323 lines
15 KiB
Python
323 lines
15 KiB
Python
|
#!/usr/bin/env python3
|
|||
|
|
# -*- coding: utf-8 -*-
|
|||
|
|
"""session-plan-check.py —— 「会话规划体检」＝ 开会话后的第一件事
|
|||
|
|
|
|||
|
|
用户口径（2026-10-02 原话逐字）：
|
|||
|
|
「这个会话和协作会话的技能包 运行的第一件事 ，就应该是检查清楚
|
|||
|
|
所有会话规划是否配置完整且生效，然后标记一个状态」
|
|||
|
|
|
|||
|
|
为什么要它（真因，不是想当然）
|
|||
|
|
──────────────────────────────
|
|||
|
|
2026-10-02 实测：本线的排期**都在册**（`[唤醒]-…` / `[跟进]-…` 两条 recurring、
|
|||
|
|
`model_is_thinking=1`、`cwds` 正确），**但三类会话一条活的都没有** ——
|
|||
|
|
"配置在册" ≠ "机制在生效"。同一天早些时候还发现过更隐蔽的：
|
|||
|
|
钩子注入文本指着 **09-28 就已合并退役**的技能名、常驻规则快照的生成脚本**写到幽灵目录**
|
|||
|
|
⇒ 全都是"看着有配置、其实没生效"。所以**开工第一件事**必须先把这个状态问清楚，
|
|||
|
|
而不是等它表现出来（那时已经卡住或被用户发现）。
|
|||
|
|
|
|||
|
|
检查六项（⛔ 每条都必须能真报出问题 —— 见 selftest 的红绿对照）：
|
|||
|
|
① 周期钟 每类是否有 recurring 排期（缺 ⇒ 没人推 / 没人收）
|
|||
|
|
② 钟会响吗 排期绑的模型能不能跑（`thinking=0` ＋ flash 系 ⇒ 服务端必拒，2026-10-02 实测）
|
|||
|
|
③ 归属同形 `cwds` 与工作区路径**逐字**比（错一字面 ⇒ 裂组且**自我强化**）
|
|||
|
|
④ 活会话 三类（唤醒 / 协作 / 跟进）当前有没有活的
|
|||
|
|
⑤ 投递在跑 协作程序（投递轮）在不在跑 ＋ 心跳新鲜度
|
|||
|
|
⑥ 死排期 `once` 且 `next_run_at` 为空 ⇒ 已消耗、**不会再触发**（静默消失的活）
|
|||
|
|
|
|||
|
|
产出（**标记一个状态**）：
|
|||
|
|
`<WS>/.workbuddy/collab/session-plan.json` —— 机器可读（看板 / 后续会话读它）
|
|||
|
|
＋ stdout 一行摘要（给人扫）
|
|||
|
|
|
|||
|
|
退出码：0 = 齐备且生效 ｜ 1 = 有缺口（⚠️ **只标记、不拦开工**）
|
|||
|
|
用法：python session-plan-check.py [--ws <工作区>] [--json] [--quiet]
|
|||
|
|
"""
|
|||
|
|
import io
|
|||
|
|
import json
|
|||
|
|
import os
|
|||
|
|
import re
|
|||
|
|
import sqlite3
|
|||
|
|
import sys
|
|||
|
|
import time
|
|||
|
|
|
|||
|
|
# ── 根目录外置（roots.env 由 install.py 生成；宿主 env 优先，本文件兜底） ──────
|
|||
|
|
def _sm_load_roots():
|
|||
|
|
here = os.path.dirname(os.path.abspath(__file__))
|
|||
|
|
for up in range(4):
|
|||
|
|
p = os.path.normpath(os.path.join(here, *([".."] * up), "roots.env"))
|
|||
|
|
if os.path.isfile(p):
|
|||
|
|
try:
|
|||
|
|
for ln in io.open(p, encoding="utf-8"):
|
|||
|
|
ln = ln.strip()
|
|||
|
|
if ln and not ln.startswith("#") and "=" in ln:
|
|||
|
|
k, v = ln.split("=", 1)
|
|||
|
|
os.environ.setdefault(k.strip(), v.strip())
|
|||
|
|
except Exception:
|
|||
|
|
pass
|
|||
|
|
return
|
|||
|
|
_sm_load_roots()
|
|||
|
|
|
|||
|
|
ARGV = sys.argv[1:]
