codebee 0.1.18 → 0.1.19
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +15 -0
- package/README.md +4 -3
- package/app/core/automation.py +19 -0
- package/app/core/backup.py +470 -0
- package/app/core/bookmeta.py +4 -1
- package/app/core/builtin_agent.py +33 -1
- package/app/core/cleanup.py +303 -0
- package/app/core/flows.py +1 -1
- package/app/core/jobs.py +22 -4
- package/app/core/knowledge.py +26 -4
- package/app/core/manager.py +25 -1
- package/app/core/market.py +14 -2
- package/app/core/modelhub.py +16 -0
- package/app/core/pipeline.py +50 -8
- package/app/core/planner.py +22 -4
- package/app/core/publish/manager.py +80 -3
- package/app/core/runner.py +127 -7
- package/app/core/selfupdate.py +80 -38
- package/app/core/settings.py +32 -1
- package/app/core/skill_scan.py +86 -0
- package/app/core/store.py +62 -8
- package/app/core/wxdigest.py +710 -0
- package/app/core/zentao.py +516 -50
- package/app/main.py +263 -1
- package/app/pet.py +1323 -0
- package/app/pet_bee.png +0 -0
- package/app/pet_bee_robot.png +0 -0
- package/app/pick_dialog.py +34 -2
- package/app/ui/app.js +11063 -10389
- package/app/ui/i18n.js +169 -9
- package/app/ui/index.html +132 -2
- package/app/ui/style.css +156 -2
- package/package.json +1 -1
|
@@ -0,0 +1,303 @@
|
|
|
1
|
+
# -*- coding: utf-8 -*-
|
|
2
|
+
"""每日垃圾清理:只清 CodeBee 自己产生的可再生数据,绝不碰用户成果。
|
|
3
|
+
|
|
4
|
+
垃圾来源(实测 data/ 923MB 构成):
|
|
5
|
+
- runs/<id>/steps/*.log 运行过程日志,单文件可达 22MB(保留 run.json
|
|
6
|
+
与 report.md,运行记录/结果页签不受影响)
|
|
7
|
+
- publish/shots/ 发布过程截图证据
|
|
8
|
+
- data 根 *.bak* 配置改写工具留下的备份残留
|
|
9
|
+
- pending/ 运行中指挥信箱的过期附件
|
|
10
|
+
- publish/profiles/ 浏览器 profile 迁移残留(新版已搬家目录,
|
|
11
|
+
旧位置 688MB 实测;仅当家目录同款平台目录
|
|
12
|
+
存在——即迁移已完成——才按废弃清理)
|
|
13
|
+
- exports/ imports/ 备份模块自己的历史产物
|
|
14
|
+
- data 根 *.log 服务日志按体量截尾,不是按天删
|
|
15
|
+
|
|
16
|
+
调度挂在 automation._tick(fire_due 自节流:启用 + 当天没清过才真跑)。
|
|
17
|
+
配置在 data/settings.json(cleanup_enabled / cleanup_retention_days),
|
|
18
|
+
状态在 data/cleanup.json(last_run/last_freed/history 供设置页展示)。
|
|
19
|
+
|
|
20
|
+
发布浏览器 profile(家目录 ~/.codebee/publish_profiles,约 1GB)含平台
|
|
21
|
+
登录态,清了要重新扫码——绝不进每日自动清理,只留手动入口
|
|
22
|
+
(run_cleanup(include_profiles=True))。
|
|
23
|
+
"""
|
|
24
|
+
from __future__ import annotations
|
|
25
|
+
|
|
26
|
+
import json
|
|
27
|
+
import os
|
|
28
|
+
import time
|
|
29
|
+
from pathlib import Path
|
|
30
|
+
|
|
31
|
+
from . import paths, settings as settings_mod
|
|
32
|
+
|
|
33
|
+
DEFAULT_RETENTION_DAYS = 14
|
|
34
|
+
LOG_MAX_BYTES = 5 * 1024 * 1024 # 服务日志超过这个体量才截
|
|
35
|
+
LOG_KEEP_BYTES = 1 * 1024 * 1024 # 截尾保留的末段
|
|
36
|
+
STATE_NAME = "cleanup.json"
|
|
37
|
+
|
|
38
|
+
|
|
39
|
+
# ---------------------------------------------------------------- 状态
|
|
40
|
+
|
|
41
|
+
def _state_path():
|
|
42
|
+
return Path(paths.DATA_DIR) / STATE_NAME
|
|
43
|
+
|
|
44
|
+
|
|
45
|
+
def _load_state():
|
|
46
|
+
try:
|
|
47
|
+
d = json.loads(_state_path().read_text(encoding="utf-8"))
|
|
48
|
+
return d if isinstance(d, dict) else {}
|
|
49
|
+
except Exception:
|
|
50
|
+
return {}
|
|
51
|
+
|
|
52
|
+
|
|
53
|
+
def _save_state(st):
|
|
54
|
+
p = _state_path()
|
|
55
|
+
