compact: retain newest ~20% of context; refactor startup into internal/app

- Replace fixed 6-message compact retention with ratio-based logic:
  retain ~20% of compactable estimated tokens, with a 6-message floor
  and user-turn boundary alignment
- Add compactRetainedTokenBudget, compactRecentStart, compactTurnBoundary
  helpers in internal/agent/agent.go
- Add TestAgentCompactKeepsTwentyPercentRecentContext
- Move startup/session/repl/bootstrap logic from root into internal/app
  so main.go is a thin binary entry point (15 lines)
- Update /compact docs in README.md
This commit is contained in:
loveuer
2026-06-24 23:59:21 -07:00
parent 9abd5c3b5f
commit dabf5bfecc
9 changed files with 408 additions and 224 deletions
+45
View File
@@ -145,6 +145,9 @@ func TestAgentCompact(t *testing.T) {
if messages[len(messages)-1].Content != "recent-answer-2" {
t.Fatalf("last recent message = %#v", messages[len(messages)-1])
}
if messages[2].Content != "recent-user-0" {
t.Fatalf("first retained recent message = %#v", messages[2])
}
if usage := a.LastUsage(); usage == nil || usage.TotalTokens != 12 {
t.Fatalf("usage = %#v", usage)
}
@@ -217,6 +220,48 @@ func longConversation() []llm.Message {
return messages
}
func manyTurnConversation(turns int) []llm.Message {
messages := []llm.Message{{Role: llm.RoleSystem, Content: "system"}}
for i := 0; i < turns; i++ {
messages = append(messages,
llm.Message{Role: llm.RoleUser, Content: fmt.Sprintf("user-%02d", i)},
llm.Message{Role: llm.RoleAssistant, Content: fmt.Sprintf("assistant-%02d", i)},
)
}
return messages
}
func TestAgentCompactKeepsTwentyPercentRecentContext(t *testing.T) {
provider := &compactProvider{}
a := New(Options{Provider: provider, Model: "test", SystemPrompt: "system"})
a.SetMessages(manyTurnConversation(25))
if err := a.Compact(context.Background()); err != nil {
t.Fatal(err)
}
messages := a.Messages()
// 1 system + 1 summary + 10 retained recent messages (5 user/assistant turns)
if len(messages) != 12 {
t.Fatalf("messages = %d, want 12; contents: %v", len(messages), joinMessageContents(messages))
}
if provider.summaryCalls != 1 {
t.Fatalf("summaryCalls = %d, want 1", provider.summaryCalls)
}
if messages[2].Content != "user-20" {
t.Fatalf("first retained message = %#v, want user-20", messages[2])
}
if messages[len(messages)-1].Content != "assistant-24" {
t.Fatalf("last retained message = %#v, want assistant-24", messages[len(messages)-1])
}
if !strings.Contains(provider.lastSummary, "user-19") {
t.Fatalf("summary should contain user-19 (part of summarized 80%%): %q", provider.lastSummary)
}
if strings.Contains(provider.lastSummary, "user-20") {
t.Fatalf("summary should not contain user-20 (part of retained 20%%): %q", provider.lastSummary)
}
}
func joinMessageContents(messages []llm.Message) string {
var b strings.Builder
for _, msg := range messages {