litloop 1.18.3__py3-none-any.whl
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.
- litloop/__init__.py +89 -0
- litloop/__main__.py +6 -0
- litloop/_source_hygiene.py +158 -0
- litloop/agents/__init__.py +13 -0
- litloop/agents/client.py +164 -0
- litloop/agents/runner.py +123 -0
- litloop/agents/system.py +54 -0
- litloop/agents/tools.py +142 -0
- litloop/app_server.py +8 -0
- litloop/browser/__init__.py +1 -0
- litloop/browser/backend.py +1009 -0
- litloop/browser/camofox_api.py +368 -0
- litloop/browser/chrome_copy.py +136 -0
- litloop/browser/cloakbrowser_compat.py +72 -0
- litloop/browser/cookies.py +347 -0
- litloop/browser/engine.py +965 -0
- litloop/browser/login.py +636 -0
- litloop/browser/pdf.py +263 -0
- litloop/browser/profile_health.py +167 -0
- litloop/browser/runtime.py +210 -0
- litloop/cache.py +104 -0
- litloop/citations/__init__.py +1 -0
- litloop/citations/bib_import.py +102 -0
- litloop/citations/bibtex_crossref.py +143 -0
- litloop/citations/formats.py +225 -0
- litloop/citations/paper_list.py +200 -0
- litloop/citations/session_export.py +94 -0
- litloop/cli/__init__.py +12 -0
- litloop/cli/_utils.py +267 -0
- litloop/cli/commands.py +1098 -0
- litloop/cli/doctor.py +141 -0
- litloop/cli/main.py +1173 -0
- litloop/cli/web.py +190 -0
- litloop/config.py +413 -0
- litloop/data/VARCON-LICENSE.txt +62 -0
- litloop/data/elements.tsv +119 -0
- litloop/data/institutional_identity_policy.json +165 -0
- litloop/data/publisher_access_catalog.json +969 -0
- litloop/data/publisher_carsi.json +105 -0
- litloop/data/spelling_search.tsv.gz +0 -0
- litloop/data/spelling_us.tsv.gz +0 -0
- litloop/data/webvpn.dat +0 -0
- litloop/data/webvpn.json +603 -0
- litloop/db.py +1922 -0
- litloop/deps.py +128 -0
- litloop/desktop/__init__.py +1 -0
- litloop/desktop/agent_runner.py +537 -0
- litloop/desktop/download_jobs.py +571 -0
- litloop/desktop/error_text.py +192 -0
- litloop/desktop/server.py +999 -0
- litloop/desktop/settings.py +552 -0
- litloop/desktop/translate.py +103 -0
- litloop/diagnostics.py +126 -0
- litloop/download/__init__.py +1 -0
- litloop/download/doi_check.py +44 -0
- litloop/download/fetcher.py +975 -0
- litloop/download/markdown.py +130 -0
- litloop/download/pdf.py +454 -0
- litloop/download/pipeline.py +1166 -0
- litloop/download/progress.py +312 -0
- litloop/download/rename.py +211 -0
- litloop/extractors/__init__.py +0 -0
- litloop/extractors/html_extractor.py +31 -0
- litloop/extractors/pdf_extractor.py +151 -0
- litloop/extractors/publisher_adapters/__init__.py +0 -0
- litloop/extractors/publisher_adapters/acs.py +167 -0
- litloop/extractors/publisher_adapters/elsevier.py +178 -0
- litloop/extractors/publisher_adapters/generic.py +191 -0
- litloop/extractors/publisher_adapters/nature.py +204 -0
- litloop/extractors/publisher_adapters/rsc.py +154 -0
- litloop/extractors/publisher_adapters/tandfonline.py +154 -0
- litloop/extractors/publisher_adapters/wiley.py +160 -0
- litloop/http_utils.py +57 -0
- litloop/identifiers.py +134 -0
- litloop/institutional/__init__.py +1 -0
- litloop/institutional/auth.py +576 -0
- litloop/institutional/carsi_login.py +569 -0
- litloop/institutional/config_adapter.py +63 -0
- litloop/institutional/proxy_check.py +98 -0
- litloop/institutional/remote_assist.py +274 -0
- litloop/institutional/schools.py +102 -0
- litloop/institutional/session_heal.py +73 -0
- litloop/institutional/session_store.py +105 -0
- litloop/log.py +47 -0
- litloop/models.py +269 -0
- litloop/network.py +417 -0
- litloop/optional.py +310 -0
- litloop/publishers/__init__.py +1 -0
- litloop/publishers/access.py +232 -0
- litloop/publishers/pdf_router.py +451 -0
- litloop/publishers/profiles.py +899 -0
- litloop/publishers/strategies/__init__.py +122 -0
- litloop/publishers/strategies/_core.py +2340 -0
- litloop/publishers/strategies/_elsevier_api.py +578 -0
- litloop/publishers/strategies/acm.py +27 -0
- litloop/publishers/strategies/acs.py +42 -0
- litloop/publishers/strategies/aip.py +27 -0
- litloop/publishers/strategies/aps.py +29 -0
- litloop/publishers/strategies/asce.py +28 -0
- litloop/publishers/strategies/base.py +108 -0
- litloop/publishers/strategies/copernicus.py +24 -0
- litloop/publishers/strategies/elsevier.py +66 -0
- litloop/publishers/strategies/generic.py +39 -0
- litloop/publishers/strategies/ieee.py +28 -0
- litloop/publishers/strategies/iop.py +27 -0
- litloop/publishers/strategies/nature.py +27 -0
- litloop/publishers/strategies/oxford.py +27 -0
- litloop/publishers/strategies/registry.py +83 -0
- litloop/publishers/strategies/royalsociety.py +28 -0
- litloop/publishers/strategies/rsc.py +27 -0
- litloop/publishers/strategies/sage.py +30 -0
- litloop/publishers/strategies/science.py +27 -0
- litloop/publishers/strategies/springer.py +34 -0