|
|||
|
|
def _opt(name, default=None):
|
|||
|
|
if name in ARGV:
|
|||
|
|
i = ARGV.index(name)
|
|||
|
|
return ARGV[i + 1] if i + 1 < len(ARGV) else default
|
|||
|
|
return default
|
|||
|
|
|
|||
|
|
WS = (_opt("--ws") or os.environ.get("DSH_COLLAB_WS") or os.environ.get("DSH_WS_ROOT")
|
|||
|
|
or os.getcwd()).replace("\\", "/").rstrip("/")
|
|||
|
|
CFG = os.environ.get("CODEBUDDY_CONFIG_DIR") or os.path.join(os.path.expanduser("~"), ".workbuddy")
|
|||
|
|
DB = os.path.join(CFG, "workbuddy.db")
|
|||
|
|
OUT = WS + "/.workbuddy/collab/session-plan.json" # ⛔ 不用 os.path.join（Windows 会出反斜杠，混排）
|
|||
|
|
AS_JSON = "--json" in ARGV
|
|||
|
|
QUIET = "--quiet" in ARGV
|
|||
|
|
|
|||
|
|
# 四类会话的标题前缀（口径 ⇒ skills/session-mechanism · 用户 2026-10-01 定「四类」）
|
|||
|
|
ROLES = [("唤醒", "[唤醒]"), ("协作", "[协作]"), ("跟进", "[跟进]")]
|
|||
|
|
# ⚠️ 只有「唤醒 / 跟进」要求**周期钟** —— 它们是"机制在跑"的两台钟（推 / 收）。
|
|||
|
|
# **协作会话是按需创建的**（用户 2026-10-01 口径「随需求确定时创建」）⇒ 要求它有 recurring
|
|||
|
|
# 排期就是**假警报**（本条判据首版就是这么写错的：首次实跑报了「协作类周期钟缺失」，
|
|||
|
|
# 而那是设计如此）。判据必须跟口径同源，⛔ 不按"四类应当对称"想当然。
|
|||
|
|
ROLES_CLOCK = [("唤醒", "[唤醒]"), ("跟进", "[跟进]")]
|
|||
|
|
# flash 系模型**不支持关闭思考** ⇒ `model_is_thinking=0` 时服务端 -32603（`automation-request-refused`）
|
|||
|
|
FLASH_HINT = ("flash",)
|
|||
|
|
# 排期"已终结"的时长上限：`next_run_at` 距现在超过这么久 ⇒ 判为"钟停了"
|
|||
|
|
STALE_FUTURE_MS = 365 * 24 * 3600 * 1000
|
|||
|
|
|
|||
|
|
CHECKS = [] # [{id, level, title, detail}]
|
|||
|
|
def rec(cid, level, title, detail):
|
|||
|
|
CHECKS.append({"id": cid, "level": level, "title": title, "detail": detail})
|
|||
|
|
|
|||
|
|
def _ro_conn():
|
|||
|
|
"""活库铁律：只读打开 ＋ busy_timeout（⛔ 绝不写宿主库）。"""
|
|||
|
|
c = sqlite3.connect("file:%s?mode=ro" % DB.replace("\\", "/"), uri=True, timeout=8)
|
|||
|
|
c.execute("PRAGMA busy_timeout=6000")
|
|||
|
|
return c
|
|||
|
|
|
|||
|
|
def _norm_cwd(p):
|
|||
|
|
"""与去重键同源：`path.trim().toLowerCase()`（⛔ 逐字比，别自作聪明归一成 basename）。"""
|
|||
|
|
return str(p or "").replace("\\", "/").strip().lower()
|
|||
|
|
|
|||
|
|
def _now_ms():
|
|||
|
|
return int(time.time() * 1000)
|
|||
|
|
|
|||
|
|
def main():
|
|||
|
|
if not os.path.isfile(DB):
|
|||
|
|
rec("db", "fail", "读不到宿主库", "库不在：%s" % DB)