p.parent.mkdir(parents=True, exist_ok=True)
|
|
56
|
+
tmp = p.with_suffix(".tmp")
|
|
57
|
+
tmp.write_text(json.dumps(st, ensure_ascii=False, indent=2), encoding="utf-8")
|
|
58
|
+
os.replace(str(tmp), str(p))
|
|
59
|
+
|
|
60
|
+
|
|
61
|
+
def _profiles_base(home=None):
|
|
62
|
+
"""家目录浏览器 profile 根(与 publish/manager 同款落点)。home 供测试注入。"""
|
|
63
|
+
if home:
|
|
64
|
+
return Path(home) / ".codebee" / "publish_profiles"
|
|
65
|
+
return Path.home() / ".codebee" / "publish_profiles"
|
|
66
|
+
|
|
67
|
+
|
|
68
|
+
def _parse_local_dt(s):
|
|
69
|
+
try:
|
|
70
|
+
return time.mktime(time.strptime(str(s)[:19], "%Y-%m-%d %H:%M:%S"))
|
|
71
|
+
except Exception:
|
|
72
|
+
return None
|
|
73
|
+
|
|
74
|
+
|
|
75
|
+
# ---------------------------------------------------------------- 扫描
|
|
76
|
+
|
|
77
|
+
def _run_ended(run_dir, cutoff_ts):
|
|
78
|
+
"""run 目录整体是否过期:看 run.json 的 ended_at;没有就退回目录 mtime。"""
|
|
79
|
+
rj = run_dir / "run.json"
|
|
80
|
+
if rj.is_file():
|
|
81
|
+
ts = None
|
|
82
|
+
try:
|
|
83
|
+
ts = _parse_local_dt(json.loads(rj.read_text(encoding="utf-8"))
|
|
84
|
+
.get("ended_at"))
|
|
85
|
+
except Exception:
|
|
86
|
+
ts = None
|
|
87
|
+
if ts:
|
|
88
|
+
return ts < cutoff_ts
|
|
89
|
+
try:
|
|
90
|
+
return run_dir.stat().st_mtime < cutoff_ts
|
|
91
|
+
except OSError:
|
|
92
|
+
return False
|
|
93
|
+
|
|
94
|
+
|
|
95
|
+
def _plan_items(retention_days=None, include_profiles=False, home=None, now=None):
|
|
96
|
+
"""扫描各类垃圾,返回条目列表。每条:key/label/note/bytes/count/files,
|
|
97
|
+
files 元素为 {path, op},op ∈ delete | rmtree | truncate。"""
|
|
98
|
+
now = now or time.time()
|
|
99
|
+
try:
|
|
100
|
+
days = int(retention_days if retention_days is not None
|
|
101
|
+
else settings_mod.load().get("cleanup_retention_days"))
|
|
102
|
+
except (TypeError, ValueError):
|
|
103
|
+
days = DEFAULT_RETENTION_DAYS
|
|
104
|
+
days = max(1, min(365, days))
|
|
105
|
+
cutoff = now - days * 86400
|
|
106
|
+
data_dir = Path(paths.DATA_DIR)
|
|
107
|
+
items = []
|
|
108
|
+
|
|
109
|
+
def add(key, label, note, files):
|
|
110
|
+
size = 0
|
|
111
|
+
good = []
|
|
112
|
+
for f in files:
|
|
113
|
+
try:
|
|
114
|
+
p = Path(f["path"])
|
|
115
|
+
if f["op"] == "truncate":
|
|
116
|
+
sz = p.stat().st_size - LOG_KEEP_BYTES
|
|
117
|
+
elif p.is_dir():
|
|
118
|
+
sz = sum(x.stat().st_size for x in p.rglob("*") if x.is_file())
|
|
119
|
+
else:
|
|
120
|
+
sz = p.stat().st_size
|
|
121
|
+
except OSError:
|
|
122
|
+
continue
|
|
123
|
+
# truncate 只在有得砍时才留;delete/rmtree 即使 0 字节(空目录)也要清
|
|
124
|
+
if sz > 0 or f["op"] != "truncate":
|
|
125
|
+
good.append(f)
|
|
126
|
+
size += max(0, sz)
|
|
127
|
+
if good:
|
|
128
|
+
items.append({"key": key, "label": label, "note": note,
|
|
129
|
+
"bytes": size, "count": len(good), "files": good})
|
|
130
|
+
|
|
131
|
+
# 1) 过期运行的过程日志(保留 run.json / report.md / 封面)
|
|
132
|
+
runs_dir = data_dir / "runs"
|
|
133
|
+
files = []
|
|
134
|
+
if runs_dir.is_dir():
|
|
135
|
+
for rd in runs_dir.iterdir():
|
|
136
|
+
if not rd.is_dir() or not _run_ended(rd, cutoff):
|
|
137
|
+
continue
|
|
138
|
+
steps = rd / "steps"
|
|
139
|
+
if steps.is_dir():
|
|
140
|
+