- litloop/publishers/strategies/tandfonline.py +32 -0
- litloop/publishers/strategies/wiley.py +32 -0
- litloop/publishers/supplementary.py +219 -0
- litloop/quota.py +202 -0
- litloop/search/__init__.py +1 -0
- litloop/search/abstracts.py +291 -0
- litloop/search/deep.py +1216 -0
- litloop/search/quick.py +1102 -0
- litloop/search/scansci_find.py +410 -0
- litloop/search/screening.py +225 -0
- litloop/search/title_to_doi.py +163 -0
- litloop/search/watches.py +124 -0
- litloop/server.py +1731 -0
- litloop/sources/__init__.py +1339 -0
- litloop/sources/_batch.py +529 -0
- litloop/sources/_index.py +221 -0
- litloop/sources/_routes.py +290 -0
- litloop/sources/altcha.py +90 -0
- litloop/sources/arxiv.py +45 -0
- litloop/sources/attempt.py +91 -0
- litloop/sources/carsi.py +74 -0
- litloop/sources/core_api.py +46 -0
- litloop/sources/crossref.py +162 -0
- litloop/sources/elsevier_api.py +620 -0
- litloop/sources/elsevier_check.py +274 -0
- litloop/sources/europepmc.py +81 -0
- litloop/sources/ezproxy.py +317 -0
- litloop/sources/instsci.py +1295 -0
- litloop/sources/libgen.py +116 -0
- litloop/sources/nature.py +39 -0
- litloop/sources/oa_discovery.py +135 -0
- litloop/sources/openaire.py +122 -0
- litloop/sources/openalex.py +100 -0
- litloop/sources/publishers.py +684 -0
- litloop/sources/sage_cn.py +139 -0
- litloop/sources/scibban.py +10 -0
- litloop/sources/scihub.py +1379 -0
- litloop/sources/scihub_domains.py +203 -0
- litloop/sources/scihub_speed.py +187 -0
- litloop/sources/scoring.py +250 -0
- litloop/sources/semantic_scholar.py +146 -0
- litloop/sources/springer_tdm.py +172 -0
- litloop/sources/unpaywall.py +116 -0
- litloop/sources/vpnsci.py +38 -0
- litloop/store.py +264 -0
- litloop/templates/index.html +347 -0
- litloop/terms/__init__.py +1 -0
- litloop/terms/glossary.py +422 -0
- litloop/terms/spelling.py +89 -0
- litloop/terms/text.py +106 -0
- litloop/tor/__init__.py +1 -0
- litloop/tor/embedded.py +371 -0
- litloop/tor/proxy.py +110 -0
- litloop/updates.py +148 -0
- litloop/zotero/__init__.py +1 -0
- litloop/zotero/bridge.py +309 -0
- litloop/zotero/client.py +317 -0
- litloop-1.18.3.dist-info/METADATA +347 -0
- litloop-1.18.3.dist-info/RECORD +176 -0
- litloop-1.18.3.dist-info/WHEEL +5 -0
- litloop-1.18.3.dist-info/entry_points.txt +2 -0
- litloop-1.18.3.dist-info/licenses/LICENSE +197 -0
- litloop-1.18.3.dist-info/top_level.txt +1 -0
litloop/deps.py
ADDED
|
@@ -0,0 +1,128 @@
|
|
|
1
|
+
"""Dependency checking and graceful degradation."""
|
|
2
|
+
|
|
3
|
+
from __future__ import annotations
|
|
4
|
+
|
|
5
|
+
import importlib
|
|
6
|
+
from typing import Any
|
|
7
|
+
|
|
8
|
+
from .log import get_logger
|
|
9
|
+
|
|
10
|
+
log = get_logger()
|
|
11
|
+
|
|
12
|
+
# Core dependencies (required)
|
|
13
|
+
CORE_DEPS = {
|
|
14
|
+
"requests": "HTTP client",
|
|
15
|
+
"bs4": "HTML parsing (beautifulsoup4)",
|
|
16
|
+
"mcp": "MCP protocol",
|
|
17
|
+
"typer": "CLI framework",
|
|
18
|
+
"uvicorn": "ASGI server",
|
|
19
|
+
"fastapi": "Web UI / HTTP transport",
|
|
20
|
+
"jinja2": "Web UI templates",
|
|
21
|
+
"pymupdf": "PDF text extraction (PyMuPDF)",
|
|
22
|
+
}
|
|
23
|
+
|
|
24
|
+
# Optional dependencies
|
|
25
|
+
OPTIONAL_DEPS = {
|
|
26
|
+
"socks": "SOCKS proxy support (requests[socks])",
|
|
27
|
+
"Crypto": "WebVPN AES encryption (pycryptodome)",
|
|
28
|
+
"patchright": "Stealth browser backend (default; open-source Playwright fork)",
|
|
29
|
+
"cloakbrowser": "Stealth browser backend (fallback; free tier pinned to Chromium 146)",
|
|
30
|
+
}
|
|
31
|
+
|
|
32
|
+
_feature_status: dict[str, bool] = {}
|
|
33
|
+
|
|
34
|
+
|
|
35
|
+
def check_dependency(module_name: str) -> bool:
|
|
36
|
+
"""Check if a Python module is available."""
|
|
37
|
+
try:
|
|
38
|
+
importlib.import_module(module_name)
|
|
39
|
+
return True
|
|
40
|
+
except ImportError:
|
|
41
|
+
return False
|
|
42
|
+
|
|
43
|
+
|
|
44
|
+
def check_all() -> dict[str, dict[str, Any]]:
|
|
45
|
+
"""Check all dependencies and return status report."""
|
|
46
|
+
result = {"core": {}, "optional": {}}
|
|
47
|
+
|
|
48
|
+
for module, desc in CORE_DEPS.items():
|
|
49
|
+
available = check_dependency(module)
|
|
50
|
+
result["core"][module] = {
|
|
51
|
+
"available": available,
|
|
52
|
+
"description": desc,
|
|
53
|
+
"required": True,
|
|
54
|
+
}
|
|
55
|
+
|
|
56
|
+
for module, desc in OPTIONAL_DEPS.items():
|
|
57
|
+
available = check_dependency(module)
|
|
58
|
+
result["optional"][module] = {
|
|
59
|
+
"available": available,
|
|
60
|
+
"description": desc,
|
|
61
|
+
"required": False,
|
|
62
|
+
}
|
|
63
|
+
|
|
64
|
+
return result
|
|
65
|
+
|
|
66
|
+
|
|
67
|
+
def is_feature_available(feature: str) -> bool:
|
|
68
|
+
"""Check if a feature's dependencies are met."""
