From 56101b46e89996e1f0c655f203c466bb6a2f8e8e Mon Sep 17 00:00:00 2001
From: Autumn-27 <53844290+Autumn-27@users.noreply.github.com>
Date: Thu, 20 Aug 2026 09:15:53 -0400
Subject: [PATCH] =?UTF-8?q?feat(worker):=20=E6=9A=82=E5=81=9C/=E9=87=8D?=
 =?UTF-8?q?=E8=B7=91=E7=BB=AD=E6=8E=A5=E5=85=88=E5=89=8D=20LLM=20=E5=AF=B9?=
 =?UTF-8?q?=E8=AF=9D=EF=BC=8C=E4=B8=8D=E5=86=8D=E4=BB=8E=E5=A4=B4=E5=BC=80?=
 =?UTF-8?q?=E5=A7=8B?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Worker 的 Execute 在创建 session 后尝试 Resume 先前的 transcript，
如果存在则续接对话而非重建全新会话，避免浪费已积累的 LLM 上下文。

Claude-Session: https://claude.ai/code/session_014TfZWeVGhjXqZD3vcj44mz
---
 agent/worker.go   | 12 ++++++++++++
 db/exploration.go | 10 +++++-----
 server/engine.go  |  3 ++-
 3 files changed, 19 insertions(+), 6 deletions(-)

diff --git a/agent/worker.go b/agent/worker.go
index 47e25e9..cdbe63d 100644
--- a/agent/worker.go
+++ b/agent/worker.go
@@ -360,6 +360,18 @@ func (w *Worker) Execute(ctx context.Context, name string, taskID int64, as *db.
 	s := agentcore.NewSession(opts)
 	defer s.Close() // release the session's background-task manager (temp dir + processes)
 
+	// Resume prior conversation if this intent was paused/blocked/exhausted and is
+	// being re-run. The transcript ID is deterministic per intent, so if a prior
+	// session exists the worker continues from where it left off instead of
+	// restarting from scratch.
+	if w.tx != nil {
+		_ = s.Resume(opts.SessionID)
+		if len(s.Messages()) > 0 {
+			seedUnlockFromHistory(s.Messages(), def.UnlockSkill)
+			input = "继续执行。"
+		}
+	}
+
 	// Budgets + settlement are owned by the SDK (MaxTurns/MaxDuration + Settlement):
 	// on hit it runs a wrap-up turn and finishes with ReasonMaxTurns/ReasonTimeout.
 	// We only add a HARD ctx backstop past the soft wall-clock budget, for the rare
diff --git a/db/exploration.go b/db/exploration.go
index be01753..dbcdb35 100644
--- a/db/exploration.go
+++ b/db/exploration.go
@@ -215,7 +215,7 @@ func (s *ExplorationStore) SetNodeState(id int64, state string) error {
 // ResetRunningIntents returns any intent left in 'running' back to 'open' so it
 // is re-claimed. Called on startup: a 'running' intent with no live worker (a
 // backend restart or crashed worker goroutine left it stuck) would otherwise spin
-// forever in the UI. Workers re-run an intent from scratch, so reopening is safe.
+// forever in the UI. Workers resume from their transcript, so reopening is safe.
 func (s *ExplorationStore) ResetRunningIntents() (int64, error) {
 	res, err := s.db.Exec(`UPDATE exploration_nodes SET state='open', completed_at=NULL, blocked_reason=NULL
 WHERE exploration_id=$1 AND kind='intent' AND state='running'`, s.expID)
@@ -227,9 +227,9 @@ WHERE exploration_id=$1 AND kind='intent' AND state='running'`, s.expID)
 }
 
 // ReopenIntent flips ONE not-successfully-finished intent (blocked/exhausted/stopped)
-// back to 'open' so a worker re-claims and re-runs it from scratch (graph writes it
-// already produced stay). done/open/running are left untouched. Returns whether a row
-// changed (false = intent absent or not in a rerunnable state). Used by the "重跑" button.
+// back to 'open' so a worker re-claims it (graph writes it already produced stay).
+// The worker resumes from its prior LLM transcript instead of restarting from scratch.
+// done/open/running are left untouched. Returns whether a row changed. Used by "重跑".
 func (s *ExplorationStore) ReopenIntent(id int64) (bool, error) {
 	res, err := s.db.Exec(`UPDATE exploration_nodes
 SET state='open', completed_at=NULL, blocked_reason=NULL
@@ -244,7 +244,7 @@ WHERE id=$1 AND exploration_id=$2 AND kind='intent'
 
 // ReopenBlockedIntents flips EVERY 'blocked' intent in this exploration back to 'open'
 // (batch rerun after e.g. an LLM/network outage that blocked many at once). Returns the
-// number reopened. Workers re-run each from scratch; kept graph writes remain.
+// number reopened. Workers resume from their transcripts; kept graph writes remain.
 func (s *ExplorationStore) ReopenBlockedIntents() (int64, error) {
 	res, err := s.db.Exec(`UPDATE exploration_nodes SET state='open', completed_at=NULL, blocked_reason=NULL
 WHERE exploration_id=$1 AND kind='intent' AND state='blocked'`, s.expID)
diff --git a/server/engine.go b/server/engine.go
index c52e76b..4606a8f 100644
--- a/server/engine.go
+++ b/server/engine.go
@@ -926,7 +926,8 @@ func (e *Engine) runWorkerStep(ctx context.Context, t *Task, name string, worker
 		return true
 	}
 	// if a pause cancelled this run mid-flight, return the intent to the frontier
-	// so it is re-claimed (and re-run from scratch) on resume — not marked done.
+	// so it is re-claimed on resume — the worker will resume the prior LLM
+	// conversation from its transcript instead of restarting from scratch.
 	if ectx.Err() != nil && e.IsPaused(t.ID) {
 		_ = t.Store.SetIntentState(intent.ID, "open")
 		return true