OpenAI-compatible chat client for the LiteLLM proxy (base_url normalization,
Bearer auth, retry/backoff on 429/5xx/transport, usage accounting) - stdlib
http only. Intake lists issues in scope from /issues.json with the task
class read from a configurable custom field. Writeback lands each turn as
REPORT-<vertical>-<task>-<ts>.md plus REPORT-latest.md (atomic rename) with
model/tier/token telemetry. The bash tool ports maki's permission semantics
without tree-sitter: segment-by-segment compound-command checks, deny beats
allow, word-boundary "cmd *" matching, $()/backtick/subshell denied, output
truncation, per-command timeout with process-group cleanup.
💘 Generated with Crush
Assisted-by: Crush:glm-5.2
242 lines
7.3 KiB
Go
242 lines
7.3 KiB
Go
package tools
|
|
|
|
import (
|
|
"context"
|
|
"encoding/json"
|
|
"errors"
|
|
"strings"
|
|
"testing"
|
|
"time"
|
|
)
|
|
|
|
func TestScopeMatches(t *testing.T) {
|
|
cases := []struct {
|
|
pattern string
|
|
scope string
|
|
want bool
|
|
}{
|
|
// "cmd *" word-boundary prefix (maki semantics).
|
|
{"pwd *", "pwd", true},
|
|
{"pwd *", "pwd -L", true},
|
|
{"pwd *", "pwdx", false},
|
|
{"pwd *", "pwd secret", true},
|
|
{"git diff *", "git diff", true},
|
|
{"git diff *", "git diff HEAD~1", true},
|
|
{"git diff *", "git difftool", false},
|
|
// Universal wildcards.
|
|
{"*", "anything at all", true},
|
|
{"**", "anything", false}, // not a universal form; exact match
|
|
// Bare "*" raw prefix.
|
|
{"pfx*", "pfxthing", true},
|
|
{"pfx*", "pfz", false},
|
|
// Exact.
|
|
{"go version", "go version", true},
|
|
{"go version", "go version -m", false},
|
|
// Path prefix.
|
|
{"/tmp/work/**", "/tmp/work", true},
|
|
{"/tmp/work/**", "/tmp/work/sub/f", true},
|
|
{"/tmp/work/**", "/tmp/workx/f", false},
|
|
{"/tmp/work/**", "/tmp/other", false},
|
|
}
|
|
for _, tc := range cases {
|
|
if got := ScopeMatches(tc.pattern, tc.scope); got != tc.want {
|
|
t.Errorf("ScopeMatches(%q, %q) = %v, want %v", tc.pattern, tc.scope, got, tc.want)
|
|
}
|
|
}
|
|
}
|
|
|
|
func TestSplitCommand(t *testing.T) {
|
|
cases := []struct {
|
|
cmd string
|
|
want []string
|
|
}{
|
|
{"ls -la", []string{"ls -la"}},
|
|
{"git diff && rm -rf /", []string{"git diff", "rm -rf /"}},
|
|
{"a || b ; c", []string{"a", "b", "c"}},
|
|
{"cat f | grep x", []string{"cat f", "grep x"}},
|
|
{`echo "a && b"`, []string{`echo "a && b"`}},
|
|
{`echo 'p | q'`, []string{`echo 'p | q'`}},
|
|
{" ", nil},
|
|
{"echo hi > out.txt", []string{"echo hi > out.txt"}},
|
|
}
|
|
for _, tc := range cases {
|
|
got := SplitCommand(tc.cmd)
|
|
if len(got) != len(tc.want) {
|
|
t.Errorf("SplitCommand(%q) = %v, want %v", tc.cmd, got, tc.want)
|
|
continue
|
|
}
|
|
for i := range got {
|
|
if got[i] != tc.want[i] {
|
|
t.Errorf("SplitCommand(%q)[%d] = %q, want %q", tc.cmd, i, got[i], tc.want[i])
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
func newTool(t *testing.T, allow, deny []string) *BashTool {
|
|
t.Helper()
|
|
b, err := NewBashTool(BashOptions{
|
|
WorkRoot: t.TempDir(),
|
|
Allow: allow,
|
|
Deny: deny,
|
|
Timeout: 5 * time.Second,
|
|
MaxOutputBytes: 100_000,
|
|
})
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
return b
|
|
}
|
|
|
|
func TestCheckGate(t *testing.T) {
|
|
cases := []struct {
|
|
name string
|
|
allow []string
|
|
deny []string
|
|
cmd string
|
|
ok bool
|
|
want string // substring of denial reason when !ok
|
|
}{
|
|
{"simple allow", []string{"echo *"}, nil, "echo hi", true, ""},
|
|
{"exact allow", []string{"pwd"}, nil, "pwd", true, ""},
|
|
{"uncovered", []string{"echo *"}, nil, "ls -la", false, "not covered by any allow rule"},
|
|
{"compound all allowed", []string{"git *"}, nil, "git diff && git status", true, ""},
|
|
{"compound one bad", []string{"git *"}, nil, "git diff && rm -rf /", false, `segment "rm -rf /"`},
|
|
{"deny beats allow", []string{"*"}, []string{"sudo *"}, "sudo id", false, `matches deny rule "sudo *"`},
|
|
{"deny beats allow compound", []string{"*"}, []string{"rm -rf *"}, "git status && rm -rf /", false, "deny rule"},
|
|
{"pipe splits", []string{"cat *", "grep *"}, nil, "cat f | grep x", true, ""},
|
|
{"pipe uncovered", []string{"cat *"}, nil, "cat f | grep x", false, `segment "grep x"`},
|
|
{"command substitution denied", []string{"*"}, nil, "echo $(rm -rf /)", false, "not decomposable"},
|
|