files.append({"path": str(steps), "op": "rmtree"})
|
|
141
|
+
add("runs_old_logs", "运行过程日志",
|
|
142
|
+
"保留运行记录与结果报告,只删过程中的原始日志", files)
|
|
143
|
+
|
|
144
|
+
# 2) 发布截图
|
|
145
|
+
shots = data_dir / "publish" / "shots"
|
|
146
|
+
add("publish_shots", "发布截图", "发布过程留档截图,超期即清",
|
|
147
|
+
[{"path": str(p), "op": "delete"}
|
|
148
|
+
for p in (shots.iterdir() if shots.is_dir() else [])
|
|
149
|
+
if p.is_file() and p.stat().st_mtime < cutoff])
|
|
150
|
+
|
|
151
|
+
# 3) 配置备份残留 *.bak*
|
|
152
|
+
add("bak_files", "配置备份残留", "改配置时自动留的 .bak 副本,超期即清",
|
|
153
|
+
[{"path": str(p), "op": "delete"}
|
|
154
|
+
for p in data_dir.glob("*.bak*")
|
|
155
|
+
if p.is_file() and p.stat().st_mtime < cutoff])
|
|
156
|
+
|
|
157
|
+
# 4) 过期指挥信箱(运行早已结束,附件永远不会被消费)
|
|
158
|
+
pend = data_dir / "pending"
|
|
159
|
+
files = []
|
|
160
|
+
for p in (pend.iterdir() if pend.is_dir() else []):
|
|
161
|
+
try:
|
|
162
|
+
if p.stat().st_mtime >= cutoff:
|
|
163
|
+
continue
|
|
164
|
+
except OSError:
|
|
165
|
+
continue
|
|
166
|
+
files.append({"path": str(p), "op": "rmtree" if p.is_dir() else "delete"})
|
|
167
|
+
add("pending_stale", "过期指挥信箱", "已结束任务的运行中留言附件",
|
|
168
|
+
files)
|
|
169
|
+
|
|
170
|
+
# 5) 浏览器 profile 迁移残留:家目录同款平台目录存在(迁移已完成)才清
|
|
171
|
+
legacy = data_dir / "publish" / "profiles"
|
|
172
|
+
files = []
|
|
173
|
+
if legacy.is_dir():
|
|
174
|
+
for p in legacy.iterdir():
|
|
175
|
+
if not p.is_dir():
|
|
176
|
+
continue
|
|
177
|
+
try:
|
|
178
|
+
if (_profiles_base(home) / p.name).exists():
|
|
179
|
+
files.append({"path": str(p), "op": "rmtree"})
|
|
180
|
+
except OSError:
|
|
181
|
+
continue
|
|
182
|
+
add("legacy_profiles", "旧版浏览器缓存残留",
|
|
183
|
+
"发布浏览器 profile 已迁往用户主目录,旧位置属废弃残留", files)
|
|
184
|
+
|
|
185
|
+
# 6) 历史备份包(导出 zip / 导入前自动备份,超期即清)
|
|
186
|
+
files = []
|
|
187
|
+
for d in ("exports", "imports"):
|
|
188
|
+
base = data_dir / d
|
|
189
|
+
for p in (base.iterdir() if base.is_dir() else []):
|
|
190
|
+
try:
|
|
191
|
+
if p.stat().st_mtime < cutoff:
|
|
192
|
+
files.append({"path": str(p),
|
|
193
|
+
"op": "rmtree" if p.is_dir() else "delete"})
|
|
194
|
+
except OSError:
|
|
195
|
+
continue
|
|
196
|
+
add("backups_old", "历史备份包", "导出与导入前自动备份的旧 zip", files)
|
|
197
|
+
|
|
198
|
+
# 7) 服务日志截尾(不删文件,砍掉前段)
|
|
199
|
+
files = [{"path": str(p), "op": "truncate"}
|
|
200
|
+
for p in data_dir.glob("*.log")
|
|
201
|
+
if p.is_file() and p.stat().st_size > LOG_MAX_BYTES]
|
|
202
|
+
add("logs_truncate", "服务日志瘦身", "只保留每个日志的末尾 1MB", files)
|
|
203
|
+
|
|
204
|
+
# 8) 手动:发布浏览器缓存(含登录态,清了要重新扫码;绝不自动清)
|
|
205
|
+
if include_profiles:
|
|
206
|
+
base = _profiles_base(home)
|
|
207
|
+
files = [{"path": str(p), "op": "rmtree"}
|
|
208
|
+
for p in (base.iterdir() if base.is_dir() else []) if p.is_dir()]
|
|
209
|
+
add("publish_profiles", "发布浏览器缓存",
|
|
210
|
+
"含平台登录态,清理后发布时需重新扫码登录", files)
|
|
211
|
+
return {"retention_days": days, "items": items,
|
|
212
|
+
"total_bytes": sum(i["bytes"] for i in items)}
|
|
213
|
+
|
|
214
|
+
|
|
215
|
+
def plan(retention_days=None, include_profiles=False, home=None):
|
|
216
|
+
return _plan_items(retention_days=retention_days,
|
|
217
|
+
include_profiles=include_profiles, home=home)
