diff --git a/.graphlemon/runs/fe8e1b3d-dea1-4fa1-aa53-8cdd77d977f3/code_diff.patch b/.graphlemon/runs/fe8e1b3d-dea1-4fa1-aa53-8cdd77d977f3/code_diff.patch new file mode 100644 index 0000000..e49ae73 --- /dev/null +++ b/.graphlemon/runs/fe8e1b3d-dea1-4fa1-aa53-8cdd77d977f3/code_diff.patch @@ -0,0 +1,342 @@ +diff --git a/docs/reference/graph-engineering-1000-agents-5-steps.diff-with-graphlemon.md b/docs/reference/graph-engineering-1000-agents-5-steps.diff-with-graphlemon.md +new file mode 100644 +index 0000000..dfe4954 +--- /dev/null ++++ b/docs/reference/graph-engineering-1000-agents-5-steps.diff-with-graphlemon.md +@@ -0,0 +1,50 @@ ++# 对照笔记：《Graph Engineering：一个提示词，跑起 1000+ 的智能体循环》vs graphlemon 架构 ++ ++- **对照对象**：[graph-engineering-1000-agents-5-steps.md](./graph-engineering-1000-agents-5-steps.md)（本文简称"5 步版"） ++- **背景**：仓库里已有同主题的 [14 步版](../Graph%20Engineering%20with%20Claude_%2014-Step%20roadmap%20from%200%20to%20graph%20architect%20(Full%20Course).md)，`docs/graphlemon_architecture.md` 第 10 节已对 14 步版做过逐条映射，并在 10.1 节据此决策"Work Graph 执行层直接复用 Claude Code 原生动态工作流"。5 步版是同一方法论的精简版，本笔记只记录 5 步版带来的**新增信息**和**尚待跟进的开放项**，不重复 14 步版已覆盖的内容。 ++- **状态**：讨论草稿，待后续会话继续展开，非最终结论。 ++- **整理日期**：2026-08-04 ++ ++--- ++ ++## 1. 已经吻合、且已在代码里落地的部分 ++ ++| 5 步版观点 | graphlemon 现状 | 依据 | ++|---|---|---| ++| 验证器（Verifier）必须用**干净的独立上下文**，不能复用执行器的对话，否则"验证"只是自己给自己点头 | Validator 是独立的 Org Agent，独立 `claude -p` 调用，不共享 Coder 的会话上下文 | `docs/graphlemon_architecture.md` §6.3 根因溯源节点设计；`.claude/agents/validator.md` | ++| 并行智能体必须用 **git worktree** 隔离工作区，否则会互相踩（文中举 Bun 团队真实翻车案例） | 并行 Coder 节点强制走独立 worktree，self-target（自举）场景无条件强制 | `src/work_node_run.py::create_worktree` / `ensure_worktree_committed`；CLAUDE.md「Isolation」小节 | ++| 图需要**锚点（Anchors）**：真跑过的测试、基于证据的判定、不可被智能体自行调整的冻结规则 | Validator 只出 PASS / FAIL / INCONCLUSIVE 三态，且要求独立复核证据，不采信 Coder 自称"测试通过" | `docs/graphlemon_architecture.md` §9.1（qa profile 映射） | ++| Work Graph 层直接用 Claude Code 原生 `agent()`/`parallel()`/`pipeline()`，编排本身零 token | 已决策并做过实测验证 | `docs/graphlemon_architecture.md` §10.1；`spikes/a1-dynamic-workflows/` | ++ ++## 2. 5 步版提到、但 graphlemon 仍是"已知简化项"的部分 ++ ++这几条是本轮对照读出来的**待办候选**，还没有决定要不要做、什么时候做： ++ ++1. **loop-until-dry（连续 K 轮无新发现才停止）** ++ 现状：Org-level Loop 固定 3 轮上限，超过未收敛直接转 `blocked_pending_human`。 ++ 出处：`docs/sdd/04-P-流程与状态流转.md`（明确标注为"未来优化候选"）；`docs/sdd/06-验收与测试锚点.md`。 ++ 5 步版的场景（第 4 步"未知规模的发现"）建议连续两轮无新发现才停，比固定轮次更适合"扫描规模未知"的任务——这跟 Validator 未来如果要承担"未知规模缺陷扫描"类任务是同一个开放问题。 ++ ++2. **按节点分层选模型**（便宜模型跑重复/有界节点，贵模型做综合裁决） ++ 现状：`src/agent_profile.py` 里 Planner / Coder / Validator 三个 Org Agent 目前**全部固定** `claude-sonnet-5`，Work Graph 内部节点也没有差异化选型。 ++ 出处：`docs/graphlemon_architecture.md` §10 表格第 12 行早就标注"应补进第 7.5 节预算护栏"，至今未落地。 ++ ++3. **1000 智能体 / 16 并发的量级验证** ++ graphlemon 目前的 Work Graph 场景（自举开发自己）是个位数到十几个节点的规模，从未真正测过大规模 fan-out。5 步版给的"先 20 个探路，看单次运行花多少钱再放开"的做法，如果以后要做"审计整个代码库"这类任务，可以直接借用作为验证节奏。 ++ ++## 3. 一处架构定位上的差异（不是缺口，是设计选择） ++ ++5 步版第 5 步专门讲"什么时候不该用 Graph"（任务太小/太孤立、需要人紧盯每一步、还在探索阶段、步骤间真正强依赖 → 别上图，用一个 Loop 就够）。 ++ ++这个判断题在 graphlemon 里**不是留给每次任务临场判断的**，而是直接被架构的两层设计固化成了硬边界： ++ ++- **Org Graph**（Planner → Coder → Validator）人工固定设计，永远不是"动态生成的图"——恰好对应 5 步版说的"步骤间真的互相依赖 + 需要锚点 + 不该让智能体自己改架构"这一类场景。 ++- **Work Graph** 才是 5 步版描述的那种"动态生成、可扇形展开"的图，且作用域被限定在单次 Org Agent 调用内部（Planner 一次规划、Coder 一次实现、Validator 一次验收）。 ++ ++也就是说，graphlemon 把"要不要上图"做成了架构里的一条硬边界，而不是运行时的判断逻辑——这一点比 5 步版单纯给"判断标准"更进一步，值得在后续讨论里明确写进架构文档，作为对 5 步版方法论的一个延伸。 ++ ++## 4. 待讨论的开放问题（供下一轮会话展开） ++ ++- [ ] loop-until-dry 是否要在 Validator 的"未知规模缺陷扫描"场景里落地？