{"backticks denied", []string{"*"}, nil, "echo `id`", false, "not decomposable"},
|
|
{"subshell denied", []string{"*"}, nil, "(cd / && rm -rf /)", false, "not decomposable"},
|
|
{"empty", []string{"*"}, nil, " ", false, "empty command"},
|
|
{"word boundary", []string{"pwd *"}, nil, "pwdx", false, "not covered"},
|
|
{"redirect is part of scope", []string{"go *"}, nil, "go test > /tmp/out", true, ""},
|
|
}
|
|
for _, tc := range cases {
|
|
t.Run(tc.name, func(t *testing.T) {
|
|
b := newTool(t, tc.allow, tc.deny)
|
|
err := b.Check(tc.cmd)
|
|
if tc.ok {
|
|
if err != nil {
|
|
t.Fatalf("Check(%q) = %v, want nil", tc.cmd, err)
|
|
}
|
|
return
|
|
}
|
|
if !errors.Is(err, ErrDenied) {
|
|
t.Fatalf("Check(%q) = %v, want ErrDenied", tc.cmd, err)
|
|
}
|
|
if !strings.Contains(err.Error(), tc.want) {
|
|
t.Fatalf("denial %q does not contain %q", err, tc.want)
|
|
}
|
|
})
|
|
}
|
|
}
|
|
|
|
func TestDefaultAllow(t *testing.T) {
|
|
b, err := NewBashTool(BashOptions{
|
|
WorkRoot: t.TempDir(),
|
|
Allow: nil,
|
|
DefaultAllow: true,
|
|
})
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
if err := b.Check("anything --at --all"); err != nil {
|
|
t.Errorf("default allow should pass uncovered scopes: %v", err)
|
|
}
|
|
}
|
|
|
|
func TestRun(t *testing.T) {
|
|
t.Run("allowed echo", func(t *testing.T) {
|
|
b := newTool(t, []string{"echo *"}, nil)
|
|
out, err := b.Run(context.Background(), json.RawMessage(`{"command":"echo hello mopac"}`))
|
|
if err != nil {
|
|
t.Fatalf("Run: %v", err)
|
|
}
|
|
if !strings.Contains(out, "hello mopac") {
|
|
t.Errorf("output = %q", out)
|
|
}
|
|
})
|
|
t.Run("denied returns ErrDenied without executing", func(t *testing.T) {
|
|
b := newTool(t, []string{"echo *"}, []string{"sudo *"})
|
|
out, err := b.Run(context.Background(), json.RawMessage(`{"command":"sudo rm -rf /"}`))
|
|
if !errors.Is(err, ErrDenied) {
|
|
t.Fatalf("err = %v, want ErrDenied", err)
|
|
}
|
|
if out != "" {
|
|
t.Errorf("denied run must not produce output, got %q", out)
|
|
}
|
|
})
|
|
t.Run("nonzero exit reports output", func(t *testing.T) {
|
|
b := newTool(t, []string{"false"}, nil)
|
|
out, err := b.Run(context.Background(), json.RawMessage(`{"command":"false"}`))
|
|
if err == nil {
|
|
t.Fatal("expected error for exit 1")
|
|
}
|
|
if out != "" {
|
|
t.Errorf("no output expected, got %q", out)
|
|
}
|
|
})
|
|
t.Run("timeout kills command", func(t *testing.T) {
|
|
b, err := NewBashTool(BashOptions{
|
|
WorkRoot: t.TempDir(),
|
|
Allow: []string{"sleep *"},
|
|
Timeout: 300 * time.Millisecond,
|
|
})
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
start := time.Now()
|
|
_, err = b.Run(context.Background(), json.RawMessage(`{"command":"sleep 30"}`))
|
|
if err == nil || !strings.Contains(err.Error(), "timed out") {
|
|
t.Fatalf("err = %v, want timeout", err)
|
|
}
|
|
if elapsed := time.Since(start); elapsed > 5*time.Second {
|
|
t.Fatalf("timeout did not fire, took %s", elapsed)
|
|
}
|
|
})
|
|
t.Run("output truncation", func(t *testing.T) {
|
|
b, err := NewBashTool(BashOptions{
|
|
WorkRoot: t.TempDir(),
|
|
Allow: []string{"echo *"},
|
|
Timeout: 5 * time.Second,
|
|
MaxOutputBytes: 100,
|
|
})
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
out, err := b.Run(context.Background(), json.RawMessage(`{"command":"echo 012345678901234567890123456789012345678901456789012345678901234567890123456789012345678901234567890123456789"}`))
|
|
if err != nil {
|
|
t.Fatalf("Run: %v", err)
|
|
}
|
|
if !strings.Contains(out, "[truncated") {
|
|
t.Errorf("expected truncation marker in %q", out)
|
|
}
|
|
if len(out) > 200 {
|
|
t.Errorf("output not truncated: %d bytes", len(out))
|
|
}
|
|
})
|
|
t.Run("bad json args", func(t *testing.T) {
|
|
b := newTool(t, []string{"echo *"}, nil)
|
|
if _, err := b.Run(context.Background(), json.RawMessage(`{"command": 42}`)); err == nil {
|
|
t.Fatal("expected arg error")
|
|
}
|
|
})
|
|
}
|
|
|
|
func TestDefinitionSchema(t *testing.T) {
|
|
b := newTool(t, nil, nil)
|
|
def := b.Definition()
|
|
if def.Function.Name != "bash" || def.Type != "function" {
|
|
t.Errorf("definition = %+v", def)
|
|
}
|
|
if def.Function.Parameters["required"] == nil {
|
|
t.Error("parameters must mark command required")
|
|
}
|
|
// Sanity: the schema must marshal (it goes straight into requests).
|
|
if _, err := json.Marshal(def); err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
}
|