|
|
69
|
+
if feature in _feature_status:
|
|
70
|
+
return _feature_status[feature]
|
|
71
|
+
|
|
72
|
+
feature_deps = {
|
|
73
|
+
"socks_proxy": ["socks"],
|
|
74
|
+
"instsci": ["Crypto", "cloakbrowser"],
|
|
75
|
+
"html_parse": ["bs4"],
|
|
76
|
+
}
|
|
77
|
+
|
|
78
|
+
deps = feature_deps.get(feature, [])
|
|
79
|
+
available = all(check_dependency(d) for d in deps)
|
|
80
|
+
_feature_status[feature] = available
|
|
81
|
+
return available
|
|
82
|
+
|
|
83
|
+
|
|
84
|
+
def warn_missing(feature: str, module: str) -> None:
|
|
85
|
+
"""Log a warning when an optional dependency is missing."""
|
|
86
|
+
log.warning(f"Feature '{feature}' disabled: install '{module}' to enable")
|
|
87
|
+
|
|
88
|
+
|
|
89
|
+
def print_status() -> None:
|
|
90
|
+
"""Print dependency status to stderr."""
|
|
91
|
+
report = check_all()
|
|
92
|
+
|
|
93
|
+
log.info("=== Dependency Status ===")
|
|
94
|
+
|
|
95
|
+
all_ok = True
|
|
96
|
+
for module, info in report["core"].items():
|
|
97
|
+
status = "OK" if info["available"] else "MISSING"
|
|
98
|
+
if not info["available"]:
|
|
99
|
+
all_ok = False
|
|
100
|
+
log.info(f" [{status}] {module} - {info['description']}")
|
|
101
|
+
|
|
102
|
+
if not all_ok:
|
|
103
|
+
log.warning("Some core dependencies are missing! Install with: pip install litloop")
|
|
104
|
+
|
|
105
|
+
has_optional = False
|
|
106
|
+
for module, info in report["optional"].items():
|
|
107
|
+
if info["available"]:
|
|
108
|
+
status = "OK"
|
|
109
|
+
else:
|
|
110
|
+
status = "optional"
|
|
111
|
+
has_optional = True
|
|
112
|
+
log.info(f" [{status}] {module} - {info['description']}")
|
|
113
|
+
|
|
114
|
+
# cloakbrowser iterates on anti-detection patches fast; flag stale installs
|
|
115
|
+
# so users know to upgrade even when the package is present.
|
|
116
|
+
if report["optional"].get("cloakbrowser", {}).get("available"):
|
|
117
|
+
try:
|
|
118
|
+
from .cli import _cloakbrowser_version_status
|
|
119
|
+
cb_status, cb_detail = _cloakbrowser_version_status()
|
|
120
|
+
if cb_status == "outdated":
|
|
121
|
+
log.warning(f" [STALE] cloakbrowser {cb_detail}")
|
|
122
|
+
except Exception:
|
|
123
|
+
pass
|
|
124
|
+
|
|
125
|
+
if has_optional:
|
|
126
|
+
log.info(" Install optional deps: pip install litloop[fast,vpnsci]")
|
|
127
|
+
|
|
128
|
+
log.info("========================")
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
"""Desktop application backend, settings, and agent runner."""
|
|
@@ -0,0 +1,537 @@
|
|
|
1
|
+
"""Searches started from the desktop app: the form becomes a message to the agent.
|
|
2
|
+
|
|
3
|
+
The app does what the user would do in a chat: it sends the question, the
|
|
4
|
+
criteria, the year and journal filters, the seeds and the scale to the agent
|
|
5
|
+
(``agent_command``, Claude Code's ``claude -p`` by default), which runs the
|
|
6
|
+
usual LitLoop workflow with the litloop tools and saves a screening session.
|
|
7
|
+
One run at a time. Each run is ``agent_runs/<id>.json`` (status, what the agent
|
|
8
|
+
is doing, the screening session it saved) plus the agent's raw output in
|
|
9
|
+
``agent_runs/<id>.log``.
|
|
10
|
+
"""
|
|
11
|
+
|
|
12
|
+
from __future__ import annotations
|
|
13
|
+
|
|
14
|
+
import json
|
|
15
|
+
import os
|
|
16
|
+
import re
|
|
17
|
+
import shlex
|
|
18
|
+
import shutil
|
|
19
|
+
import subprocess
|
|
20
|
+
import threading
|
|
21
|
+
import time
|
|
22
|
+
from datetime import datetime
|
|
23
|
+
from pathlib import Path
|
|
24
|
+
from typing import Any
|
|
25
|
+
|
|
26
|
+
from .. import config as _config
|
|
27
|
+
from .. import db, store
|
|
28
|
+
|
|
29
|
+
# What the agent tells the app, by tool: shown under the run as it works.
|
|
30
|
+
_TOOL_TEXT = {
|
|
31
|
+
"litloop_search": "正在检索",
|
|
32
|
+
"litloop_deep_search": "正在深度检索",
|
|
33
|
+
"litloop_screening": "正在保存评分",
|
|
34
|
+
"litloop_find": "正在查找文献",
|
|
35
|
+
"litloop_config": "正在读取设置",
|
|
36
|
+
"litloop_diagnostics": "正在检测设置",
|
|
37
|
+
}
|
|
38
|
+
_SESSION_RE = re.compile(r'"session"\s*:\s*"([^"]+)"')
|
|
39
|
+
_LOCK = threading.Lock()
|
|
40
|
+
_DONE: set[str] = set() # builtin runs whose final state is written
|
|
41
|
+
_PROCS: dict[str, subprocess.Popen] = {}
|
|
42
|
+
_CANCELLED: set[str] = set()
|
|
43
|
+
_SESSION_TOOLS = frozenset({"litloop_screening", "litloop_deep_search"})
