mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-15 00:41:59 +00:00
ThearchyHelios
eddeded2a4
- 第一批（安全加固与生命周期止血）：隔离测试宿主机密钥与 .env 泄漏；修复 SQLite 取消查询挂死、:memory: 连接池报错与 EngineCache 世代竞态；补齐运行与评测删除时的 Checkpointer 线程、Artifact 磁盘文件与无主附件回收；加固三套沙箱路径越界、进程组清理与 Seatbelt 数据目录隔离，封堵 SSRF IPv4-mapped IPv6 元数据绕过与响应体 OOM；修复推理模型多轮 reasoning_content、结构化输出 Token 漏计与缓存折扣计费 - 第二批（数据库/集群/问数提速与首屏分包）：补齐核心表索引与 SQLite WAL 调优，列表接口按需 defer 大列，收敛启动扫描并打通跨进程 slot 槽位唤醒；优化跨源全局选表、哈希桶数字绑定、流式 CSV 导入及多方言分析算子；剥离首屏静态引入的 @xyflow/react 与全局样式（首屏主包降至 461.59 kB），补充流式 rAF 攒批、Store 精细订阅与表单无障碍绑定 - 第三批（架构演进与工程自动化闭环）：记录库超挂接阈值历史版本自动增量归档至单库多表（解除 SQLite 10 库 ATTACH 硬上限）；重构知识库为词面 BM25 与全集合归一化向量矩阵真·双路独立召回并支持游标分批重建；升级多方言感知 SQL 模板词法分析（E'...' / $$...$$ / q'[...]' / # / 嵌套注释）与 Decimal / bool / NaN 参数校验，消除 ledger 层对 api 路由层的反向依赖；拆分 3101 行 CaliberDrafts 页面为 7 个领域子模块，消除检查脚本本机路径硬编码并新增 GitHub Actions CI 门禁与三批回归测试套件
498 lines
26 KiB
JavaScript
498 lines
26 KiB
JavaScript
// 界面文案防回潮：静态扫描前后端会上界面的字符串，不开浏览器、不连后端。
|
||
//
|
||
// node scripts/check-copy.mjs 扫一遍，命中禁用写法（且不在白名单里）的报出来
|
||
// node scripts/check-copy.mjs --list 列出全部命中，含被白名单放过的（调白名单时用）
|
||
//
|
||
// 守的是 2026-09 文案整改定下的写法（术语表和语体规范见 frontend/src/lib/terms.ts 开头）：
|
||
// - 「跑」作动词（工作流写「运行」，节点、工具写「执行」）、后端（写「服务端」）、多半、照样、眼下、压根、
|
||
// 取不到、没取回来、读不懂、连不上、写坏了（写「无法 / 未能 / 失败」）、
|
||
// 令牌 / tok / tokens（写「token」）、快照链接（写「分享链接」）、钉住（写「固定版本」「引用版本」）、智能体（写「Agent」）、
|
||
// 供应商（写「模型接入」）、界面上的 Copilot（写「助手」）；
|
||
// - 句尾语气词（吧、呢、哦……）；
|
||
// - 句子里裸写的配置键名和枚举值（report_from、require_citation、cells: true……），要写界面上的中文标签；
|
||
// - 中文句子里的半角标点、英文引号和「...」（用全角标点、「」和「…」）。
|
||
//
|
||
// 扫哪些字符串：
|
||
// - 前端：frontend/src 下的 .ts / .tsx，不含 __tests__、dev/、.d.ts。用 vite 自带的解析器（rolldown 的
|
||
// oxc）读成语法树，取字符串字面量、模板字符串和 JSX 文本，注释天然不在里面。按原文匹配用的字符串
|
||
// （includes / startsWith / === / new RegExp 的参数、case 标签、对象的键、import 路径、类型里的字面量、
|
||
// console 输出）不算界面文案，跳过。
|
||
// - 后端：backend/app 下的 .py，用 Python 的 ast 读，取带中文的字符串和 f-string（docstring、日志、
|
||
// 按原文比较和正则匹配用的跳过）。写给模型的提示词不上界面，按常量名、函数名排除（见 MODEL_ONLY）。
|
||
//
|
||
// 确实该保留的写在 scripts/copy-allow.json：每条写明文件、原文片段、规则和理由。白名单里有一条
|
||
// 已经对不上任何命中（原文改掉了、挪走了），也算不通过：白名单只收现在还需要的例外。
|
||
import { spawnSync } from 'node:child_process'
|
||
import { readdirSync, readFileSync, realpathSync, statSync } from 'node:fs'
|
||
import { createRequire } from 'node:module'
|
||
import { dirname, join, relative } from 'node:path'
|
||
import { fileURLToPath, pathToFileURL } from 'node:url'
|
||
import { resolvePython } from './lib/env.mjs'
|
||
|
||
const HERE = dirname(fileURLToPath(import.meta.url))
|
||
const ROOT = dirname(HERE)
|
||
const FRONT = join(ROOT, 'frontend/src')
|
||
const BACK = join(ROOT, 'backend/app')
|
||
const LIST = process.argv.includes('--list')