|
|||
|
|
return finish()
|
|||
|
|
try:
|
|||
|
|
c = _ro_conn()
|
|||
|
|
except Exception as e:
|
|||
|
|
rec("db", "fail", "打不开宿主库（只读）", str(e)[:160])
|
|||
|
|
return finish()
|
|||
|
|
|
|||
|
|
# ── 取数：全部未删排期 ＋ 本工作区相关的会话 ─────────────────────────
|
|||
|
|
try:
|
|||
|
|
autos = list(c.execute(
|
|||
|
|
"select id,name,status,schedule_type,next_run_at,rrule,model_id,"
|
|||
|
|
"model_is_thinking,cwds,last_run_at from automations where deleted_at is null"))
|
|||
|
|
except Exception as e:
|
|||
|
|
rec("db", "fail", "读 automations 失败", str(e)[:160])
|
|||
|
|
return finish()
|
|||
|
|
try:
|
|||
|
|
sess = list(c.execute(
|
|||
|
|
"select id,title,status,cwd,updated_at from sessions "
|
|||
|
|
"where title like '[唤醒]%' or title like '[协作]%' or title like '[跟进]%'"))
|
|||
|
|
except Exception:
|
|||
|
|
sess = []
|
|||
|
|
|
|||
|
|
ws_norm = _norm_cwd(WS)
|
|||
|
|
mine = [a for a in autos if any(_norm_cwd(x) == ws_norm
|
|||
|
|
for x in (_safe_list(a[8])))] or []
|
|||
|
|
|
|||
|
|
# ── ① 周期钟：唤醒 / 跟进 两类必须有 recurring 排期 ────────────────────
|
|||
|
|
clocks = {}
|
|||
|
|
for a in autos:
|
|||
|
|
nm = a[1] or ""
|
|||
|
|
if a[3] != "recurring":
|
|||
|
|
continue
|
|||
|
|
if not any(_norm_cwd(x) == ws_norm for x in _safe_list(a[8])):
|
|||
|
|
continue
|
|||
|
|
for label, pre in ROLES_CLOCK:
|
|||
|
|
if nm.startswith(pre):
|
|||
|
|
clocks.setdefault(label, []).append(a)
|
|||
|
|
missing_clock = [lb for lb, _ in ROLES_CLOCK if not clocks.get(lb)]
|
|||
|
|
if missing_clock:
|
|||
|
|
rec("clock", "fail", "周期钟缺失",
|
|||
|
|
"这些类别没有任何 recurring 排期 ⇒ 没人在推 / 收：%s" % "、".join(missing_clock))
|
|||
|
|
else:
|
|||
|
|
rec("clock", "ok", "周期钟在册",
|
|||
|
|
"；".join("%s×%d" % (lb, len(clocks[lb])) for lb, _ in ROLES_CLOCK)
|
|||
|
|
+ "（协作按需创建 ⇒ 不要求周期钟）")
|
|||
|
|
|
|||
|
|
# ── ② 钟会响吗：模型可用性（thinking=0 ＋ flash ⇒ 必被拒） ─────────────
|
|||
|
|
deaf = []
|
|||
|
|
for lb, lst in clocks.items():
|
|||
|
|
for a in lst:
|
|||
|
|
mid = (a[6] or "").lower()
|
|||
|
|
think = a[7]
|
|||
|
|
if not think and any(h in mid for h in FLASH_HINT):
|
|||
|
|
deaf.append("%s（%s）" % (a[1][:28], a[6]))
|
|||
|
|
if deaf:
|
|||
|
|