|
|
44
|
+
# Backend B runs in a thread of this process instead of a child process.
|
|
45
|
+
_BUILTIN: dict[str, threading.Thread] = {}
|
|
46
|
+
_STOP: dict[str, threading.Event] = {}
|
|
47
|
+
_BUILTIN_TOOLS = {
|
|
48
|
+
"search": "正在检索",
|
|
49
|
+
"deep_search_start": "深度检索:正在构建候选池",
|
|
50
|
+
"deep_search_rate": "深度检索:正在保存本批评分",
|
|
51
|
+
"deep_search_next": "深度检索:正在生成下一批",
|
|
52
|
+
"deep_search_status": "深度检索:正在查看进度",
|
|
53
|
+
"screening_save": "正在保存评分",
|
|
54
|
+
}
|
|
55
|
+
|
|
56
|
+
|
|
57
|
+
class _Cancelled(Exception):
|
|
58
|
+
pass
|
|
59
|
+
|
|
60
|
+
|
|
61
|
+
def runs_dir() -> Path:
|
|
62
|
+
return _config.DATA_DIR / "agent_runs"
|
|
63
|
+
|
|
64
|
+
|
|
65
|
+
def _path(run: str) -> Path:
|
|
66
|
+
if not re.fullmatch(r"[0-9-]+", run):
|
|
67
|
+
raise ValueError("无法打开检索:编号不正确")
|
|
68
|
+
return runs_dir() / f"{run}.json"
|
|
69
|
+
|
|
70
|
+
|
|
71
|
+
def _lines(text: str) -> list[str]:
|
|
72
|
+
return [s.strip() for s in re.split(r"[\n;;]+", text or "") if s.strip()]
|
|
73
|
+
|
|
74
|
+
|
|
75
|
+
def build_prompt(form: dict[str, Any], skill: bool = True) -> str:
|
|
76
|
+
"""The message the user would have typed, from the app's form.
|
|
77
|
+
|
|
78
|
+
``skill=False`` is for the built-in agent: it is told what to do by the system
|
|
79
|
+
prompt in ``litloop.agents.system`` instead of by a skill it cannot read.
|
|
80
|
+
"""
|
|
81
|
+
question = str(form.get("question") or "").strip()
|
|
82
|
+
if not question:
|
|
83
|
+
raise ValueError("填写检索问题")
|
|
84
|
+
deep = form.get("mode") == "deep"
|
|
85
|
+
parts = [f"{'用 litloop 技能' if skill else ''}{'做深度检索' if deep else '快速查找文献'}:{question}"]
|
|
86
|
+
criteria = _lines(str(form.get("criteria") or ""))
|
|
87
|
+
if criteria:
|
|
88
|
+
parts.append("纳入标准(按这些判断,不要改):\n" + "\n".join(f"- {c}" for c in criteria))
|
|
89
|
+
else:
|
|
90
|
+
parts.append("纳入标准由你按问题拟定 2–5 条,保存时写进 criteria。")
|
|
91
|
+
y0, y1 = form.get("year_from"), form.get("year_to")
|
|
92
|
+
if y0 or y1:
|
|
93
|
+
parts.append(f"年份:{y0 or '不限'}–{y1 or '至今'}。")
|
|
94
|
+
rank = []
|
|
95
|
+
if form.get("max_cas_zone"):
|
|
96
|
+
rank.append(f"中科院分区 {form['max_cas_zone']} 区及以上(max_cas_zone={form['max_cas_zone']})")
|
|
97
|
+
if form.get("max_jcr_quartile"):
|
|
98
|
+
rank.append(f"JCR Q{form['max_jcr_quartile']} 及以上(max_jcr_quartile={form['max_jcr_quartile']})")
|
|
99
|
+
if form.get("exclude_warning"):
|
|
100
|
+
rank.append("排除预警期刊(exclude_warning=true)")
|
|
101
|
+
if rank:
|
|
102
|
+
parts.append("期刊要求:" + ";".join(rank) + "。")
|
|
103
|
+
seeds = _lines(str(form.get("seeds") or "").replace(",", "\n").replace(",", "\n"))
|
|
104
|
+
if seeds:
|
|
105
|
+
parts.append("种子文献:" + "、".join(seeds))
|
|
106
|
+
if form.get("seed_collection"):
|
|
107
|
+
parts.append(f"Zotero 分类「{form['seed_collection']}」里的文献也作种子(seed_collection)。")
|
|
108
|
+
if deep:
|
|
109
|
+
batches = int(form.get("batches") or 0)
|
|
110
|
+
parts.append(
|
|
111
|
+
f"每批逐篇评分后取下一批,{'最多 ' + str(batches) + ' 批,' if batches else ''}"
|
|
112
|
+
"返回 stop_suggested=true 时停止。")
|
|
113
|
+
else:
|
|
114
|
+
parts.append(f"返回约 {int(form.get('limit') or 20)} 篇,逐篇评分后保存。")
|
|
115
|
+
parts.append(
|
|
116
|
+
"这是从 LitLoop 桌面 app 发起的任务,用户不在对话里:不要提问、不要等确认,"
|
|
117
|
+
"需要用户决定的事写进最后的汇报;不要下载、不要推送 Zotero。"
|
|
118
|
+
"结束时用两三句中文汇报筛选会话名、篇数和各相关度数量。")
|
|
119
|
+
return "\n\n".join(parts)
|
|
120
|
+
|
|
121
|
+
|
|
122
|
+
def _command() -> list[str]:
|
|
123
|
+
cfg = _config.load_config()
|
|
124
|
+
cmd = str(cfg.get("agent_command") or "claude").strip()
|
|
125
|
+
exe = shutil.which(cmd) or cmd
|
|
126
|
+
# Windows paths keep their backslashes; quotes only group words.
|
|
127
|
+
args = [a.strip('"') for a in shlex.split(str(cfg.get("agent_args") or ""), posix=os.name != "nt")]
|
|
128
|
+
return [exe, *args]
|
|
129
|
+
|
|
130
|
+
|
|
131
|
+
def _backend() -> str:
|
|
132
|
+
"""Which backend runs a search started here: the agent command or a model API."""
