Files
ARTEX-chixuweihu/intercept/jev/client_test.go
T

327 lines
14 KiB
Go

package jev
import (
"context"
"encoding/json"
"fmt"
"math"
"net/http"
"net/http/httptest"
"strings"
"testing"
"time"
"unicode/utf8"
"github.com/Autumn-27/artex/intercept"
)
func TestBuildRequestMapsReviewInputAndChoiceCriteria(t *testing.T) {
req, err := BuildRequest(intercept.ReviewInput{Version: 5, Tool: "Bash", Arguments: json.RawMessage(`{"cmd":"ls"}`)}, "policy text", "jev-test")
if err != nil {
t.Fatal(err)
}
if req.Model != "jev-test" || req.Questions["decision"].Type != "choice" {
t.Fatalf("unexpected request: %#v", req)
}
b, _ := json.Marshal(req.State)
if !strings.Contains(string(b), "Bash") || strings.Contains(string(b), "policy text") {
t.Fatalf("state should contain evidence only: %s", b)
}
if !strings.Contains(req.Questions["decision"].Instructions, "policy text") {
t.Fatalf("custom policy missing from instructions: %s", req.Questions["decision"].Instructions)
}
if len(b) > maxStateBytes+2000 {
t.Fatalf("state exceeded bound: %d", len(b))
}
}
func TestBuildRequestUsesCompactPolicyForDefaultPrompts(t *testing.T) {
for _, prompt := range []string{"", intercept.DefaultJudgePrompt, intercept.EffectiveJudgePrompt(intercept.DefaultJudgePrompt)} {
req, err := BuildRequest(intercept.ReviewInput{Version: 5, Tool: "Bash", Arguments: json.RawMessage(`{"cmd":"cat /etc/shadow"}`)}, prompt, "")
if err != nil {
t.Fatal(err)
}
instructions := req.Questions["decision"].Instructions
if !strings.Contains(instructions, "漏洞利用 payload") || !strings.Contains(instructions, "其余默认 allow") {
t.Fatalf("default policy lost authorized-pentest semantics: %s", instructions)
}
if strings.Contains(instructions, intercept.JudgeContextBoundary) || strings.Contains(instructions, intercept.JudgeOutputContract) {
t.Fatalf("default prompt was copied into Jev instructions: %s", instructions)
}
}
}
func TestBuildRequestBoundsLargeEvidence(t *testing.T) {
args := json.RawMessage(`{"payload":"` + strings.Repeat("x", 50000) + `"}`)
req, err := BuildRequest(intercept.ReviewInput{Version: 5, Tool: "Bash", Arguments: args}, strings.Repeat("policy ", 3000), "")
if err != nil {
t.Fatal(err)
}
b, _ := json.Marshal(req.State)
if len(b) > maxStateBytes+500 {
t.Fatalf("bounded state size=%d", len(b))
}
if !req.EvidenceTruncated || !strings.Contains(string(b), `"evidence_truncated":true`) {
t.Fatalf("large evidence was not marked truncated: %s", b[:min(len(b), 200)])
}
if !json.Valid(b) || len(b) > maxStateBytes {
t.Fatalf("truncated evidence state invalid or too large: %d", len(b))
}
}
func TestTruncateUTF8KeepsBoundary(t *testing.T) {
got, truncated := truncateUTF8(strings.Repeat("审查", 20), 20)
if !truncated || !utf8.ValidString(got) || len(got) > 20 {
t.Fatalf("truncateUTF8 result=%q truncated=%v len=%d", got, truncated, len(got))
}
}
func TestDecisionFromResponseLowConfidenceForcesAsk(t *testing.T) {
d, err := DecisionFromResponse(Response{Answers: map[string]Answer{
"decision": {Type: "choice", Choice: "allow", Confidence: 0.3, Probabilities: map[string]float64{"allow": 0.8, "ask": 0.1, "deny": 0.1}},
}}, 0.55, 42)
if err != nil {
t.Fatal(err)
}
if d.Action != "ask" || d.ProfileID != 42 || !strings.Contains(d.Message, "置信度") {
t.Fatalf("unexpected decision: %#v", d)
}
}