|
|
218
|
+
|
|
219
|
+
|
|
220
|
+
def status():
|
|
221
|
+
"""设置页一次拉全:配置 + 上次清理状态 + 当前可清理预估。"""
|
|
222
|
+
cfg = settings_mod.load()
|
|
223
|
+
return {"config": {"enabled": bool(cfg.get("cleanup_enabled")),
|
|
224
|
+
"retention_days": cfg.get("cleanup_retention_days")},
|
|
225
|
+
"state": _load_state(), "plan": plan()}
|
|
226
|
+
|
|
227
|
+
|
|
228
|
+
# ---------------------------------------------------------------- 执行
|
|
229
|
+
|
|
230
|
+
def _delete_path(p: Path, op):
|
|
231
|
+
if op == "rmtree":
|
|
232
|
+
import shutil
|
|
233
|
+
shutil.rmtree(str(p), ignore_errors=False)
|
|
234
|
+
elif op == "truncate":
|
|
235
|
+
with open(str(p), "rb") as f:
|
|
236
|
+
f.seek(-LOG_KEEP_BYTES, 2)
|
|
237
|
+
tail = f.read()
|
|
238
|
+
tmp = p.with_suffix(".tmp-cl")
|
|
239
|
+
tmp.write_bytes(tail)
|
|
240
|
+
os.replace(str(tmp), str(p))
|
|
241
|
+
else:
|
|
242
|
+
p.unlink()
|
|
243
|
+
|
|
244
|
+
|
|
245
|
+
def run_cleanup(retention_days=None, include_profiles=False):
|
|
246
|
+
"""按当前扫描结果执行清理。返回汇总并落状态(供设置页展示)。"""
|
|
247
|
+
scan = _plan_items(retention_days=retention_days,
|
|
248
|
+
include_profiles=include_profiles)
|
|
249
|
+
freed = 0
|
|
250
|
+
done = []
|
|
251
|
+
errors = []
|
|
252
|
+
for item in scan["items"]:
|
|
253
|
+
got = 0
|
|
254
|
+
n = 0
|
|
255
|
+
for f in item["files"]:
|
|
256
|
+
try:
|
|
257
|
+
p = Path(f["path"])
|
|
258
|
+
if f["op"] == "truncate":
|
|
259
|
+
sz = p.stat().st_size - LOG_KEEP_BYTES
|
|
260
|
+
elif p.is_dir():
|
|
261
|
+
sz = sum(x.stat().st_size for x in p.rglob("*") if x.is_file())
|
|
262
|
+
else:
|
|
263
|
+
sz = p.stat().st_size
|
|
264
|
+
_delete_path(p, f["op"])
|
|
265
|
+
got += max(0, sz)
|
|
266
|
+
n += 1
|
|
267
|
+
except OSError as e:
|
|
268
|
+
errors.append("%s:%s" % (f["path"], e.strerror or e))
|
|
269
|
+
if n:
|
|
270
|
+
done.append({"key": item["key"], "label": item["label"],
|
|
271
|
+
"bytes": got, "count": n})
|
|
272
|
+
freed += got
|
|
273
|
+
result = {"ts": time.strftime("%Y-%m-%d %H:%M:%S"), "freed": freed,
|
|
274
|
+
"items": done, "errors": errors[:10],
|
|
275
|
+
"include_profiles": bool(include_profiles)}
|
|
276
|
+
st = _load_state()
|
|
277
|
+
st["last_run"] = result["ts"]
|
|
278
|
+
st["last_freed"] = freed
|
|
279
|
+
st["last_result"] = result
|
|
280
|
+
hist = st.get("history") or []
|
|
281
|
+
hist.insert(0, {"ts": result["ts"], "freed": freed})
|
|
282
|
+
st["history"] = hist[:30]
|
|
283
|
+
_save_state(st)
|
|
284
|
+
try:
|
|
285
|
+
from . import store
|
|
286
|
+
store.bump_state()
|
|
287
|
+
except Exception:
|
|
288
|
+
pass
|
|
289
|
+
return result
|
|
290
|
+
|
|
291
|
+
|
|
292
|
+
def fire_due():
|
|
293
|
+
"""automation._tick 每拍调用:启用且当天没清过才真跑,否则零开销返回。"""
|
|
294
|
+
try:
|
|
295
|
+
cfg = settings_mod.load()
|
|
296
|
+
except Exception:
|
|
297
|
+
return None
|
|
298
|
+
if not cfg.get("cleanup_enabled"):
|
|
299
|
+
return None
|
|
300
|
+
today = time.strftime("%Y-%m-%d")
|
|
301
|
+
if str(_load_state().get("last_run") or "").startswith(today):
|
|
302
|
+
return None
|
|
303
|
+
return run_cleanup()
|
package/app/core/flows.py
CHANGED
|
@@ -76,7 +76,7 @@ BUILTIN_FLOWS = [