rec("model", "fail", "钟会被服务端拒（静默失效）",
|
|||
|
|
"模型不支持关闭思考却传了 thinking=0 ⇒ 每触发必拒：%s" % "、".join(deaf))
|
|||
|
|
else:
|
|||
|
|
rec("model", "ok", "排期模型可用",
|
|||
|
|
"周期排期均未命中「flash ＋ 关思考」这一必拒组合")
|
|||
|
|
|
|||
|
|
# ── ③ 归属同形：cwds 逐字比 ──────────────────────────────────────────
|
|||
|
|
off = []
|
|||
|
|
for a in autos:
|
|||
|
|
for x in _safe_list(a[8]):
|
|||
|
|
s = str(x or "")
|
|||
|
|
if not s.strip():
|
|||
|
|
continue
|
|||
|
|
if _norm_cwd(s) != s.replace("\\", "/").strip().lower():
|
|||
|
|
pass
|
|||
|
|
# 只报"本线的排期但字面不符"（别人的 cwds 不关我事）
|
|||
|
|
if os.path.basename(ws_norm) in _norm_cwd(s) and _norm_cwd(s) != ws_norm:
|
|||
|
|
off.append("%s → %s" % (a[1][:24], s))
|
|||
|
|
if off:
|
|||
|
|
rec("cwd", "fail", "cwds 与本工作区不同形",
|
|||
|
|
"裂组且会自我强化（去重键＝trim+toLowerCase）：%s" % "；".join(off[:3]))
|
|||
|
|
else:
|
|||
|
|
rec("cwd", "ok", "cwds 归属同形", "本线排期的 cwds 与工作区路径逐字一致")
|
|||
|
|
|
|||
|
|
# ── ④ 活会话：三类当前有没有活的 ─────────────────────────────────────
|
|||
|
|
live = {}
|
|||
|
|
for lb, pre in ROLES:
|
|||
|
|
hits = [s for s in sess if (s[1] or "").startswith(pre) and (s[2] or "") == "working"]
|
|||
|
|
live[lb] = hits
|
|||
|
|
dead_roles = [lb for lb, _ in ROLES if not live.get(lb)]
|
|||
|
|
if dead_roles:
|
|||
|
|
rec("live", "warn", "当前无活会话",
|
|||
|
|
"这些类别此刻没有任何活着的会话（按需创建属正常；但主会话开工阶段须建齐）：%s"
|
|||
|
|
% "、".join(dead_roles))
|
|||
|
|
else:
|
|||
|
|
rec("live", "ok", "三类会话均有活的",
|
|||
|
|
"；".join("%s×%d" % (lb, len(live[lb])) for lb, _ in ROLES))
|
|||
|
|
|
|||
|
|
# ── ⑤ 投递在跑：协作程序 / 心跳新鲜度 ────────────────────────────────
|
|||
|
|
hb = os.path.join(WS, ".workbuddy", "collab", "logs", "supervise-heartbeat.json")
|
|||
|
|
hb_age = None
|
|||
|
|
for cand in (hb, os.path.join(WS, ".workbuddy", "collab", "supervise-heartbeat.json")):
|
|||
|
|
try:
|
|||
|
|
if os.path.isfile(cand):
|
|||
|
|
hb_age = time.time() - os.path.getmtime(cand)
|
|||
|
|
break
|
|||
|
|
except Exception:
|
|||
|
|
pass
|
|||
|
|
if hb_age is None:
|
|||
|
|
rec("deliver", "warn", "投递（常驻）未见心跳",
|
|||
|
|
"没找到心跳文件 ⇒ 无法证明投递在跑（⛔ 不等于它一定没跑；载体＝专用容器会话）")
|
|||
|
|
elif hb_age > 900:
|
|||
|
|
rec("deliver", "fail", "投递心跳陈旧",
|
|||
|
|
"最后心跳 %.0f 分钟前 ⇒ 常驻可能已掉线" % (hb_age / 60.0))