优先级几何？ ++- [ ] 按节点分层选模型：Org Graph 三节点要不要差异化？Work Graph 内部节点（Planning Skill 生成的编排脚本）要不要在 Prompt 里显式要求按任务复杂度选模型？ ++- [ ] 是否需要在 `docs/graphlemon_architecture.md` 里补一节，显式写明"Org Graph 固定 / Work Graph 动态"这条边界与 5 步版"何时不该用 Graph"判断标准的对应关系？ +diff --git a/src/backend_server.py b/src/backend_server.py +index 8716596..2b442d4 100644 +--- a/src/backend_server.py ++++ b/src/backend_server.py +@@ -821,8 +821,19 @@ class _Handler(BaseHTTPRequestHandler): + self._send_json(200, result) + + def _send_file_download(self, file_path, node_exec_id): +- with open(file_path, "rb") as f: +- body = f.read() ++ # execution_log_file rows are created up front (work_node_run.py) but ++ # the backing file itself is only ever written by execution_log. ++ # append_log() -- a node that errored before its first log write (or ++ # a file removed out-of-band) leaves a row with no file on disk yet; ++ # that's a clean 404, not an uncaught FileNotFoundError bubbling up ++ # as a bare 500 (bug A2, same root cause as A1: see execution_log.py ++ # and work_graph_orchestrate.py's DU-01 notes). ++ try: ++ with open(file_path, "rb") as f: ++ body = f.read() ++ except FileNotFoundError: ++ self._send_json(404, {"error": "LOG_FILE_MISSING"}) ++ return + self.send_response(200) + self.send_header("Content-Type", "application/octet-stream") + self.send_header("Content-Disposition", 'attachment; filename="' + node_exec_id + '.log"') +diff --git a/src/test_backend_server.py b/src/test_backend_server.py +index daad261..c156be1 100644 +--- a/src/test_backend_server.py ++++ b/src/test_backend_server.py +@@ -1145,6 +1145,19 @@ class ExecutionLogRouteTest(unittest.TestCase): + self.assertEqual(status, 404) + self.assertEqual(json.loads(body)["error"], "LOG_NOT_FOUND") + ++ def test_download_when_backing_file_missing_returns_clean_404_not_uncaught_exception(self): ++ # Bug A2: execution_log_file row exists (inserted in setUp) but its ++ # file_path was never actually written to disk -- append_log() is ++ # what creates the file, and this test deliberately never calls it, ++ # matching the exact real-world state before DU-01's fix (a node ++ # that errored before its first log write, or any other row whose ++ # backing file went missing). _send_file_download() must catch this ++ # itself, not let a bare FileNotFoundError escape as an uncaught 500. ++ self.assertFalse(os.path.exists(os.path.join(self.tmp.name, "node-1.jsonl"))) ++ status, body, _ = self._get("/api/work-nodes/node-1/log?download=true") ++ self.assertEqual(status, 404) ++ self.assertEqual(json.loads(body)["error"], "LOG_FILE_MISSING") ++ + def test_purged_log_returns_410(self): + with self.server.conn_lock: + self.server.conn.execute("UPDATE execution_log_file SET status = 'purged' WHERE log_file_id = 'log-1'") +diff --git a/src/test_work_graph_orchestrate.py b/src/test_work_graph_orchestrate.py +index 31c724a..a9c4ceb 100644 +--- a/src/test_work_graph_orchestrate.py ++++ b/src/test_work_graph_orchestrate.py +@@ -7,10 +7,13 @@ spike -- this module's own job is orchestration, not re-proving that). + Run: python3 -m unittest src.test_work_graph_orchestrate -v + """ + ++import json ++import os ++import tempfile + import unittest + from unittest.mock import patch + +-from src import paths, schema, work_graph_execute, work_graph_orchestrate as wgo, work_node_run ++from src import execution_log, paths, schema, work_graph_execute, work_graph_orchestrate as wgo, work_node_run + + + class