|
|
|
76
76
|
"note": "起草 → 多维评审 → 修订循环 → 发布门禁"},
|
|
77
77
|
{"id": "rank_scan", "name": "扫榜选材", "icon": "i-chart", "engine": "direct", "builtin": True,
|
|
78
78
|
"goal_hint": "想写哪个方向(一句话,可留空默认分析总榜热门题材)",
|
|
79
|
-
"note": "
|
|
79
|
+
"note": "抓取七猫+番茄排行榜公开数据 → AI 提炼跨平台热门题材/人设/差异化切入点(快档直出报告)"},
|
|
80
80
|
{"id": "research", "name": "调研报告", "icon": "i-file-search", "engine": "review", "builtin": True,
|
|
81
81
|
"manuscript": "report.md",
|
|
82
82
|
"rubric": ["全面性", "深度", "论据可靠", "可读性", "结论质量"],
|
package/app/core/jobs.py
CHANGED
|
@@ -156,6 +156,18 @@ def cancel(run_id):
|
|
|
156
156
|
ev = CANCELS.get(run_id)
|
|
157
157
|
if ev:
|
|
158
158
|
ev.set()
|
|
159
|
+
# 强制终止语义:置位事件后立刻把 running 落成 cancelled 终态,
|
|
160
|
+
# UI 即刻翻牌,不等流水线走到下一个检查点(CLI 子进程树由取消
|
|
161
|
+
# 事件在 run_process 的 0.4s 轮询里秒杀;内置智能体生成中的 HTTP
|
|
162
|
+
# 由取消等待方放弃)。expected_status 守卫保证已正常结束的运行
|
|
163
|
+
# 不被误改;流水线随后照常收尾,终态幂等。
|
|
164
|
+
try:
|
|
165
|
+
from . import store
|
|
166
|
+
store.update_run(run_id, expected_status="running",
|
|
167
|
+
status="cancelled", ended_at=_now(),
|
|
168
|
+
error="用户主动取消")
|
|
169
|
+
except Exception:
|
|
170
|
+
pass
|
|
159
171
|
return True
|
|
160
172
|
# 事件不存在=任务还在队列里没被 worker 拿起:直接落终态(取消事件在
|
|
161
173
|
# worker 起跑时才创建,排队任务点取消会在这里漏掉——起跑后再杀一遍)。
|
|
@@ -576,12 +588,13 @@ def _do_mgmt(job, ev):
|
|
|
576
588
|
cost_usd=res.get("cost_usd", 0.0), tokens=res.get("tokens", 0))
|
|
577
589
|
else:
|
|
578
590
|
store.finish_step(run_id, step["n"], "failed", summary="未知操作 %s" % op)
|
|
579
|
-
# 以步骤状态汇总 run
|
|
591
|
+
# 以步骤状态汇总 run 状态;expected_status 守卫:用户已强制终止的运行
|
|
592
|
+
# 保持 cancelled,不被步骤汇总改写成 failed(被杀的步骤记的就是非 done)
|
|
580
593
|
run = store.get_run(run_id)
|
|
581
594
|
statuses = [s["status"] for s in (run.get("steps") if run else [])] or ["failed"]
|
|
582
595
|
final = "done" if all(s == "done" for s in statuses) else "failed"
|
|
583
596
|
suffix = "(AI 修复成功)" if ok and len((run.get("steps") if run else [])) > 1 else ""
|
|
584
|
-
store.update_run(run_id, status=final, ended_at=_now(),
|
|
597
|
+
store.update_run(run_id, expected_status="running", status=final, ended_at=_now(),
|
|
585
598
|
summary=("%s %s %s%s" % (entry.get("name"), op,
|
|
586
599
|
"完成" if final == "done" else "失败", suffix)))
|
|
587
600
|
|
|
@@ -600,7 +613,9 @@ def _do_selfupgrade(job):
|
|
|
600
613
|
summary="升级完成,点「重启」生效" if res["ok"]
|
|
601
614
|
else ("升级失败: " + res["error"][:300]),
|
|
602
615
|
exit_code=res.get("exit_code"))
|
|
603
|
-
|
|
616
|
+
# expected_status 守卫:用户已强制终止的运行保持 cancelled,不被改写
|
|
617
|
+
store.update_run(run_id, expected_status="running",
|
|
618
|
+
status="done" if res["ok"] else "failed", ended_at=_now(),
|
|
604
619
|
summary="CodeBee selfupgrade %s" % ("完成" if res["ok"] else "失败"))
|
|
605
620
|
|
|
606
621
|
|
|
@@ -619,7 +634,10 @@ def _ai_repair(run_id, entry, ev, failed_cmd, orig_log):
|
|
|
619
634
|
if agent is None or agent.get("mode") != "real":
|
|
620
635
|
return False # 无真实智能体可用,维持原失败
|
|
621
636
|
try:
|
|
622
|
-
|
|
637
|
+
# 日志尾巴喂进 AI 修复提示词:先洗终端噪声,否则 ANSI 转义/乱码墙会被
|
|
638
|
+
# 模型当成"日志原文"照抄进修复推理里
|
|
639
|
+
log_tail = runner.clean_cli_text(
|
|
640
|
+