func TestDecisionFromResponseRejectsMissingOrInvalidAnswer(t *testing.T) {
for _, resp := range []Response{
{Answers: map[string]Answer{}},
{Answers: map[string]Answer{"decision": {Choice: "maybe", Confidence: 1}}},
} {
if _, err := DecisionFromResponse(resp, 0.55, 1); err == nil {
t.Fatalf("expected invalid response error for %#v", resp)
}
}
}
func TestDecisionFromResponseThresholdZeroAllowsLowConfidence(t *testing.T) {
d, err := DecisionFromResponse(Response{Answers: map[string]Answer{
"decision": {Type: "choice", Choice: "allow", Confidence: 0.01, Probabilities: map[string]float64{"allow": 0.8, "ask": 0.1, "deny": 0.1}},
}}, 0, 1)
if err != nil || d.Action != "allow" {
t.Fatalf("threshold zero decision=%#v err=%v", d, err)
}
}
func TestReviewForcesAskWhenEvidenceWasTruncated(t *testing.T) {
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.Header().Set("Content-Type", "application/json")
_, _ = w.Write([]byte(`{"answers":{"decision":{"type":"choice","choice":"allow","confidence":1,"probabilities":{"allow":1,"ask":0,"deny":0}}}}`))
}))
defer srv.Close()
args := json.RawMessage(`{"payload":"` + strings.Repeat("x", 50000) + `"}`)
d, _, err := (Config{BaseURL: srv.URL, APIKey: "secret", Client: srv.Client()}).ReviewWithUsage(
context.Background(), intercept.ReviewInput{Version: 5, Tool: "Bash", Arguments: args}, "", 0, 8)
if err != nil || d.Action != "ask" || !strings.Contains(d.Message, "证据已截断") {
t.Fatalf("decision=%#v err=%v", d, err)
}
}
func TestReviewLowConfidenceTruncatedEvidenceKeepsBothReasons(t *testing.T) {
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.Header().Set("Content-Type", "application/json")
_, _ = w.Write([]byte(`{"answers":{"decision":{"type":"choice","choice":"allow","confidence":0.2,"probabilities":{"allow":0.8,"ask":0.1,"deny":0.1}}}}`))
}))
defer srv.Close()
args := json.RawMessage(`{"payload":"` + strings.Repeat("x", 50000) + `"}`)
d, _, err := (Config{BaseURL: srv.URL, APIKey: "secret", Client: srv.Client()}).ReviewWithUsage(
context.Background(), intercept.ReviewInput{Version: 5, Tool: "Bash", Arguments: args}, "", 0.55, 8)
if err != nil || d.Action != "ask" || !strings.Contains(d.Message, "置信度") || !strings.Contains(d.Message, "证据已截断") {
t.Fatalf("decision=%#v err=%v", d, err)
}
}
func TestReviewForcesAskForIncompleteScriptEvidence(t *testing.T) {
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.Header().Set("Content-Type", "application/json")
_, _ = w.Write([]byte(`{"answers":{"decision":{"type":"choice","choice":"allow","confidence":1,"probabilities":{"allow":1,"ask":0,"deny":0}}}}`))
}))
defer srv.Close()
d, _, err := (Config{BaseURL: srv.URL, APIKey: "secret", Client: srv.Client()}).ReviewWithUsage(
context.Background(), intercept.ReviewInput{Version: 5, Tool: "Bash", ScriptCaptureStatus: "incomplete", Arguments: json.RawMessage(`{"cmd":"run"}`)}, "", 0, 8)
if err != nil || d.Action != "ask" || !strings.Contains(d.Message, "证据已截断") {
t.Fatalf("decision=%#v err=%v", d, err)
}
}
func TestReviewWithUsageKeepsParseableErrorUsage(t *testing.T) {
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.WriteHeader(http.StatusBadGateway)
_, _ = w.Write([]byte(`{"usage":{"input_tokens":17,"output_tokens":3},"error":"busy"}`))
}))
defer srv.Close()
_, usage, err := (Config{BaseURL: srv.URL, APIKey: "secret", Client: srv.Client()}).ReviewWithUsage(