|
||
|
||
let failed = 0
|
||
const check = (name, cond, detail = '') => {
|
||
console.log(` ${cond ? '✓' : '✗'} ${name}${detail ? ` — ${detail}` : ''}`)
|
||
if (!cond) failed++
|
||
}
|
||
|
||
// ---------------------------------------------------------------- 规则
|
||
|
||
const CJK = /[㐀-鿿]/
|
||
const H = '[\\u3400-\\u9fff]'
|
||
// 句子里不该裸写的配置键名和枚举值（配置界面字段旁边附的键名是单独的字符串，不带中文，不会命中）
|
||
const KEYS = ['report_from', 'require_citation', 'on_violation', 'on_uncited', 'on_unsupported', 'on_exhausted',
|
||
'allow_numbers', 'caliber_from', 'metrics_from', 'cite_fields', 'upgrade_policy', 'evidence_role', 'max_repairs',
|
||
'max_rounds', 'skip_if', 'budget_tokens', 'workflow_version', 'output_schema',
|
||
// 应用平台的新配置键（人工节点的超时处理、表单来自、派给谁，能力的无人值守写入）：句子里写中文标签
|
||
'on_timeout', 'form_from', 'assignee', 'unattended_write']
|
||
// 前面是「{」的是 str.format 的占位符（{budget_tokens:,}），不算
|
||
const KEY_RE = new RegExp(`(?<![A-Za-z0-9_.${'$'}{])(?:${KEYS.join('|')})(?![A-Za-z0-9_])`
|
||
+ '|cells\\s*:\\s*true|numbers\\s*:\\s*(?:strict|off|flag)\\b|claims\\s*:\\s*(?:off|judge|require_citation)\\b')
|
||
|
||
/**
|
||
* 每条规则：id（白名单按它对）、说明、test(text, prose) → 命中的片段或 null。
|
||
* prose：这段文字带中文，或者是 JSX 文本——只有这样的才按「句子」查英文写法。
|
||
*/
|
||
const RULES = [
|
||
{ id: '跑', what: '「跑」作动词：工作流写「运行」，节点、工具写「执行」', test: (t) => t.match(/.{0,6}跑.{0,6}/)?.[0] },
|
||
{ id: '后端', what: '「后端」：写「服务端」，连接类写「无法连接服务」', test: (t) => t.match(/.{0,6}后端.{0,6}/)?.[0] },
|
||
{
|
||
id: '口语', what: '口语副词和补语：多半、照样、眼下、压根；取不到、没取回来、读不懂、连不上、写坏了',
|
||
test: (t) => t.match(/.{0,6}(?:多半|照样|眼下|压根|取不到|没取回来|读不懂|连不上|写坏了).{0,6}/)?.[0],
|
||
},
|
||
{ id: '令牌', what: '「令牌」：写「token」', test: (t) => t.match(/.{0,6}令牌.{0,6}/)?.[0] },
|
||
// 总方案 X10：不记名的只读链接叫「分享链接」，「快照」专指查询快照
|
||
{ id: '快照链接', what: '「快照链接」：写「分享链接」（「快照」专指查询快照）', test: (t) => t.match(/.{0,6}快照链接.{0,6}/)?.[0] },
|
||
{
|
||
id: 'tok', what: '「tok」「tokens」：写「token」（如「2.8k token」）',
|
||
test: (t, prose) => (prose
|
||
? t.match(/.{0,8}(?<![A-Za-z_])(?:toks?|tokens)(?![A-Za-z_]).{0,4}/)?.[0]
|
||
: t.match(/(?:\d|\{x\})\s*(?:toks?|tokens)(?![A-Za-z_])/)?.[0]),
|
||
},
|
||
{ id: '钉', what: '「钉住」「钉」：写「固定版本」「引用版本」', test: (t) => t.match(/.{0,6}钉.{0,6}/)?.[0] },
|
||
{ id: '智能体', what: '「智能体」：写「Agent」', test: (t) => t.match(/.{0,6}智能体.{0,6}/)?.[0] },
|
||
{ id: '供应商', what: '「供应商」：写「模型接入」', test: (t) => t.match(/.{0,6}供应商.{0,6}/)?.[0] },
|
||
{
|
||
id: 'Copilot', what: '界面上的「Copilot」：写「助手」',
|
||
test: (t, prose) => (prose ? t.match(/.{0,6}copilot.{0,6}/i)?.[0] : t.match(/.{0,6}Copilot.{0,6}/)?.[0]),
|
||
},
|
||
{
|
||
id: '语气词', what: '句尾语气词（吧、呢、哦、啊、呀、嘛、啦）',
|
||
test: (t) => t.match(new RegExp(`.{0,6}${H}[吧呢哦啊呀嘛啦](?=$|[\\s。！？!?，,；;：:」』）)\\]…—])`))?.[0],