runner.tail_decoded(orig_log.read_bytes(), 2000)) if orig_log.exists() else "(无输出)"
|
|
623
641
|
except Exception:
|
|
624
642
|
log_tail = "(日志不可读)"
|
|
625
643
|
import shutil
|
package/app/core/knowledge.py
CHANGED
|
@@ -6,8 +6,10 @@
|
|
|
6
6
|
- 知识(knowledge)记「已知是这样」——领域事实/平台规则/结论/方法论,
|
|
7
7
|
来源=run 产出材料(调研报告/文档)。两者输入源不同,提炼互不双写。
|
|
8
8
|
|
|
9
|
-
|
|
10
|
-
|
|
9
|
+
质量闸门:条目默认直接转正(approved)参与注入——用户拍板:人工把关太重,
|
|
10
|
+
提炼提示词里的「只保留有明确复用价值的」约束兜质量。status 字段保留 draft
|
|
11
|
+
枚举向后兼容,手动新建同样直接 approved;migrate_drafts_approved() 在启动时
|
|
12
|
+
把历史草稿一次性转正。
|
|
11
13
|
|
|
12
14
|
注入纪律(与教训反着来):教训是全量小注(top-8),知识默认不注入——
|
|
13
15
|
scope 命中且库里有 approved 条目才成块,按与任务目标的相关性取 top,
|
|
@@ -299,12 +301,15 @@ def learn_from_run(run_id):
|
|
|
299
301
|
"""运行结束后由编排者从产出材料提炼知识条目(草稿态)。返回写入条数。
|
|
300
302
|
|
|
301
303
|
知识没有便宜的兜底路径:无编排者/无产出/非真实运行(mock)一律静默跳过,
|
|
302
|
-
|
|
304
|
+
宁缺毋滥——教训库的规则兜底搬到这里只会制造垃圾知识。只有正常跑完(done)
|
|
305
|
+
的运行才提炼:失败/取消的运行产物是半成品,据此沉淀的知识会污染知识库。
|
|
303
306
|
"""
|
|
304
307
|
from . import store
|
|
305
308
|
run = store.get_run(run_id)
|
|
306
309
|
if not run:
|
|
307
310
|
return 0
|
|
311
|
+
if (run.get("status") or "") != "done":
|
|
312
|
+
return 0
|
|
308
313
|
task = store.get_task(run.get("task_id")) if run.get("task_id") else None
|
|
309
314
|
if not task:
|
|
310
315
|
return 0
|
|
@@ -339,7 +344,7 @@ def learn_from_run(run_id):
|
|
|
339
344
|
as_of = _now()[:10]
|
|
340
345
|
if upsert_entry(task.get("type") or "*", x["title"], x["body"],
|
|
341
346
|
tags=x.get("tags"), source=run_id,
|
|
342
|
-
as_of=as_of, status="
|
|
347
|
+
as_of=as_of, status="approved"):
|
|
343
348
|
n += 1
|
|
344
349
|
except Exception:
|
|
345
350
|
return n
|
|
@@ -356,6 +361,23 @@ def learn_async(run_id):
|
|
|
356
361
|
threading.Thread(target=_run, name="kb-learn", daemon=True).start()
|
|
357
362
|
|
|
358
363
|
|
|
364
|
+
def migrate_drafts_approved():
|
|
365
|
+
"""一次性迁移:草稿闸门退役(默认直接转正),把历史 draft 全部转正。
|
|
366
|
+
|
|
367
|
+
幂等——没有 draft 时零写入。返回转正条数。"""
|
|
368
|
+
with _LOCK:
|
|
369
|
+
data = _load()
|
|
370
|
+
items = data.get("entries") or []
|
|
371
|
+
dirty = [x for x in items if (x.get("status") or "draft") != "approved"]
|
|
372
|
+
if not dirty:
|
|
373
|
+
return 0
|
|
374
|
+
for x in dirty:
|
|
375
|
+
x["status"] = "approved"
|
|
376
|
+
x["updated_at"] = _now()
|
|
377
|
+
_save(data)
|
|
378
|
+
return len(dirty)
|
|
379
|
+
|
|
380
|
+
|
|
359
381
|
def view():
|
|
360
382
|
"""知识库总览(给 UI/API):条目 + 标签聚合 + 草稿数 + 账龄标记。"""
|
|
361
383
|
entries = list_entries()
|
package/app/core/manager.py
CHANGED
|
@@ -859,6 +859,21 @@ def _toml_top_set(text, key, value):
|
|
|
859
859
|
return eol.join(lines).rstrip("\r\n") + eol, True
|
|
860
860
|
|
|
861
861
|
|
|
862
|
+
_TEST_HOST_SUFFIXES = (".test", ".example", ".invalid", ".localhost")
|
|
863
|
+
|
|
864
|
+
|
|
865
|
+
def _is_dead_endpoint(base):
|
|
866
|
+
"""明显打不通的测试/保留地址(*.test、*.example、example.com 等)。
|
|
867
|
+
这类端点落进 config.toml 后 codex 每次请求都撞死墙——2026-09-20
|
|
868
|
+
orch/p1.test 残留反复劫持全局配置的事故教训:宁可不写,不写必死端点。"""
|
|
869
|
+
host = (base or "").split("://", 1)[-1].split("/", 1)[0].split("@")[-1]
|
|
870
|
+
host = host.split(":")[0].lower().rstrip(".")