context.Background(), intercept.ReviewInput{Version: 5, Tool: "Read", Arguments: json.RawMessage(`{}`)}, "", 0.55, 3)
if err == nil || usage.InputTokens != 17 || usage.OutputTokens != 3 {
t.Fatalf("usage=%+v err=%v", usage, err)
}
}
func TestSystemOneUsesConfiguredProxy(t *testing.T) {
proxy := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if r.URL.Path != "/v1/systemone" {
t.Errorf("proxy path=%s", r.URL.Path)
}
w.Header().Set("Content-Type", "application/json")
_, _ = w.Write([]byte(`{"answers":{"decision":{"type":"choice","choice":"deny","confidence":1,"probabilities":{"allow":0,"ask":0,"deny":1}}}}`))
}))
defer proxy.Close()
lat, _, err := (Config{BaseURL: "http://target.invalid", Proxy: proxy.URL, APIKey: "secret"}).TestConnection(context.Background())
if err != nil || lat <= 0 {
t.Fatalf("proxy test latency=%v err=%v", lat, err)
}
}
func TestTestConnectionAddsThirtySecondDeadline(t *testing.T) {
client := &http.Client{Transport: roundTripFunc(func(req *http.Request) (*http.Response, error) {
deadline, ok := req.Context().Deadline()
if !ok || time.Until(deadline) > 30*time.Second || time.Until(deadline) < 29*time.Second {
return nil, fmt.Errorf("unexpected deadline: %v", deadline)
}
return nil, context.DeadlineExceeded
})}
_, _, err := (Config{BaseURL: "http://target.invalid", APIKey: "secret", Client: client}).TestConnection(context.Background())
if err == nil || !strings.Contains(err.Error(), "context deadline exceeded") {
t.Fatalf("expected context deadline error, got %v", err)
}
}
type roundTripFunc func(*http.Request) (*http.Response, error)
func (f roundTripFunc) RoundTrip(req *http.Request) (*http.Response, error) { return f(req) }
func TestSystemOneReviewAndConnectionTest(t *testing.T) {
var gotAuth string
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if r.URL.Path != "/v1/systemone" {
t.Errorf("path=%s", r.URL.Path)
}
gotAuth = r.Header.Get("Authorization")
var req Request
if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
t.Errorf("decode: %v", err)
}
if req.Questions["decision"].Criteria["deny"] == "" {
t.Error("missing deny criterion")
}
w.Header().Set("Content-Type", "application/json")
_, _ = w.Write([]byte(`{"answers":{"decision":{"type":"choice","choice":"deny","confidence":0.91,"probabilities":{"allow":0.02,"ask":0.07,"deny":0.91}}}}`))
}))
defer srv.Close()
c := Config{BaseURL: srv.URL, APIKey: "secret", Model: "jev-latest", Client: srv.Client()}
d, err := c.Review(context.Background(), intercept.ReviewInput{Version: 5, Tool: "Write", Arguments: json.RawMessage(`{}`)}, "policy", 0.55, 9)
if err != nil || d.Action != "deny" {
t.Fatalf("review decision=%#v err=%v", d, err)
}
if gotAuth != "Bearer secret" {
t.Fatalf("auth=%q", gotAuth)
}
if _, _, err := c.TestConnection(context.Background()); err != nil {
t.Fatalf("connection test: %v", err)
}
}
func TestEndpointNormalizesAndValidatesBaseURL(t *testing.T) {
for _, tc := range []struct{ base, want string }{
{"", "https://api.typesafe.ai/v1/systemone"},
{"https://example.test/", "https://example.test/v1/systemone"},
{"https://example.test/v1/", "https://example.test/v1/systemone"},
{"https://example.test/v1/systemone", "https://example.test/v1/systemone"},
{"http://example.test/gateway/v1", "http://example.test/gateway/v1/systemone"},