|
||
},
|
||
{
|
||
id: '键名', what: '句子里裸写配置键名或枚举值：写界面上的中文标签（如「报告来自」「单元格引用」）',
|
||
test: (t, prose) => (prose && CJK.test(t) ? t.match(KEY_RE)?.[0] : null),
|
||
},
|
||
{
|
||
id: '半角标点', what: '中文句子里的半角标点、英文引号或「...」：用全角标点、「」和「…」',
|
||
test: (t, prose) => {
|
||
if (!prose) return null
|
||
const m = t.match(new RegExp(`.{0,4}(?:${H}[,;!?]|[,;!?]${H}|${H}:(?![/\\d])|:${H}|${H}\\(|\\)${H}`
|
||
+ `|${H}"|"${H}|[“”]|\\.\\.\\.).{0,4}`))
|
||
return m?.[0] ?? null
|
||
},
|
||
},
|
||
]
|
||
|
||
/** 一段文字命中了哪些规则 → [{ rule, snippet }] */
|
||
function lint(text, prose) {
|
||
const out = []
|
||
for (const r of RULES) {
|
||
const hit = r.test(text, prose)
|
||
if (hit) out.push({ rule: r.id, snippet: hit })
|
||
}
|
||
return out
|
||
}
|
||
|
||
// ---------------------------------------------------------------- 前端：取界面字符串
|
||
|
||
// pnpm 的依赖是软链：从 vite 的真实路径出发才解析得到它的依赖 rolldown
|
||
const require = createRequire(realpathSync(join(ROOT, 'frontend/node_modules/vite/package.json')))
|
||
const { parseAst } = await import(pathToFileURL(require.resolve('rolldown/parseAst')).href)
|
||
|
||
// 这些方法的参数是「拿来匹配的原文」，不是显示给人的
|
||
const MATCH_METHODS = new Set(['includes', 'startsWith', 'endsWith', 'indexOf', 'lastIndexOf', 'match', 'matchAll',
|
||
'search', 'split', 'test'])
|
||
const FIRST_ARG_METHODS = new Set(['replace', 'replaceAll'])
|
||
// 不上界面的 JSX 属性
|
||
const HIDDEN_ATTR = /^(?:className|class|style|key|id|role|type|href|to|src|htmlFor|name|value|d|viewBox|fill|stroke|data-.+|aria-hidden|aria-controls|aria-labelledby|aria-describedby)$/
|
||
|
||
function memberName(n) {
|
||
if (!n) return ''
|
||
if (n.type === 'Identifier') return n.name
|
||
if (n.type === 'MemberExpression' && !n.computed) return `${memberName(n.object)}.${n.property.name}`
|
||
return ''
|
||
}
|
||
|
||
/** 这个字面量是不是「按原文匹配 / 代码里的键」而不是界面文字 */
|
||
function skipLiteral(node, parent, key) {
|
||
if (!parent) return false
|
||
switch (parent.type) {
|
||
case 'ImportDeclaration': case 'ExportNamedDeclaration': case 'ExportAllDeclaration': case 'ImportExpression':
|
||
case 'TSLiteralType': case 'TSExternalModuleReference': case 'TSImportType':
|
||
return true
|
||
case 'Property': case 'PropertyDefinition': case 'MethodDefinition': case 'TSPropertySignature':
|
||
return key === 'key' && !parent.computed
|
||
case 'BinaryExpression':
|
||
return ['===', '!==', '==', '!=', 'in'].includes(parent.operator)
|
||
case 'SwitchCase':
|
||
return key === 'test'
|
||
case 'JSXAttribute':
|
||
return HIDDEN_ATTR.test(parent.name?.name ?? '')
|
||
case 'NewExpression':
|
||
case 'CallExpression': {
|
||
if (key !== 'arguments') return false
|
||
const callee = parent.callee
|
||