TopologicalOrderTest(unittest.TestCase): +@@ -64,6 +67,17 @@ class TopologicalOrderTest(unittest.TestCase): + + class RunGraphTestBase(unittest.TestCase): + def setUp(self): ++ # DU-01: run_fn now really calls execution_log.append_log(), which ++ # writes to paths.log_file_path() (default ~/.graphlemon/logs/...) -- ++ # isolate that under a throwaway tempdir so real (non-mocked) ++ # run_work_node/run_fn calls in this test module never touch the ++ # developer's actual home directory, same pattern as test_paths.py. ++ self.home_tmp = tempfile.TemporaryDirectory() ++ self.addCleanup(self.home_tmp.cleanup) ++ self.old_graphlemon_home = os.environ.get("GRAPHLEMON_HOME") ++ os.environ["GRAPHLEMON_HOME"] = self.home_tmp.name ++ self.addCleanup(self._restore_graphlemon_home) ++ + self.conn = schema.connect(":memory:") + schema.create_schema(self.conn) + self.conn.execute( +@@ -80,6 +94,12 @@ class RunGraphTestBase(unittest.TestCase): + ) + self.conn.commit() + ++ def _restore_graphlemon_home(self): ++ if self.old_graphlemon_home is None: ++ os.environ.pop("GRAPHLEMON_HOME", None) ++ else: ++ os.environ["GRAPHLEMON_HOME"] = self.old_graphlemon_home ++ + def tearDown(self): + self.conn.close() + +@@ -306,5 +326,83 @@ class SelfTargetForcesWorktreeTest(RunGraphTestBase): + self.assertTrue(mock_run.call_args.kwargs["needs_worktree"]) + + ++class RunFnLogsExecutionEventsTest(RunGraphTestBase): ++ """DU-01 (bug A1/A2 root cause): `_make_run_fn()`'s `run_fn` closure must ++ actually call `execution_log.append_log()` around the real ++ `execute_work_graph()` call, so `execution_log_file`'s backing file on ++ disk genuinely has content -- not just a DB row tracking its lifecycle.""" ++ ++ def _node_exec_id(self, node_id="brd"): ++ return self.conn.execute( ++ "SELECT node_exec_id FROM work_node_execution WHERE run_id = 'run-1' AND node_id = ?", (node_id,) ++ ).fetchone()[0] ++ ++ def _log_lines(self, node_exec_id): ++ content = execution_log.get_log(self.conn, node_exec_id)["content"] ++ return [json.loads(line) for line in content.splitlines() if line] ++ ++ def test_success_writes_node_start_then_node_completed(self): ++ manifest = {"nodes": [{"node_id": "brd", "depends_on": []}], "edges": []} ++ with patch.object(work_graph_execute, "write_workflow_script"), \ ++ patch.object(work_graph_execute, "execute_work_graph", return_value={"content": "brd text"}): ++ wgo.run_graph( ++ self.conn, "run-1", "planner", "planner-doc-writing", "/tmp/proj-1", manifest, ++ lambda node, outputs: "p", _IN_SCHEMA, _OUT_SCHEMA, ++ ) ++ ++ events = self._log_lines(self._node_exec_id()) ++ self.assertEqual([e["event"] for e in events], ["node_start", "node_completed"]) ++ self.assertEqual(events[1]["output"], {"content": "brd text"}) ++ ++ # also exercise the real HTTP-facing read path end to end (not just ++ # the module function) -- an empty file was exactly bug A1. ++ row = self.conn.execute( ++ "SELECT size_bytes FROM execution_log_file WHERE node_exec_id = ?", (self._node_exec_id(),) ++ ).fetchone() ++ self.assertGreater(row[0], 0) ++ ++ def test_classifier_blocked_writes_node_failed_and_still_propagates(self): ++ manifest = {"nodes": [{"node_id": "brd", "depends_on": []}], "edges": []} ++ with patch.object(work_graph_execute, "write_workflow_script"), \ ++ patch.object( ++ work_graph_execute, "execute_work_graph", ++ side_effect=work_graph_execute.ClassifierBlockedError("blocked by classifier"), ++ ): ++ with