|
|||
|
|
else:
|
|||
|
|
rec("deliver", "ok", "投递心跳新鲜", "最后心跳 %.0f 秒前" % hb_age)
|
|||
|
|
|
|||
|
|
# ── ⑥ 死排期：`once` 已消耗 **且从未运行** ⇒ 活静默消失 ────────────────
|
|||
|
|
# 🔴 判据要点两条（都是实测校正出来的，⛔ 别想当然）：
|
|||
|
|
# (a) **"已跑完"不是问题、"从未跑却失效"才是**（首版把两者混报 ⇒ 18 条正常痕迹被当 18 个问题）；
|
|||
|
|
# (b) **"有没有跑过"必须查 `automation_runs`，⛔ 不能查 `automations.last_run_at`** ——
|
|||
|
|
# 该字段宿主**根本不写**（实测：本会话自己那条 `接续 · 会话机制合并包 · 任务4b-4d-6`
|
|||
|
|
# 明明在跑，`last_run_at` 仍是 `None`）⇒ 拿它当判据 = 把"跑过的"全报成"没跑过"。
|
|||
|
|
try:
|
|||
|
|
ran = {r[0] for r in c.execute(
|
|||
|
|
"select distinct automation_id from automation_runs limit 5000")}
|
|||
|
|
except Exception:
|
|||
|
|
ran = None
|
|||
|
|
never_ran, consumed, unverifiable = [], [], []
|
|||
|
|
for a in mine:
|
|||
|
|
if a[3] != "once" or a[4]:
|
|||
|
|
continue
|
|||
|
|
if ran is None:
|
|||
|
|
unverifiable.append((a[1] or "")[:30])
|
|||
|
|
elif a[0] in ran:
|
|||
|
|
consumed.append((a[1] or "")[:30])
|
|||
|
|
else:
|
|||
|
|
never_ran.append((a[1] or "")[:30])
|
|||
|
|
if never_ran:
|
|||
|
|
rec("zombie", "fail", "一次性排期从未运行就失效",
|
|||
|
|
"共 %d 条 ⇒ 这些活会静默消失（消耗掉的一次性排期不会自动重跑）：%s"
|
|||
|
|
% (len(never_ran), "、".join(never_ran[:4])))
|
|||
|
|
elif unverifiable:
|
|||
|
|
rec("zombie", "warn", "无法核对一次性排期是否运行过",
|
|||
|
|
"读不到 automation_runs（⛔ 不等于它们没跑）；本线有 %d 条 once＋无下次触发：%s"
|
|||
|
|
% (len(unverifiable), "、".join(unverifiable[:3])))
|
|||
|
|
elif consumed:
|
|||
|
|
rec("zombie", "ok", "无「从未运行」的一次性排期",
|
|||
|
|
"本线 %d 条已跑完的一次性排期（正常痕迹，不计问题）" % len(consumed))
|
|||
|
|
else:
|
|||
|
|
rec("zombie", "ok", "无残留的一次性排期", "本线没有「once 且不会再触发」的排期")
|
|||
|
|
|
|||
|
|
return finish()
|
|||
|
|
|
|||
|
|
|
|||
|
|
def _safe_list(v):
|
|||
|
|
"""cwds 字段形态不稳（JSON 数组 / 逗号串 / 单值）⇒ 一律解成 list[str]。"""
|
|||
|
|
if v is None:
|
|||
|
|
return []
|
|||
|
|
if isinstance(v, (list, tuple)):
|
|||
|
|
return [str(x) for x in v]
|
|||
|
|
s = str(v).strip()
|
|||
|
|
if s.startswith("["):
|
|||
|
|
try:
|
|||
|
|
r = json.loads(s)
|
|||
|
|
return [str(x) for x in r] if isinstance(r, list) else [s]
|
|||
|
|
except Exception:
|
|||
|
|
return [s]
|
|||
|
|
return [x.strip() for x in s.split(",") if x.strip()]
|