|
|
133
|
+
return "builtin" if str(_config.load_config().get("agent_backend") or "") == "builtin" else "command"
|
|
134
|
+
|
|
135
|
+
|
|
136
|
+
def _write(run: str, data: dict[str, Any]) -> None:
|
|
137
|
+
store.write_json(_path(run), {k: v for k, v in data.items() if not k.startswith("_")}, indent=1)
|
|
138
|
+
|
|
139
|
+
|
|
140
|
+
def get(run: str) -> dict[str, Any] | None:
|
|
141
|
+
p = _path(run)
|
|
142
|
+
if not p.exists():
|
|
143
|
+
return None
|
|
144
|
+
d = store.read_json(p)
|
|
145
|
+
if d.get("state") == "running" and not _alive(run):
|
|
146
|
+
# The reader may have caught the file just before the run wrote its
|
|
147
|
+
# final state; if the run is gone now, look once more before calling it
|
|
148
|
+
# interrupted.
|
|
149
|
+
d = store.read_json(p)
|
|
150
|
+
if d.get("state") == "running":
|
|
151
|
+
d["state"] = "interrupted" # the app was closed while it ran
|
|
152
|
+
return d
|
|
153
|
+
|
|
154
|
+
|
|
155
|
+
def _alive(run: str) -> bool:
|
|
156
|
+
with _LOCK:
|
|
157
|
+
return run in _PROCS or run in _BUILTIN or run in _DONE
|
|
158
|
+
|
|
159
|
+
|
|
160
|
+
def list_runs(limit: int = 20) -> list[dict[str, Any]]:
|
|
161
|
+
out = []
|
|
162
|
+
for p in sorted(runs_dir().glob("*.json"), reverse=True)[:limit]:
|
|
163
|
+
d = get(p.stem)
|
|
164
|
+
if d:
|
|
165
|
+
d.pop("events", None)
|
|
166
|
+
out.append(d)
|
|
167
|
+
return out
|
|
168
|
+
|
|
169
|
+
|
|
170
|
+
def running() -> str | None:
|
|
171
|
+
with _LOCK:
|
|
172
|
+
for r, p in _PROCS.items():
|
|
173
|
+
if p.poll() is None:
|
|
174
|
+
return r
|
|
175
|
+
return next((r for r, t in _BUILTIN.items() if t.is_alive() and r not in _DONE), None)
|
|
176
|
+
|
|
177
|
+
|
|
178
|
+
def _new_run_id() -> str:
|
|
179
|
+
base = datetime.now().strftime("%Y%m%d-%H%M%S")
|
|
180
|
+
run = base
|
|
181
|
+
idx = 2
|
|
182
|
+
while _path(run).exists():
|
|
183
|
+
run = f"{base}-{idx}"
|
|
184
|
+
idx += 1
|
|
185
|
+
return run
|
|
186
|
+
|
|
187
|
+
|
|
188
|
+
def build_resume_prompt(session: str, *, batches: int = 0, skill: bool = True) -> str:
|
|
189
|
+
"""Build the prompt to resume an interrupted or paused deep-search session.
|
|
190
|
+
|
|
191
|
+
Reads only the small screening record and optional summary; never reads the
|
|
192
|
+
(potentially huge) candidate pool file (docs/APP_DATA.md).
|
|
193
|
+
"""
|
|
194
|
+
from ..search import deep as deep_search, screening
|
|
195
|
+
|
|
196
|
+
if not deep_search.has_pool(session):
|
|
197
|
+
raise ValueError("无法继续检索:这不是深度检索会话")
|
|
198
|
+
record = screening.load(session)
|
|
199
|
+
summary = db.load_summary(session)
|
|
200
|
+
|
|
201
|
+
question = str(record.get("question") or "").strip()
|
|
202
|
+
head = f"{'用 litloop 技能' if skill else ''}继续深度检索会话 {session}" + (f":{question}" if question else "")
|
|
203
|
+
parts = [head]
|
|
204
|
+
criteria = [str(c).strip() for c in (record.get("criteria") or []) if str(c).strip()]
|
|
205
|
+
if criteria:
|
|
206
|
+
parts.append("纳入标准(按这些判断,不要改):\n" + "\n".join(f"- {c}" for c in criteria))
|
|
207
|
+
if isinstance(summary, dict):
|
|
208
|
+
screened = int(summary.get("screened") or 0)
|
|
209
|
+
pool_size = int(summary.get("pool") or 0)
|
|
210
|
+
done_batches = len(summary.get("batches") or [])
|
|
211
|
+
parts.append(f"已有进度:已生成 {done_batches} 批,已评分 {screened}/{pool_size} 篇。")
|
|
212
|
+
else:
|
|
213
|
+
parts.append("上次检索在生成首批结果前中断,候选池基础记录已保存。")
|
|
214
|
+
limit_note = f"本次最多再筛 {int(batches)} 批," if batches else ""
|
|
215
|
+
if skill:
|
|
216
|
+
parts.append(
|
|
217
|
+
f"不要重新调用 litloop_deep_search(action='start')。请从这个会话接着做:"
|
|
218
|
+
f"先调用 litloop_deep_search(action='status', session={session!r}) 查看进度与 unrated;"
|
|
219
|
+
f"如果 unrated 非空,先按编号用 action='rate' 评完并保存,再调用 action='next' 继续生成并筛选下一批,"
|
|
220
|
+
f"{limit_note}返回 stop_suggested=true 时停止。"
|
|
221
|
+
)
|
|
222
|
+
else:
|
|
223
|
+
parts.append(
|
|
224
|
+
f"不要重新调用 deep_search_start。请从这个会话接着做:"
|
|
225
|
+
f"先调用 deep_search_status(session={session!r}) 查看进度与 unrated;"