if (callee.type === 'Identifier' && callee.name === 'RegExp') return true
|
||
if (callee.type === 'MemberExpression' && !callee.computed) {
|
||
const m = callee.property.name
|
||
if (MATCH_METHODS.has(m)) return true
|
||
if (FIRST_ARG_METHODS.has(m) && parent.arguments[0] === node) return true
|
||
if (memberName(callee.object) === 'console') return true
|
||
}
|
||
return false
|
||
}
|
||
default:
|
||
return false
|
||
}
|
||
}
|
||
|
||
function lineOf(src) {
|
||
const starts = [0]
|
||
for (let i = 0; i < src.length; i++) if (src.charCodeAt(i) === 10) starts.push(i + 1)
|
||
return (pos) => {
|
||
let lo = 0, hi = starts.length - 1
|
||
while (lo < hi) {
|
||
const mid = (lo + hi + 1) >> 1
|
||
if (starts[mid] <= pos) lo = mid; else hi = mid - 1
|
||
}
|
||
return lo + 1
|
||
}
|
||
}
|
||
|
||
/** 一段 TS / TSX 源码里会上界面的字符串 → [{ line, text, prose }] */
|
||
export function frontStrings(src, lang = 'tsx') {
|
||
const ast = parseAst(src, { lang })
|
||
const at = lineOf(src)
|
||
const out = []
|
||
const visit = (node, parent, key) => {
|
||
if (!node || typeof node.type !== 'string') return
|
||
if (node.type === 'Literal' && typeof node.value === 'string') {
|
||
if (!skipLiteral(node, parent, key)) out.push({ line: at(node.start), text: node.value, prose: CJK.test(node.value) })
|
||
return
|
||
}
|
||
if (node.type === 'TemplateLiteral') {
|
||
if (parent?.type !== 'TaggedTemplateExpression' && !skipLiteral(node, parent, key)) {
|
||
const text = node.quasis.map((q) => q.value.cooked ?? q.value.raw).join('{x}')
|
||
out.push({ line: at(node.start), text, prose: CJK.test(text) })
|
||
}
|
||
for (const e of node.expressions) visit(e, node, 'expressions')
|
||
return
|
||
}
|
||
if (node.type === 'JSXText') {
|
||
const text = node.value.replace(/\s+/g, ' ').trim()
|
||
if (text) out.push({ line: at(node.start), text, prose: true })
|
||
return
|
||
}
|
||
for (const k of Object.keys(node)) {
|
||
if (k === 'parent') continue
|
||
const v = node[k]
|
||
if (Array.isArray(v)) for (const c of v) visit(c, node, k)
|
||
else if (v && typeof v === 'object' && typeof v.type === 'string') visit(v, node, k)
|
||
}
|
||
}
|
||
visit(ast, null, null)
|
||
return out
|
||
}
|
||
|
||
function walk(dir, keep, out = []) {
|
||
for (const name of readdirSync(dir)) {
|
||
const p = join(dir, name)
|
||
if (statSync(p).isDirectory()) walk(p, keep, out)
|
||
else if (keep(p)) out.push(p)
|
||
}
|
||
return out
|
||
}
|
||
|
||
const frontFiles = walk(FRONT, (p) => /\.tsx?$/.test(p) && !/\.d\.ts$/.test(p)
|
||
&& !/\/__tests__\//.test(p) && !/\/dev\//.test(p) && !/\.test\.tsx?$/.test(p))
|
||
|
||
// ---------------------------------------------------------------- 后端：取界面字符串
