feat(exec): add Claude verifier for local model output quality gate
This commit is contained in:
74
internal/exec/verifier_test.go
Normal file
74
internal/exec/verifier_test.go
Normal file
@@ -0,0 +1,74 @@
|
||||
package exec_test
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
iexec "github.com/mathiasbq/supervisor/internal/exec"
|
||||
"github.com/stretchr/testify/assert"
|
||||
"github.com/stretchr/testify/require"
|
||||
)
|
||||
|
||||
func fakeVerifierClaude(t *testing.T, verdict iexec.Verdict) string {
|
||||
t.Helper()
|
||||
data, err := json.Marshal(verdict)
|
||||
require.NoError(t, err)
|
||||
dir := t.TempDir()
|
||||
script := filepath.Join(dir, "claude")
|
||||
content := fmt.Sprintf("#!/bin/sh\necho '%s'\n", string(data))
|
||||
require.NoError(t, os.WriteFile(script, []byte(content), 0755))
|
||||
return script
|
||||
}
|
||||
|
||||
func TestVerifierAccepts(t *testing.T) {
|
||||
claude := fakeVerifierClaude(t, iexec.Verdict{Accept: true, Feedback: ""})
|
||||
v := iexec.NewVerifier(claude, "claude-sonnet-4-6", 5*time.Second)
|
||||
|
||||
verdict, err := v.Verify(context.Background(), "skill rules", "do the task", iexec.Result{
|
||||
Status: "pass", Phase: "review", Skill: "review", Message: "ok",
|
||||
})
|
||||
require.NoError(t, err)
|
||||
assert.True(t, verdict.Accept)
|
||||
assert.Empty(t, verdict.Feedback)
|
||||
}
|
||||
|
||||
func TestVerifierEscalates(t *testing.T) {
|
||||
claude := fakeVerifierClaude(t, iexec.Verdict{Accept: false, Feedback: "missing line references"})
|
||||
v := iexec.NewVerifier(claude, "claude-sonnet-4-6", 5*time.Second)
|
||||
|
||||
verdict, err := v.Verify(context.Background(), "skill rules", "do the task", iexec.Result{
|
||||
Status: "pass", Phase: "review", Skill: "review", Message: "incomplete",
|
||||
})
|
||||
require.NoError(t, err)
|
||||
assert.False(t, verdict.Accept)
|
||||
assert.Equal(t, "missing line references", verdict.Feedback)
|
||||
}
|
||||
|
||||
func TestVerifierErrorOnUnparsableOutput(t *testing.T) {
|
||||
dir := t.TempDir()
|
||||
script := filepath.Join(dir, "claude")
|
||||
require.NoError(t, os.WriteFile(script, []byte("#!/bin/sh\necho 'not json'\n"), 0755))
|
||||
|
||||
v := iexec.NewVerifier(script, "claude-sonnet-4-6", 5*time.Second)
|
||||
_, err := v.Verify(context.Background(), "rules", "task", iexec.Result{
|
||||
Status: "pass", Phase: "review", Skill: "review", Message: "ok",
|
||||
})
|
||||
assert.Error(t, err)
|
||||
}
|
||||
|
||||
func TestVerifierErrorOnNonZeroExit(t *testing.T) {
|
||||
dir := t.TempDir()
|
||||
script := filepath.Join(dir, "claude")
|
||||
require.NoError(t, os.WriteFile(script, []byte("#!/bin/sh\nexit 1\n"), 0755))
|
||||
|
||||
v := iexec.NewVerifier(script, "claude-sonnet-4-6", 5*time.Second)
|
||||
_, err := v.Verify(context.Background(), "rules", "task", iexec.Result{
|
||||
Status: "pass", Phase: "review", Skill: "review", Message: "ok",
|
||||
})
|
||||
assert.Error(t, err)
|
||||
}
|
||||
Reference in New Issue
Block a user