self.assertRaises(work_graph_execute.ClassifierBlockedError): ++ wgo.run_graph( ++ self.conn, "run-1", "planner", "planner-doc-writing", "/tmp/proj-1", manifest, ++ lambda node, outputs: "p", _IN_SCHEMA, _OUT_SCHEMA, ++ ) ++ ++ events = self._log_lines(self._node_exec_id()) ++ self.assertEqual([e["event"] for e in events], ["node_start", "node_failed"]) ++ self.assertIn("blocked by classifier", events[1]["error"]) ++ ++ def test_usage_limit_writes_quota_limit_then_retries(self): ++ manifest = {"nodes": [{"node_id": "brd", "depends_on": []}], "edges": []} ++ attempts = {"n": 0} ++ ++ def fake_execute(cwd, name, args, output_schema, timeout=600): ++ attempts["n"] += 1 ++ if attempts["n"] == 1: ++ raise work_graph_execute.UsageLimitReachedError("hit your session limit", "session") ++ return {"content": "recovered"} ++ ++ with patch.object(work_graph_execute, "write_workflow_script"), \ ++ patch.object(work_graph_execute, "execute_work_graph", side_effect=fake_execute): ++ outputs = wgo.run_graph( ++ self.conn, "run-1", "planner", "planner-doc-writing", "/tmp/proj-1", manifest, ++ lambda node, outputs: "p", _IN_SCHEMA, _OUT_SCHEMA, ++ sleep=lambda s: None, now=lambda: 0.0, ++ ) ++ self.assertEqual(outputs["brd"], {"content": "recovered"}) ++ ++ events = self._log_lines(self._node_exec_id()) ++ self.assertEqual( ++ [e["event"] for e in events], ["node_start", "quota_limit", "node_start", "node_completed"] ++ ) ++ self.assertEqual(events[1]["quota_limit_type"], "session") ++ ++ + if __name__ == "__main__": + unittest.main() +diff --git a/src/work_graph_orchestrate.py b/src/work_graph_orchestrate.py +index fb0b95b..d2562f5 100644 +--- a/src/work_graph_orchestrate.py ++++ b/src/work_graph_orchestrate.py +@@ -39,13 +39,42 @@ subgraph IS just itself), escalate to `work_graph_run.status='blocked'` + + raise if the regenerated content still fails. Not a general multi-node + re-planning mechanism -- that would be new, unscoped complexity 04-P itself + doesn't describe either. ++ ++**DU-01 (2026-08-04, bug A1/A2 root-cause fix)**: `_make_run_fn()`'s `run_fn` ++closure is the natural place to call `execution_log.append_log()` -- it's ++the one spot that both has `node_exec_id` and wraps the actual blocking ++`execute_work_graph()` call, and `work_node_run.run_work_node()` already ++guarantees an `execution_log_file` row (status='active') exists before ++`run_fn` is ever invoked. Before this, `append_log()` was fully implemented ++(U-013/U-027) but never called by any production code path, so ++`execution_log_file` rows existed with zero bytes of real content forever ++-- the console's "暂无日志内容" was a correct rendering of genuinely empty ++data, and downloading it crashed `backend_server._send_file_download()` ++with an uncaught `FileNotFoundError` (also fixed here, see that function). ++This is one blocking `agent()` call per node (03-C's own dynamic-workflow ++design, see this module's own docstring above) with no mid-flight ++progress to report, so logging is 4 discrete events around that one call, ++not a real character-by-character stream: `node_start` (before the call), ++`node_completed` (after a successful return), `node_failed` (any other ++exception, re-raised unchanged after logging), `quota_limit` (the ++`UsageLimitReachedError` case, logged before translating to ++`work_node_run`'s own `_UsageLimitSignal`). A node retried by ++`run_work_node`'s own retry loop calls `run_fn` again, appending another ++`node_start`/... pair per