|
||
|
||
// 写给模型的提示词：不上界面（backend-surfaced 分片按要求未改）。按赋值的常量名、所在函数名排除。
|
||
// 既给人看又交回模型的，给人看的那句单独存在（for_model、_UNKNOWN_ENTITY_FOR_MODEL 是模型那份）
|
||
const MODEL_ONLY = {
|
||
// 模块级常量：系统提示、输出协议、写作 / 裁判 / 复核规则、交给模型的 JSON Schema、正则片段
|
||
names: new Set([
|
||
'NODE_REFERENCE', 'ROLE', 'MARKUP_NUDGE', 'JUDGE_RULES', 'JUDGE_ROLE', 'JUDGE_RULE', 'REVIEW_SYSTEM', 'MARKER_RULES',
|
||
'GATE_SYSTEM', 'TOOL_MARKUP_NUDGE', '_UNKNOWN_ENTITY_FOR_MODEL', 'NODE_FIELD_FIX', '_FIX_TAIL',
|
||
'_STREAM_PROTOCOL', '_RULES_HEAD', '_NO_LOWERING', '_ASK_HUMAN', '_UPGRADE_TASK', 'CELL_RULES', 'ENTITY_RULES',
|
||
'EXTRACT_SYSTEM', 'GRAPH_SCHEMA', 'REVIEW_SCHEMA', '_CN_N', 'OMITTED_RULES', 'ABSENCE_RULE',
|
||
]),
|
||
// 拼提示词的函数：生成 / 修正 / 升级请求、抽取消息、交给抽取模型的输出结构；
|
||
// 助手的数据源上下文（api/copilot_context.py 的各段渲染、挑表请求，copilot.py 里挑表用的需求）
|
||
funcs: new Set(['_user_message', '_repair_request', '_publish_fix_request', '_upgrade_assist_request',
|
||
'_extract_messages', '_from_rows', 'catalog_prompt', 'writer_catalog', '_trimmed_catalog', '_metric_prompt',
|
||
'_fields_prompt', '_field_prompt', '_query_brief',
|
||
'_pick_messages', '_pick_need', '_publish_fix_need', '_upgrade_need',
|
||
'_detail_block', '_selected_block', '_join_lines', '_route_line', '_source_head']),
|
||
// 第二个参数是交回模型改写的指令（第一个是给人看的那句）：engine/evidence.py 的 _Reason(text, fix)、fail(text, fix)
|
||
fixArg: new Set(['_Reason', 'fail']),
|
||
names_re: /(?:^|_)(?:SYSTEM|PROMPT|NUDGE)(?:_|$)/,
|
||
}
|
||
|
||
const PY_EXTRACT = String.raw`
|
||
import ast, json, os, re, sys
|
||
CJK = re.compile(r'[㐀-鿿]')
|
||
root = sys.argv[1]
|
||
only = sys.argv[2:] # 给了就只读这几个文件（自检用）
|
||
out = []
|
||
def dotted(n):
|
||
if isinstance(n, ast.Name): return n.id
|
||
if isinstance(n, ast.Attribute):
|
||
d = dotted(n.value)
|
||
return (d + '.' if d else '') + n.attr
|
||
return ''
|
||
def scan(path, rel):
|
||
src = open(path, encoding='utf-8').read()
|
||
tree = ast.parse(src)
|
||
for node in ast.walk(tree):
|
||
for child in ast.iter_child_nodes(node):
|
||
child._p = node
|
||
skip = set()
|
||
for node in ast.walk(tree):
|
||
# docstring 和当注释用的裸字符串
|
||
if isinstance(node, ast.Expr) and isinstance(node.value, (ast.Constant, ast.JoinedStr)):
|
||
skip.add(id(node.value))
|
||
def ctx(node):
|
||
funcs, assign, call, kw, arg, compare, matcher, logger, dictkey = [], None, None, None, None, False, False, False, None
|
||
cur, prev = node, None
|
||
while hasattr(cur, '_p'):
|
||
prev, cur = cur, cur._p
|
||
if isinstance(cur, (ast.FunctionDef, ast.AsyncFunctionDef, ast.ClassDef)):
|
||
funcs.append(cur.name)
|
||
elif isinstance(cur, (ast.Assign, ast.AnnAssign, ast.AugAssign)) and assign is None:
|
||