|
|
226
|
+
f"如果 unrated 非空,先按编号用 deep_search_rate 评完并保存,再调用 deep_search_next(session={session!r}) 继续生成并筛选下一批,"
|
|
227
|
+
f"{limit_note}返回 stop_suggested=true 时停止。"
|
|
228
|
+
)
|
|
229
|
+
parts.append(
|
|
230
|
+
"这是从 LitLoop 桌面 app 发起的任务,用户不在对话里:不要提问、不要等确认,"
|
|
231
|
+
"需要用户决定的事写进最后的汇报;不要下载、不要推送 Zotero。"
|
|
232
|
+
"结束时用两三句中文汇报筛选会话名、总篇数和各相关度数量。"
|
|
233
|
+
)
|
|
234
|
+
return "\n\n".join(parts)
|
|
235
|
+
|
|
236
|
+
|
|
237
|
+
def _launch(run: str, data: dict[str, Any], prompt: str, builtin: bool) -> dict[str, Any]:
|
|
238
|
+
_write(run, data)
|
|
239
|
+
if builtin:
|
|
240
|
+
if not str(_config.load_config().get("builtin_agent_model") or "").strip():
|
|
241
|
+
data.update(state="failed", doing=None,
|
|
242
|
+
error="无法开始检索:还没有填内置 Agent 的模型名。在「设置 → 内置 Agent」里填写")
|
|
243
|
+
_write(run, data)
|
|
244
|
+
return data
|
|
245
|
+
thread = threading.Thread(target=_run_builtin, args=(run, data, prompt), daemon=True)
|
|
246
|
+
with _LOCK:
|
|
247
|
+
_BUILTIN[run] = thread
|
|
248
|
+
thread.start()
|
|
249
|
+
return data
|
|
250
|
+
cmd = _command()
|
|
251
|
+
try:
|
|
252
|
+
proc = subprocess.Popen(
|
|
253
|
+
cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.STDOUT,
|
|
254
|
+
cwd=str(Path.home()), env={**os.environ, "PYTHONUTF8": "1"},
|
|
255
|
+
creationflags=getattr(subprocess, "CREATE_NO_WINDOW", 0))
|
|
256
|
+
except OSError:
|
|
257
|
+
data.update(state="failed", doing=None, error=f"无法启动 Agent({cmd[0]})。在「设置 → Agent」中检查 Agent 命令")
|
|
258
|
+
_write(run, data)
|
|
259
|
+
return data
|
|
260
|
+
with _LOCK:
|
|
261
|
+
_PROCS[run] = proc
|
|
262
|
+
threading.Thread(target=_follow, args=(run, proc, prompt, data), daemon=True).start()
|
|
263
|
+
return data
|
|
264
|
+
|
|
265
|
+
|
|
266
|
+
def start(form: dict[str, Any]) -> dict[str, Any]:
|
|
267
|
+
builtin = _backend() == "builtin"
|
|
268
|
+
prompt = build_prompt(form, skill=not builtin)
|
|
269
|
+
if running():
|
|
270
|
+
raise RuntimeError("已有检索在进行,完成或取消后再开始新的检索")
|
|
271
|
+
run = _new_run_id()
|
|
272
|
+
runs_dir().mkdir(parents=True, exist_ok=True)
|
|
273
|
+
data = {"run": run, "state": "running", "started": datetime.now().isoformat(timespec="seconds"),
|
|
274
|
+
"question": str(form.get("question")).strip(), "mode": form.get("mode") or "quick",
|
|
275
|
+
"form": form, "prompt": prompt, "doing": "正在启动 Agent", "events": [],
|
|
276
|
+
"session": None, "agent_session": None, "result": None, "error": None}
|
|
277
|
+
return _launch(run, data, prompt, builtin)
|
|
278
|
+
|
|
279
|
+
|
|
280
|
+
def resume(session: str, *, batches: int = 0) -> dict[str, Any]:
|
|
281
|
+
builtin = _backend() == "builtin"
|
|
282
|
+
prompt = build_resume_prompt(session, batches=batches, skill=not builtin)
|
|
283
|
+
if running():
|
|
284
|
+
raise RuntimeError("已有检索在进行,完成或取消后再开始新的检索")
|
|
285
|
+
from ..search import screening
|
|
286
|
+
record = screening.load(session)
|
|
287
|
+
run = _new_run_id()
|
|
288
|
+
runs_dir().mkdir(parents=True, exist_ok=True)
|
|
289
|
+
question = str(record.get("question") or "").strip() or session
|
|
290
|
+
data = {"run": run, "state": "running", "started": datetime.now().isoformat(timespec="seconds"),
|
|
291
|
+
"question": question, "mode": "deep",
|
|
292
|
+
"form": {"question": question, "mode": "deep", "batches": int(batches or 0), "session": session},
|
|
293
|
+
"prompt": prompt, "doing": "正在继续深度检索", "events": [],
|
|
294
|
+
"session": session, "resumed_from": session, "agent_session": None,
|
|
295
|
+
"result": None, "error": None}
|
|
296
|
+
return _launch(run, data, prompt, builtin)
|
|
297
|
+
|
|
298
|
+
|
|
299
|
+
def _event(data: dict[str, Any], text: str) -> None:
|
|
300
|
+
data["events"].append({"t": datetime.now().strftime("%H:%M:%S"), "text": text})
|
|
301
|
+
data["events"] = data["events"][-200:]
|
|
302
|
+
|
|
303
|
+
|
|
304
|
+
def _read(data: dict[str, Any], line: str) -> None:
|
|
305
|
+
"""One line of ``--output-format stream-json``; other output is kept in the log only."""