attempt -- that's an accurate log of what ++actually happened (N attempts), not a bug. + """ + + import json + import time + import uuid + +-from src import paths, work_graph_execute, work_node_run ++from src import execution_log, paths, work_graph_execute, work_node_run ++ ++ ++def _log_event(conn, node_exec_id, event_type, **payload): ++ execution_log.append_log( ++ conn, node_exec_id, json.dumps({"event": event_type, **payload}) + "\n" ++ ) + + + class ConsistencyCheckFailedError(Exception): +@@ -138,19 +167,28 @@ def _render_node_script(script_name, org_node, prompt, output_schema, model): + ) + + +-def _make_run_fn(target_folder_path, node_exec_id, org_node, prompt, output_schema, timeout): ++def _make_run_fn(conn, target_folder_path, node_exec_id, org_node, prompt, output_schema, timeout): + script_name = "graphlemon-node-" + node_exec_id.replace("-", "") + + def run_fn(cwd, model=None): ++ _log_event(conn, node_exec_id, "node_start", org_node=org_node, model=model) + script_content = _render_node_script(script_name, org_node, prompt, output_schema, model) + work_graph_execute.write_workflow_script(cwd, script_name, script_content) + try: +- return work_graph_execute.execute_work_graph(cwd, script_name, {}, output_schema, timeout=timeout) ++ output = work_graph_execute.execute_work_graph(cwd, script_name, {}, output_schema, timeout=timeout) + except work_graph_execute.UsageLimitReachedError as e: ++ _log_event(conn, node_exec_id, "quota_limit", quota_limit_type=e.quota_limit_type) + work_node_run.raise_usage_limit_signal(e.quota_limit_type) +- # ClassifierBlockedError / generic WorkGraphExecuteError propagate +- # as-is -- run_work_node's own `except Exception` clause marks the +- # node 'blocked' and re-raises, no special handling needed here. ++ except Exception as e: ++ # ClassifierBlockedError / generic WorkGraphExecuteError -- logged ++ # then propagated as-is; run_work_node's own `except Exception` ++ # clause marks the node 'blocked' and re-raises, no special ++ # handling needed here beyond recording what actually happened. ++ _log_event(conn, node_exec_id, "node_failed", error=str(e)) ++ raise ++ else: ++ _log_event(conn, node_exec_id, "node_completed", output=output) ++ return output + + return run_fn + +@@ -215,7 +253,7 @@ def run_graph( + {"node_id": node_id}, + ) + prompt = build_prompt(node, outputs) +- run_fn = _make_run_fn(target_folder_path, node_exec_id, org_node, prompt, output_schema, timeout) ++ run_fn = _make_run_fn(conn, target_folder_path, node_exec_id, org_node, prompt, output_schema, timeout) + output = work_node_run.run_work_node( + conn, node_exec_id, {"node_id": node_id}, input_schema, output_schema, run_fn, + needs_worktree=effective_needs_worktree, repo_path=target_folder_path, +@@ -253,7 +291,9 @@ def _verify_and_maybe_regenerate( + prompt + "\n\nA consistency reviewer flagged issues with your previous draft: " + + "; ".join(feedback or []) + "\nProduce a corrected version resolving these issues." + ) +- reflect_run_fn = _make_run_fn(target_folder_path, reflect_node_exec_id, org_node, reflect_prompt, output_schema, timeout) ++ reflect_run_fn = _make_run_fn( ++ conn, target_folder_path, reflect_node_exec_id, org_node, reflect_prompt, output_schema, timeout ++ ) + regenerated = work_node_run.run_work_node( + conn, reflect_node_exec_id, {"node_id": node_id}, input_schema, output_schema, reflect_run_fn, + needs_worktree=needs_worktree, repo_path=target_folder_path,