agent/loop: send full compacted history instead of recent-only tail

After compaction, the loop was discarding cold+warm zones and only
sending the last 8 messages to the backend. This caused the agent to
lose context after manual or auto compaction.

Remove the recentHistory path entirely: state.history() already
returns the compacted (small) history post-compact. The TaskState
render block remains as additional steering injected after the system
prompt.

Removed: StepContext struct, recentHistory from state interface and
Session.

pkg/agent/loop.go

@@ -81,36 +81,18 @@ func run(ctx context.Context, cfg agentConfig, state state) error {
 	// sync.Map is required because execOne may run in concurrent goroutines.
 	var resultCache sync.Map
 
-	// stepCtx carries per-turn context assembled once at the top of each iteration.
-	type StepContext struct {
-		Goal      string
-		TaskState *TaskState
-		Recent    []backend.Message // last N messages (hot window)
-	}
-
 	for {
 		// Proactive context gate: strip cold material before calling the backend.
 		if budget := contextBudget(ctx, cfg.Backend); budget > 0 && state.estimateTokens() > budget {
 			state.stripCold()
 		}
 
-		sc := StepContext{
-			Goal:      cfg.preamble,
-			TaskState: state.taskState(),
-			Recent:    state.recentHistory(hotTailSize),
+		history := state.history()
+		if cfg.preamble != "" {
+			history = append([]backend.Message{{Role: "system", Content: cfg.preamble}}, history...)
 		}
-
-		var history []backend.Message
-		if sc.TaskState != nil {
-			history = sc.Recent
-		} else {
-			history = state.history()
-		}
-		if sc.Goal != "" {
-			history = append([]backend.Message{{Role: "system", Content: sc.Goal}}, history...)
-		}
-		if sc.TaskState != nil {
-			if msg := sc.TaskState.render(); msg != "" && len(history) > 0 {
+		if ts := state.taskState(); ts != nil {
+			if msg := ts.render(); msg != "" && len(history) > 0 {
 				// Insert after system prompt, before conversation.
 				history = append([]backend.Message{history[0], {Role: "user", Content: msg}}, history[1:]...)
 			}

pkg/agent/loop_maxsteps_helpers_test.go

@@ -48,13 +48,6 @@ func (s *simpleState) history() []backend.Message {
 	return s.msgs
 }
 
-func (s *simpleState) recentHistory(n int) []backend.Message {
-	if n >= len(s.msgs) {
-		return s.msgs
-	}
-	return s.msgs[len(s.msgs)-n:]
-}
-
 func (s *simpleState) taskState() *TaskState { return nil }
 
 func (s *simpleState) updateTaskState(TaskState) {}

pkg/agent/session.go

@@ -158,13 +158,6 @@ func (s *Session) history() []backend.Message {
 	return s.messages
 }
 
-func (s *Session) recentHistory(n int) []backend.Message {
-	if n >= len(s.messages) {
-		return s.messages
-	}
-	return s.messages[len(s.messages)-n:]
-}
-
 func (s *Session) taskState() *TaskState {
 	return s.TaskState
 }

pkg/agent/state.go

@@ -27,9 +27,6 @@ type state interface {
 	// The first entry is always the initial user message containing the goal.
 	history() []backend.Message
 
-	// recentHistory returns the last n messages from the history.
-	recentHistory(n int) []backend.Message
-
 	// taskState returns the current structured task state, or nil if not set.
 	taskState() *TaskState