|
|
871
|
+
if not host:
|
|
872
|
+
return True
|
|
873
|
+
return (host in ("example.com", "example.org", "example.net", "localhost")
|
|
874
|
+
or host.endswith(_TEST_HOST_SUFFIXES))
|
|
875
|
+
|
|
876
|
+
|
|
862
877
|
def _sync_codex_settings(entry, model, cp):
|
|
863
878
|
"""codex 专属:把绑定供应商与模型写进 ~/.codex/config.toml
|
|
864
879
|
([model_providers.orch] 段 + 顶层 model_provider/model)。
|
|
@@ -867,6 +882,8 @@ def _sync_codex_settings(entry, model, cp):
|
|
|
867
882
|
没有 config.toml 里的 provider 段,绑定模型根本无处可用;而 model 单写
|
|
868
883
|
不写 provider 会指到 codex 自带 openai 官方端点上(401)。与编排的
|
|
869
884
|
_codex_provider_args 同构,但落 config 文件。返回错误串或 None。"""
|
|
885
|
+
if _is_dead_endpoint(cp.get("base_url")):
|
|
886
|
+
return "供应商端点 %r 是测试/保留地址,拒绝写入 config.toml" % (cp.get("base_url"),)
|
|
870
887
|
path = _config_path(entry)
|
|
871
888
|
if not path:
|
|
872
889
|
return "codex 配置路径无效"
|
|
@@ -1459,6 +1476,12 @@ def sync_runtime_config(agent):
|
|
|
1459
1476
|
break
|
|
1460
1477
|
if entry is None or not (entry.get("config") or {}).get("path"):
|
|
1461
1478
|
return
|
|
1479
|
+
if entry["id"] in ("codex-cli", "codex"):
|
|
1480
|
+
return # codex 绝不在运行防线上落盘(2026-09-20 orch 劫持事故):
|
|
1481
|
+
# 编排步骤走 runner 的 -c 一次性注入,不依赖 config.toml;
|
|
1482
|
+
# 在这里写盘会把全局 ~/.codex/config.toml 的 model_provider
|
|
1483
|
+
# 顶掉(CC Switch / 用户手动选的供应商被劫持)。交互 TUI
|
|
1484
|
+
# 场景由 launch() 自行同步,且端点防线在 _sync_codex_settings。
|
|
1462
1485
|
from . import modelhub
|
|
1463
1486
|
binding = modelhub.resolve_binding(entry["id"]) or {}
|
|
1464
1487
|
model = (binding.get("model") or "").strip()
|
|
@@ -1631,7 +1654,8 @@ def run_mgmt_command(entry, op, cancel_event=None, log_path=None):
|
|
|
1631
1654
|
# 2026-09-18 dsh 案)。
|
|
1632
1655
|
refresh_update_async(entry)
|
|
1633
1656
|
return {"ok": res["ok"], "exit_code": res["exit_code"], "command": cmd,
|
|
1634
|
-
"error": "" if res["ok"] else (res["stderr"][-800:]
|
|
1657
|
+
"error": "" if res["ok"] else (runner.clean_cli_text(res["stderr"])[-800:]
|
|
1658
|
+
or "退出码 %s" % res["exit_code"])}
|
|
1635
1659
|
|
|
1636
1660
|
|
|
1637
1661
|
def refresh_update_async(entry):
|
package/app/core/market.py
CHANGED
|
@@ -218,9 +218,19 @@ def install_files(pack_id, name, files, extra=None):
|
|
|
218
218
|
"""通用安装入口:任意 {安装相对路径: 文本内容} 写进用户包目录并记账。
|
|
219
219
|
内置市场包(install)与外部目录插件(market_remote)共用这一条落地通道,
|
|
220
220
|
避让 / 安装标记 / market.json 记账 / 幂等纪律完全一致;extra 追加进记账记录
|
|
221
|
-
(如外部插件的来源与版本)。
|
|
221
|
+
(如外部插件的来源与版本)。
|
|
222
|
+
|
|
223
|
+
装前静态扫描(借鉴 NVIDIA SkillSpector):安装内容扫危险模式,风险行写进
|
|
224
|
+
返回值与记账——提示不拦阻(用户仍可装),但危险必须被看见。"""
|
|
222
225
|
if not files or not any(str(v).strip() for v in files.values()):
|
|
223
226
|
return None, "包内容为空: %s" % pack_id