targets = cur.targets if isinstance(cur, ast.Assign) else [cur.target]
|
||
for t in targets:
|
||
name = dotted(t) or (dotted(t.value) if isinstance(t, ast.Subscript) else '')
|
||
if name:
|
||
assign = name.split('.')[-1]
|
||
break
|
||
elif isinstance(cur, ast.Compare) and call is None:
|
||
compare = True
|
||
elif isinstance(cur, ast.Dict) and dictkey is None and prev in cur.values:
|
||
k = cur.keys[cur.values.index(prev)]
|
||
if isinstance(k, ast.Constant) and isinstance(k.value, str): dictkey = k.value
|
||
elif isinstance(cur, ast.keyword) and kw is None and call is None:
|
||
kw = cur.arg
|
||
elif isinstance(cur, ast.Call) and call is None:
|
||
call = dotted(cur.func)
|
||
if prev in cur.args: arg = cur.args.index(prev)
|
||
last = call.split('.')[-1]
|
||
head = call.split('.')[0]
|
||
if last in ('startswith', 'endswith', 'find', 'rfind', 'index', 'count', 'split', 'rsplit',
|
||
'partition', 'removeprefix', 'removesuffix', 'strip', 'lstrip', 'rstrip') \
|
||
or (head == 're' and last in ('compile', 'search', 'match', 'fullmatch', 'sub', 'subn',
|
||
'findall', 'finditer', 'split')) \
|
||
or (last == 'replace' and cur.args and prev is cur.args[0]):
|
||
matcher = True
|
||
if head in ('logger', 'log', '_log', 'LOG', 'logging', 'warnings') or last in ('debug', 'exception'):
|
||
logger = True
|
||
return dict(funcs=funcs[::-1], assign=assign, call=call, kw=kw, arg=arg, compare=compare, matcher=matcher,
|
||
logger=logger, dictkey=dictkey)
|
||
for node in ast.walk(tree):
|
||
if id(node) in skip: continue
|
||
if isinstance(node, ast.JoinedStr):
|
||
text = ''.join(str(v.value) if isinstance(v, ast.Constant) else '{x}' for v in node.values)
|
||
elif isinstance(node, ast.Constant) and isinstance(node.value, str):
|
||
if isinstance(getattr(node, '_p', None), (ast.JoinedStr, ast.FormattedValue)): continue
|
||
text = node.value
|
||
else:
|
||
continue
|
||
if not CJK.search(text) and 'Copilot' not in text: continue
|
||
out.append(dict(file=rel, line=node.lineno, text=text, **ctx(node)))
|
||
files = []
|
||
if only:
|
||
files = [(p, os.path.basename(p)) for p in only]
|
||
else:
|
||
for d, _, fs in os.walk(root):
|
||
for f in sorted(fs):
|
||
if f.endswith('.py'):
|
||
p = os.path.join(d, f)
|
||
files.append((p, os.path.relpath(p, root)))
|
||
for p, rel in files:
|
||
scan(p, rel)
|
||
print(json.dumps(out, ensure_ascii=False))
|
||
`
|
||
|
||
const PYTHON = resolvePython()
|
||
|
||
function backStrings(root, only = []) {
|
||
const r = spawnSync(PYTHON, ['-c', PY_EXTRACT, root, ...only], { encoding: 'utf8', maxBuffer: 256 << 20 })
|
||
if (r.status !== 0) throw new Error(`读后端源码失败：${r.stderr || r.error}`)
|
||
return JSON.parse(r.stdout)
|
||
}
|
||
|
||
