||||
}
|
||||
for _, v := range enum {
|
||||
s, _ := v.(string)
|
||||
if !want[s] {
|
||||
t.Fatalf("enum 里有闭集之外的取值 %#v（want %v）", v, want)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// TestThinkTool_ThroughDispatcher 走**真实分发器**（注册表 → JSON-Schema 校验 →
|
||||
// 工具执行），证明三件事：
|
||||
// 1. 注册进注册表后模型真的调得到，且合法入参一路走到 Waker；
|
||||
// 2. 群聊环境在**执行前**被第二道闸拒掉（Waker 不被触碰）；
|
||||
// 3. 闭集之外的 urgency 被 schema 拦在 Execute 之前（工具面本来就是闭集），
|
||||
// 因此 Execute 里的闭集兜底只服务直接调用。
|
||||
func TestThinkTool_ThroughDispatcher(t *testing.T) {
|
||||
t.Parallel()
|
||||
reg := registry.New()
|
||||
d, err := tool.New(reg)
|
||||
if err != nil {
|
||||
t.Fatalf("tool.New: %v", err)
|
||||
}
|
||||
w := &recordingWaker{receipt: SelfWakeReceipt{Accepted: true, Used: 1, Limit: 4, WindowSec: 3600}}
|
||||
if err := d.Register(NewThinkTool(w)); err != nil {
|
||||
t.Fatalf("Register: %v", err)
|
||||
}
|
||||
cap, err := reg.Get(ThinkToolName)
|
||||
if err != nil {
|
||||
t.Fatalf("%s 没进能力注册表: %v", ThinkToolName, err)
|
||||
}
|
||||
if cap.Kind != registry.KindTool {
|
||||
t.Fatalf("能力类型 = %q, want %q", cap.Kind, registry.KindTool)
|
||||
}
|
||||
// 策略标记保持零值 = owner-only（design/11 §3 的 fail-closed 默认）。
|
||||
if cap.Policy.AllowInGroup || cap.Policy.AllowInStranger {
|
||||
t.Fatalf("%s 的策略必须保持 owner-only，got %+v", ThinkToolName, cap.Policy)
|
||||
}
|
||||
|
||||
val, err := d.Invoke(ownerCtx(), ThinkToolName, map[string]any{
|
||||
thinkArgReason: "稍后回看她的日程", thinkArgUrgency: "warning",
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("Invoke(owner): %v", err)
|
||||
}
|
||||
if m := okResult(t, val); m["ok"] != true {
|
||||
t.Fatalf("Invoke(owner) 结果 = %#v", m)
|
||||
}
|
||||
if len(w.calls()) != 1 || w.calls()[0].Declared != urgency.LevelWarning {
|
||||
t.Fatalf("Waker 收到的请求 = %+v", w.calls())
|
||||
}
|
||||
|
||||
if _, err := d.Invoke(groupCtx(), ThinkToolName, map[string]any{thinkArgReason: "偷排"}); !errors.Is(err, ErrSelfWakeNotOwner) {
|
||||
t.Fatalf("Invoke(group) 错误 = %v, want ErrSelfWakeNotOwner", err)
|
||||
}
|
||||
if len(w.calls()) != 1 {
|
||||
t.Fatalf("群聊调用触到了 Waker：%+v", w.calls())
|
||||
}
|
||||
|
||||
if _, err := d.Invoke(ownerCtx(), ThinkToolName, map[string]any{
|
||||
thinkArgReason: "x", thinkArgUrgency: "urgent",
|
||||
}); err == nil {
|
||||
t.Fatal("闭集之外的 urgency 必须被 schema 拒绝，而不是静默当成 routine")
|
||||
}
|
||||
if len(w.calls()) != 1 {
|
||||
t.Fatalf("被 schema 拒掉的调用触到了 Waker：%+v", w.calls())
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user