|
|
306
|
+
try:
|
|
307
|
+
msg = json.loads(line)
|
|
308
|
+
except ValueError:
|
|
309
|
+
return
|
|
310
|
+
kind = msg.get("type")
|
|
311
|
+
if kind == "system" and msg.get("session_id"):
|
|
312
|
+
data["agent_session"] = msg["session_id"]
|
|
313
|
+
for c in ((msg.get("message") or {}).get("content") or []) if kind in ("assistant", "user") else []:
|
|
314
|
+
if not isinstance(c, dict):
|
|
315
|
+
continue
|
|
316
|
+
if c.get("type") == "tool_use":
|
|
317
|
+
name = str(c.get("name") or "").split("__")[-1]
|
|
318
|
+
data["_last_tool"] = name
|
|
319
|
+
if c.get("id"):
|
|
320
|
+
data.setdefault("_tools", {})[c["id"]] = name
|
|
321
|
+
action = (c.get("input") or {}).get("action")
|
|
322
|
+
text = _TOOL_TEXT.get(name, f"正在调用 {name}")
|
|
323
|
+
if name == "litloop_deep_search" and action:
|
|
324
|
+
text = {"start": "深度检索:正在构建候选池", "rate": "深度检索:正在保存本批评分",
|
|
325
|
+
"next": "深度检索:正在生成下一批", "status": "深度检索:正在查看进度"}.get(action, text)
|
|
326
|
+
data["doing"] = text
|
|
327
|
+
_event(data, text)
|
|
328
|
+
elif c.get("type") == "tool_result":
|
|
329
|
+
body = c.get("content")
|
|
330
|
+
body = body if isinstance(body, str) else " ".join(
|
|
331
|
+
x.get("text", "") for x in body or [] if isinstance(x, dict))
|
|
332
|
+
tool_id = c.get("tool_use_id")
|
|
333
|
+
tool = (data.get("_tools") or {}).get(tool_id) if tool_id else None
|
|
334
|
+
if not tool:
|
|
335
|
+
tool = data.get("_last_tool")
|
|
336
|
+
session = _session_from(tool, body or "")
|
|
337
|
+
if session:
|
|
338
|
+
data["session"] = session
|
|
339
|
+
elif c.get("type") == "text" and kind == "assistant" and c.get("text", "").strip():
|
|
340
|
+
data["doing"] = "Agent 正在评分"
|
|
341
|
+
if kind == "result":
|
|
342
|
+
data["result"] = str(msg.get("result") or "").strip() or None
|
|
343
|
+
if msg.get("is_error"):
|
|
344
|
+
data["error"] = data["result"] or "无法完成检索:Agent 出错"
|
|
345
|
+
if re.search(r"authenticat|log ?in|OAuth", data["error"], re.I):
|
|
346
|
+
cmd = _config.load_config().get("agent_command") or "claude"
|
|
347
|
+
data["error"] = f"无法完成检索:Agent 未登录。在终端运行 {cmd} 登录后重试"
|
|
348
|
+
data["result"] = None
|
|
349
|
+
data["cost_usd"] = msg.get("total_cost_usd")
|
|
350
|
+
|
|
351
|
+
|
|
352
|
+
def _session_from(tool: str | None, body: str) -> str | None:
|
|
353
|
+
"""The screening session a LitLoop tool reported: its JSON "session" field.
|
|
354
|
+
|
|
355
|
+
Only litloop_screening and litloop_deep_search name the session the app
|
|
356
|
+
opens; a session-like string in another tool's output (a web page, a
|
|
357
|
+
file the agent read) is not it (code review, 2026-10-18)."""
|
|
358
|
+
if tool not in _SESSION_TOOLS:
|
|
359
|
+
return None
|
|
360
|
+
try:
|
|
361
|
+
obj = json.loads(body)
|
|
362
|
+
except ValueError:
|
|
363
|
+
obj = None
|
|
364
|
+
if isinstance(obj, dict) and isinstance(obj.get("session"), str) and obj["session"]:
|
|
365
|
+
return obj["session"]
|
|
366
|
+
m = _SESSION_RE.search(body) # a result wrapped in text around the JSON
|
|
367
|
+
return m.group(1) if m else None
|
|
368
|
+
|
|
369
|
+
|
|
370
|
+
def _idle_minutes() -> float:
|
|
371
|
+
raw = _config.load_config().get("agent_idle_minutes", 15)
|
|
372
|
+
if isinstance(raw, bool):
|
|
373
|
+
return 15.0
|
|
374
|
+
try:
|
|
375
|
+
val = float(raw)
|
|
376
|
+
except (TypeError, ValueError):
|
|
377
|
+
return 15.0
|
|
378
|
+
if val < 0:
|
|
379
|
+
return 15.0
|
|
380
|
+
return val
|
|
381
|
+
|
|
382
|
+
|
|
383
|
+
def _watch_idle(run: str, proc: subprocess.Popen, data: dict[str, Any], seen: list[float], limit_minutes: float | None = None) -> None:
|
|
384
|
+
"""Stop an agent that has printed nothing for agent_idle_minutes (it hung,
|
|
385
|
+
or waits on a prompt nobody sees); the run would otherwise stay "running"."""
|
|
386
|
+
minutes = _idle_minutes() if limit_minutes is None else limit_minutes
|
|
387
|
+
if minutes <= 0:
|
|
388
|
+
return
|
|
389
|
+
limit = minutes * 60
|
|
390
|
+
while proc.poll() is None:
|
|
391
|
+
time.sleep(min(5.0, max(0.2, limit / 4)))
|
|
392
|
+
if time.time() - seen[0] > limit and proc.poll() is None:
|
|
393
|
+
data["error"] = f"检索已停止:Agent 超过 {limit / 60:.0f} 分钟没有输出"
|
|
394
|
+
_kill(proc)
|
|
395
|
+
return
|
|
396
|
+
|
|
397
|
+
|
|
398
|
+
def _kill(proc: subprocess.Popen) -> None:
|
|
399
|
+
if os.name == "nt": # the agent's own tool servers go with it
|
|
400
|
+
subprocess.run(["taskkill", "/T", "/F", "/PID", str(proc.pid)], capture_output=True,
|
|
401
|
+
creationflags=getattr(subprocess, "CREATE_NO_WINDOW", 0))
|
|
402
|
+
else:
|
|
403
|
+
proc.terminate()
|
|
404
|
+
|
|
405
|
+
|
|
406
|
+
def _follow(run: str, proc: subprocess.Popen, prompt: str, data: dict[str, Any]) -> None:
|
|
407
|
+
log = _path(run).with_suffix(".log")
|
|
408
|
+
last = 0.0
|
|
409
|
+
seen = [time.time()] # when the agent last printed anything
|
|
410
|
+
idle = _idle_minutes()
|
|
411
|
+
if idle > 0:
|
|
412
|
+
threading.Thread(target=_watch_idle, args=(run, proc, data, seen, idle), daemon=True).start()
|
|
413
|
+
try:
|
|
414
|
+
assert proc.stdin and proc.stdout
|
|
415
|
+
proc.stdin.write(prompt.encode("utf-8"))
|
|
416
|
+
proc.stdin.close()
|
|
417
|
+
with log.open("ab") as fh:
|
|
418
|
+
for raw in proc.stdout:
|
|
419
|
+
seen[0] = time.time()
|
|
420
|
+
fh.write(raw)
|
|
421
|
+
_read(data, raw.decode("utf-8", errors="replace").strip())
|
|
422
|
+
if time.time() - last > 1.0:
|
|
423
|
+
_write(run, data)
|
|
424
|
+
last = time.time()
|
|
425
|
+
code = proc.wait()
|
|
426
|
+
if run in _CANCELLED:
|
|
427
|
+
data["state"] = "cancelled"
|
|
428
|
+
else:
|
|
429
|
+
ok = code == 0 and not data.get("error")
|
|
430
|
+
data["state"] = "done" if ok else "failed"
|
|
431
|
+
if not ok and not data.get("error"):
|
|
432
|
+
data["error"] = f"无法完成检索:Agent 异常退出(退出码 {code})。详情已写入日志"
|
|
433
|
+
except Exception as e:
|
|
434
|
+
data.update(state="failed", error=str(e))
|
|
435
|
+
finally:
|
|
436
|
+
data["doing"] = None
|
|
437
|
+
data.pop("_tools", None)
|
|
438
|
+
data.pop("_last_tool", None)
|
|
439
|
+
data["finished"] = datetime.now().isoformat(timespec="seconds")
|
|
440
|
+
_write(run, data)
|
|
441
|
+
with _LOCK:
|
|
442
|
+
_PROCS.pop(run, None)
|
|
443
|
+
_CANCELLED.discard(run)
|
|
444
|
+
|
|
445
|
+
|
|
446
|
+
def _session_of(result: Any) -> str | None:
|
|
447
|
+
"""The screening session a built-in tool reported (deep search or screening save)."""