/** 后端字符串是不是会上界面的那种（不是日志、不是匹配用的原文、不是写给模型的提示词） */
|
||
function backShown(s) {
|
||
if (s.logger || s.matcher || s.compare) return false
|
||
if (s.assign && (MODEL_ONLY.names.has(s.assign) || MODEL_ONLY.names_re.test(s.assign))) return false
|
||
if (s.call && MODEL_ONLY.fixArg.has(s.call.split('.').pop()) && s.arg === 1) return false
|
||
// 接口参数说明只进 OpenAPI 文档，不上界面
|
||
if (s.call && /(?:^|\.)(?:Query|Path|Body|Header)$/.test(s.call) && s.kw === 'description') return false
|
||
if (s.funcs.some((f) => MODEL_ONLY.funcs.has(f))) return false
|
||
if (s.kw === 'fix' || s.kw === 'for_model' || s.dictkey === 'for_model') return false
|
||
return true
|
||
}
|
||
|
||
// ---------------------------------------------------------------- 汇总
|
||
|
||
const hits = []
|
||
for (const file of frontFiles) {
|
||
const rel = `frontend/src/${relative(FRONT, file)}`
|
||
const src = readFileSync(file, 'utf8')
|
||
let strings
|
||
try {
|
||
strings = frontStrings(src, file.endsWith('.tsx') ? 'tsx' : 'ts')
|
||
} catch (e) {
|
||
check(`读得懂 ${rel}`, false, String(e?.message ?? e).slice(0, 200))
|
||
continue
|
||
}
|
||
for (const s of strings) for (const h of lint(s.text, s.prose)) hits.push({ file: rel, line: s.line, text: s.text, ...h })
|
||
}
|
||
const backAll = backStrings(BACK)
|
||
const backShownList = backAll.filter(backShown)
|
||
for (const s of backShownList) {
|
||
const where = [s.funcs.join('.'), s.assign && `= ${s.assign}`, s.call && `${s.call}(${s.kw ? `${s.kw}=` : ''})`,
|
||
s.dictkey && `{${s.dictkey}}`].filter(Boolean).join(' ')
|
||
for (const h of lint(s.text, true)) hits.push({ file: `backend/app/${s.file}`, line: s.line, text: s.text, where, ...h })
|
||
}
|
||
|
||
const ALLOW_PATH = join(HERE, 'copy-allow.json')
|
||
const allow = JSON.parse(readFileSync(ALLOW_PATH, 'utf8')).entries
|
||
const used = new Set()
|
||
const allowed = (h) => allow.findIndex((a, i) => {
|
||
const ok = a.file === h.file && (a.rule === h.rule || a.rule === '*') && h.text.includes(a.text)
|
||
if (ok) used.add(i)
|
||
return ok
|
||
}) >= 0
|
||
const bad = hits.filter((h) => !allowed(h))
|
||
|
||
console.log('=== 扫描范围 ===')
|
||
check(`前端 ${frontFiles.length} 个文件都读得懂`, frontFiles.length > 60)
|
||
check(`后端取到 ${backAll.length} 条带中文的字符串，其中 ${backShownList.length} 条按界面文案查`,
|
||
backAll.length > 1000 && backShownList.length > 500)
|
||
|
||
console.log('\n=== 禁用写法 ===')
|
||
for (const r of RULES) {
|
||
const mine = bad.filter((h) => h.rule === r.id)
|
||
check(`${r.what}：没有`, mine.length === 0,
|
||
mine.slice(0, 40).map((h) => `${h.file}:${h.line}「${h.snippet}」`).join(' ｜ ')
|
||
+ (mine.length > 40 ? ` ｜ ……另有 ${mine.length - 40} 处` : ''))
|
||
}
|
||
|
||