|||
|
|
|
|||
|
|
|
|||
|
|
def finish():
|
|||
|
|
bad = [c for c in CHECKS if c["level"] == "fail"]
|
|||
|
|
warn = [c for c in CHECKS if c["level"] == "warn"]
|
|||
|
|
verdict = "fail" if bad else ("warn" if warn else "ok")
|
|||
|
|
summary = {
|
|||
|
|
"ok": "会话规划齐备且生效",
|
|||
|
|
"warn": "配置在册，运行未齐（见 warn 项）",
|
|||
|
|
"fail": "会话规划有硬缺口（见 fail 项）",
|
|||
|
|
}[verdict]
|
|||
|
|
state = {
|
|||
|
|
"ts": time.strftime("%Y-%m-%d %H:%M:%S"),
|
|||
|
|
"ts_epoch": int(time.time()),
|
|||
|
|
"ws": WS,
|
|||
|
|
"verdict": verdict,
|
|||
|
|
"summary": summary,
|
|||
|
|
"counts": {"fail": len(bad), "warn": len(warn), "ok": len(CHECKS) - len(bad) - len(warn)},
|
|||
|
|
"checks": CHECKS,
|
|||
|
|
}
|
|||
|
|
try:
|
|||
|
|
os.makedirs(os.path.dirname(OUT), exist_ok=True)
|
|||
|
|
tmp = OUT + ".tmp"
|
|||
|
|
with io.open(tmp, "w", encoding="utf-8", newline="\n") as f:
|
|||
|
|
f.write(json.dumps(state, ensure_ascii=False, indent=2))
|
|||
|
|
os.replace(tmp, OUT) # 原子替换（本机文件锁会卡死 ⇒ 见 MEMORY）
|
|||
|
|
except Exception as e:
|
|||
|
|
state["write_error"] = str(e)[:160]
|
|||
|
|
|
|||
|
|
if AS_JSON:
|
|||
|
|
sys.stdout.buffer.write(json.dumps(state, ensure_ascii=False, indent=2).encode("utf-8"))
|
|||
|
|
sys.stdout.buffer.write(b"\n")
|
|||
|
|
elif not QUIET:
|
|||
|
|
icon = {"ok": "✅", "warn": "⚠️", "fail": "🔴"}[verdict]
|
|||
|
|
buf = ["%s [会话规划] %s（fail %d / warn %d）" % (icon, summary, len(bad), len(warn))]
|
|||
|
|
for c in CHECKS:
|
|||
|
|
m = {"ok": "✓", "warn": "⚠", "fail": "✗"}[c["level"]]
|
|||
|
|
buf.append(" %s %s —— %s" % (m, c["title"], c["detail"]))
|
|||
|
|
buf.append(" · 状态标记 → %s" % OUT)
|
|||
|
|
sys.stdout.buffer.write(("\n".join(buf) + "\n").encode("utf-8"))
|
|||
|
|
sys.stdout.flush()
|
|||
|
|
# 退出码语义：**只有"齐备且生效"才 0** —— warn 也要能被机器判出来（⛔ 不用它拦开工，只做标记）
|
|||
|
|
return 0 if verdict == "ok" else 1
|
|||
|
|
|
|||
|
|
|
|||
|
|
if __name__ == "__main__":
|
|||
|
|
try:
|
|||
|
|
sys.exit(main())
|
|||
|
|
except Exception as e:
|
|||
|
|
try:
|
|||
|
|
sys.stdout.buffer.write(("⚠️ [会话规划] 体检自身异常（⛔ 不影响开工）：%s\n"
|
|||
|
|
% str(e)[:200]).encode("utf-8"))
|
|||
|
|
except Exception:
|
|||
|
|
pass
|
|||
|
|
sys.exit(0)
|