|
|
448
|
+
if isinstance(result, dict):
|
|
449
|
+
obj: Any = result
|
|
450
|
+
else:
|
|
451
|
+
try:
|
|
452
|
+
obj = json.loads(str(result or ""))
|
|
453
|
+
except ValueError:
|
|
454
|
+
return None
|
|
455
|
+
if isinstance(obj, dict):
|
|
456
|
+
for key in ("session", "id"):
|
|
457
|
+
value = obj.get(key)
|
|
458
|
+
if isinstance(value, str) and value:
|
|
459
|
+
return value
|
|
460
|
+
return None
|
|
461
|
+
|
|
462
|
+
|
|
463
|
+
def _run_builtin(run: str, data: dict[str, Any], prompt: str) -> None:
|
|
464
|
+
"""Backend B: this process asks the model and runs the tools itself."""
|
|
465
|
+
stop = threading.Event()
|
|
466
|
+
with _LOCK:
|
|
467
|
+
_STOP[run] = stop
|
|
468
|
+
last = [0.0]
|
|
469
|
+
|
|
470
|
+
def on_event(kind: str, ev: dict[str, Any]) -> None:
|
|
471
|
+
if stop.is_set():
|
|
472
|
+
raise _Cancelled()
|
|
473
|
+
if kind == "tool":
|
|
474
|
+
name = str(ev.get("name") or "")
|
|
475
|
+
data["doing"] = _BUILTIN_TOOLS.get(name, f"正在调用 {name}")
|
|
476
|
+
_event(data, data["doing"])
|
|
477
|
+
elif kind == "result":
|
|
478
|
+
session = _session_of(ev.get("result"))
|
|
479
|
+
if session:
|
|
480
|
+
data["session"] = session
|
|
481
|
+
elif kind == "error":
|
|
482
|
+
_event(data, f"{ev.get('name')} 出错:{str(ev.get('detail') or '')[:200]}")
|
|
483
|
+
elif kind == "text":
|
|
484
|
+
data["doing"] = "正在整理结论"
|
|
485
|
+
if time.time() - last[0] > 1.0: # the app polls this file
|
|
486
|
+
_write(run, data)
|
|
487
|
+
last[0] = time.time()
|
|
488
|
+
|
|
489
|
+
try:
|
|
490
|
+
from ..agents import FUNCTIONS, TOOLS, run_task
|
|
491
|
+
from ..agents.system import system_prompt
|
|
492
|
+
|
|
493
|
+
form = data.get("form") or {}
|
|
494
|
+
deep = form.get("mode") == "deep"
|
|
495
|
+
batches = int(form.get("batches") or 0) or 5
|
|
496
|
+
# One batch costs a rate + a next call; building the pool and the final
|
|
497
|
+
# answer need a few more (2026-11-06: the old flat 8 stopped deep runs).
|
|
498
|
+
steps = min(4 + 2 * batches, 24) if deep else 8
|
|
499
|
+
out = run_task(prompt, TOOLS, functions=FUNCTIONS, system=system_prompt(),
|
|
500
|
+
max_steps=steps, on_event=on_event)
|
|
501
|
+
data["result"] = str(out.get("text") or "").strip() or None
|
|
502
|
+
data["state"] = "cancelled" if stop.is_set() else "done"
|
|
503
|
+
except _Cancelled:
|
|
504
|
+
data["state"] = "cancelled"
|
|
505
|
+
except Exception as e:
|
|
506
|
+
data.update(state="failed", error=f"无法完成检索:{type(e).__name__}: {e}")
|
|
507
|
+
finally:
|
|
508
|
+
data["doing"] = None
|
|
509
|
+
data["finished"] = datetime.now().isoformat(timespec="seconds")
|
|
510
|
+
# Mark it finished before the last write: whoever reads the finished
|
|
511
|
+
# state may start the next run at once, while this thread is still
|
|
512
|
+
# alive for a moment and must not count as running.
|
|
513
|
+
with _LOCK:
|
|
514
|
+
_DONE.add(run)
|
|
515
|
+
_write(run, data)
|
|
516
|
+
with _LOCK:
|
|
517
|
+
_BUILTIN.pop(run, None)
|
|
518
|
+
_STOP.pop(run, None)
|
|
519
|
+
_CANCELLED.discard(run)
|
|
520
|
+
_DONE.discard(run)
|
|
521
|
+
|
|
522
|
+
|
|
523
|
+
def cancel(run: str) -> bool:
|
|
524
|
+
with _LOCK:
|
|
525
|
+
proc = _PROCS.get(run)
|
|
526
|
+
thread = _BUILTIN.get(run)
|
|
527
|
+
stop = _STOP.get(run)
|
|
528
|
+
if thread is not None and thread.is_alive():
|
|
529
|
+
_CANCELLED.add(run)
|
|
530
|
+
if stop is not None:
|
|
531
|
+
stop.set() # the loop checks it before the next step
|
|
532
|
+
return True
|
|
533
|
+
if not proc or proc.poll() is not None:
|
|
534
|
+
return False
|
|
535
|
+
_CANCELLED.add(run)
|
|
536
|
+
_kill(proc)
|
|
537
|
+
return True
|