|
|
227
|
+
# 装前扫描:全文件合并扫一遍(纯内存静态规则)
|
|
228
|
+
scan_note = ""
|
|
229
|
+
try:
|
|
230
|
+
from . import skill_scan
|
|
231
|
+
scan_note = skill_scan.scan_summary("\n".join(str(v) for v in files.values()))
|
|
232
|
+
except Exception:
|
|
233
|
+
scan_note = ""
|
|
224
234
|
with _LOCK:
|
|
225
235
|
udir = _user_pack_dir()
|
|
226
236
|
reg = _load_registry()
|
|
@@ -249,12 +259,14 @@ def install_files(pack_id, name, files, extra=None):
|
|
|
249
259
|
except OSError as e:
|
|
250
260
|
return None, "写入用户技能库失败: %s" % e
|
|
251
261
|
record = {"file": target, "files": written, "installed_at": _now()}
|
|
262
|
+
if scan_note:
|
|
263
|
+
record["scan"] = scan_note # 危险模式扫描结果随包记账
|
|
252
264
|
if extra:
|
|
253
265
|
record.update(extra)
|
|
254
266
|
installed[pack_id] = record
|
|
255
267
|
_save_registry(reg)
|
|
256
268
|
return {"ok": True, "id": pack_id, "name": name, "file": target,
|
|
257
|
-
"already": already}, None
|
|
269
|
+
"already": already, "scan": scan_note}, None
|
|
258
270
|
|
|
259
271
|
|
|
260
272
|
def remove(pack_id):
|
package/app/core/modelhub.py
CHANGED
|
@@ -835,6 +835,12 @@ def _is_codex_target(target):
|
|
|
835
835
|
return (target or "").strip().lower() in ("codex-cli", "codex", "codex-code")
|
|
836
836
|
|
|
837
837
|
|
|
838
|
+
def _is_aider_target(target):
|
|
839
|
+
"""aider 的绑定键与 orch.kind 同名(catalog id=aider);兜底前缀匹配防变体。"""
|
|
840
|
+
t = (target or "").strip().lower()
|
|
841
|
+
return t == "aider" or t.startswith("aider-")
|
|
842
|
+
|
|
843
|
+
|
|
838
844
|
def note_codex_wire_dead(provider_id, minutes=30):
|
|
839
845
|
"""codex 撞上 wire 不兼容的供应商 → 供应商级冷却(自动绕开的记账位)。
|
|
840
846
|
|
|
@@ -2037,8 +2043,18 @@ def _chain_entry_env(prov, model, target="", endpoint=None, key="", key_id="",
|
|
|
2037
2043
|
"ANTHROPIC_AUTH_TOKEN": use_key}
|
|
2038
2044
|
if model:
|
|
2039
2045
|
out["env"]["ANTHROPIC_MODEL"] = model
|
|
2046
|
+
if _is_aider_target(target):
|
|
2047
|
+
# aider(litellm)不认 AUTH_TOKEN,只认 ANTHROPIC_API_KEY——缺了它
|
|
2048
|
+
# 直接报 LLM Provider NOT provided(2026-09-20 连载评审 aider 全灭根因)。
|
|
2049
|
+
# claude 绝不注入:x-api-key 与 Bearer 两种鉴权头网关挑食,不能混。
|
|
2050
|
+
out["env"]["ANTHROPIC_API_KEY"] = use_key
|
|
2040
2051
|
else:
|
|
2041
2052
|
out["env"] = {"ORCH_API_KEY": use_key}
|
|
2053
|
+
if _is_aider_target(target):
|
|
2054
|
+
# litellm 的 openai 通道读这对 env;base 语义与 codex chat wire 一致
|
|
2055
|
+
# (调用方拼 /chat/completions)
|
|
2056
|
+
out["env"]["OPENAI_API_KEY"] = use_key
|
|
2057
|
+
out["env"]["OPENAI_API_BASE"] = base
|
|
2042
2058
|
wire_api = endpoint[2] if endpoint else prov.get("wire_api", "responses")
|
|
2043
2059
|
out["codex_provider"] = {
|
|
2044
2060
|
"name": "orch", "base_url": base,
|