} {
got, err := (Config{BaseURL: tc.base}).endpoint()
if err != nil || got != tc.want {
t.Fatalf("base=%q got=%q err=%v", tc.base, got, err)
}
}
for _, base := range []string{"/v1", "/v1/systemone", "ftp://example.test/v1", "https://example.test?secret=x", "https://example.test/#fragment"} {
if _, err := (Config{BaseURL: base}).endpoint(); err == nil {
t.Errorf("accepted invalid URL %q", base)
}
}
}
func TestListModelsNativeSchemaAndProxy(t *testing.T) {
proxy := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if r.Method != http.MethodGet || r.URL.Path != "/v1/models" {
t.Errorf("request = %s %s", r.Method, r.URL)
}
if r.Header.Get("Authorization") != "Bearer secret" {
t.Error("missing authorization")
}
_, _ = w.Write([]byte(`{"models":[{"name":"jev-latest"},{"name":"jev-preview"},{"name":""}]}`))
}))
defer proxy.Close()
for _, suffix := range []string{"", "/v1", "/v1/systemone"} {
models, err := (Config{BaseURL: "http://target.invalid" + suffix, Proxy: proxy.URL, APIKey: " secret "}).ListModels(context.Background())
if err != nil || strings.Join(models, ",") != "jev-latest,jev-preview" {
t.Fatalf("models=%v err=%v", models, err)
}
}
}
func TestListModelsErrorsAndEmptyList(t *testing.T) {
for _, tc := range []struct {
code int
body string
wantErr bool
}{
{401, `{"detail":"invalid key"}`, true},
{200, `not json`, true},
{200, `{"models":[]}`, false},
} {
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { w.WriteHeader(tc.code); _, _ = w.Write([]byte(tc.body)) }))
models, err := (Config{BaseURL: srv.URL, APIKey: "key"}).ListModels(context.Background())
srv.Close()
if (err != nil) != tc.wantErr || (!tc.wantErr && models == nil) {
t.Fatalf("status=%d models=%v err=%v", tc.code, models, err)
}
}
}
func TestDecisionRejectsMalformedChoiceAnswer(t *testing.T) {
valid := func() Answer {
return Answer{Type: "choice", Choice: "allow", Confidence: 0.9, Probabilities: map[string]float64{"allow": 0.9, "ask": 0.05, "deny": 0.05}}
}
for _, mutate := range []func(*Answer){
func(a *Answer) { a.Type = "score" },
func(a *Answer) { a.Confidence = 1.1 },
func(a *Answer) { a.Confidence = -0.1 },
func(a *Answer) { a.Confidence = math.NaN() },
func(a *Answer) { a.Confidence = math.Inf(1) },
func(a *Answer) { a.Probabilities = nil },
func(a *Answer) { delete(a.Probabilities, "deny") },
func(a *Answer) { a.Probabilities["allow"] = 1.1 },
func(a *Answer) { a.Probabilities["ask"] = math.NaN() },
} {
a := valid()
mutate(&a)
if _, err := DecisionFromResponse(Response{Answers: map[string]Answer{"decision": a}}, 0, 1); err == nil {
t.Fatalf("accepted malformed answer %+v", a)
}
}
for _, raw := range []string{
`{"answers":{"decision":{"type":"choice","choice":"allow","probabilities":{"allow":1,"ask":0,"deny":0}}}}`,
`{"answers":{"decision":{"type":"choice","choice":"allow","confidence":null,"probabilities":{"allow":1,"ask":0,"deny":0}}}}`,
} {
var resp Response
if err := json.Unmarshal([]byte(raw), &resp); err == nil {
t.Fatalf("accepted missing confidence: %s", raw)
}
}
}
func TestTruncateUTF8AlwaysHonorsByteLimit(t *testing.T) {
for limit := 0; limit < 80; limit++ {
got, _ := truncateUTF8(strings.Repeat("审查", 30), limit)
if len(got) > limit || !utf8.ValidString(got) {
t.Fatalf("limit=%d bytes=%d", limit, len(got))
}
}
}