console.log('\n=== 给模型的改写指令（for_model）不上界面 ===')
|
||
// 后端在报告核对的 violations、助手自查的问题里另带 for_model：交回模型用的原话。前端在请求层（api/http.ts）的
|
||
// parseBody 里一进来就去掉，别处不该再碰它——原始事件、工件原文这些直接显示 JSON 的地方才不会露出来。
|
||
// REST、上传的入口在 http.ts，助手流、运行事件流在 client.ts，两个文件合起来查
|
||
{
|
||
const transport = ['api/http.ts', 'api/client.ts'].map((f) => join(FRONT, f))
|
||
const readers = frontFiles.filter((f) => !transport.includes(f) && readFileSync(f, 'utf8').includes('for_model'))
|
||
check('前端只有 api/http.ts、api/client.ts 提到 for_model', readers.length === 0, readers.map((f) => relative(ROOT, f)).join('、'))
|
||
const code = transport.map((f) => readFileSync(f, 'utf8')).join('\n')
|
||
check('api/http.ts、api/client.ts：响应体一律经 parseBody 解析（不直接 res.json()），REST、上传、助手流、运行事件流四个入口都在',
|
||
!/\.json\(\)/.test(code) && (code.match(/parseBody\(/g) ?? []).length >= 5 && /delete[^\n]*MODEL_NOTE/.test(code))
|
||
}
|
||
|
||
console.log('\n=== 白名单 ===')
|
||
const bare = allow.filter((a) => !a.file || !a.text || !a.rule || !a.reason || !RULES.some((r) => r.id === a.rule || a.rule === '*'))
|
||
check(`${allow.length} 条都写明了文件、原文片段、规则和理由`, bare.length === 0, bare.map((a) => JSON.stringify(a)).join(' ｜ '))
|
||
const stale = allow.filter((_, i) => !used.has(i))
|
||
check('每条都还对得上一处命中（失效的要删掉）', stale.length === 0,
|
||
stale.map((a) => `${a.file}「${a.text}」(${a.rule})`).join(' ｜ '))
|
||
|
||
if (LIST) {
|
||
console.log('\n=== 全部命中（--list）===')
|
||
for (const h of hits) {
|
||
console.log(` ${allowed(h) ? '·' : '!'} [${h.rule}] ${h.file}:${h.line}「${h.snippet}」${h.where ? ` @ ${h.where}` : ''}`
|
||
+ ` ← ${JSON.stringify(h.text).slice(0, 120)}`)
|
||
}
|
||
}
|
||
|
||
// ---------------------------------------------------------------- 自检：扫描器自己认得对不对
|
||
|
||
console.log('\n=== 扫描器自检（拿构造的源码喂一遍）===')
|
||
{
|
||
const src = [
|
||
"// 注释里写接着跑不算",
|
||
"/* 块注释：后端 */",
|
||
"const a = <button title=\"接着跑\">接着跑</button>",
|
||
"const b = '后端没有响应'",
|
||
"const c = `已用 ${n} tok`",
|
||
"const d = /跑完/.test(x)",
|
||
"const e = x.includes('收尾轮仍想调用工具，接着跑')",
|
||
"const f = x === '在跑'",
|
||
"const g = { '可续跑': 1 }",
|
||
"const h = <div className=\"跑\" data-x=\"跑\" aria-label=\"不能再跑\" />",
|
||
"const i = '让 Copilot 改'",
|
||
"const j = '好的吧。'",
|
||
"const k = '按 report_from 取'",
|
||
"const l = '提示:请稍后...'",
|
||
"const m = '保存并继续运行'",
|
||
"import z from './跑.ts'",
|
||
"switch (y) { case '跑': break }",
|
||
"console.log('后端')",
|
||
].join('\n')
|
||
const strings = frontStrings(src)
|
||
const flagged = (needle, rule) => strings.some((s) => s.text.includes(needle) && lint(s.text, s.prose).some((h) => h.rule === rule))
|
||
const seen = (needle) => strings.some((s) => s.text.includes(needle))
|
||
check('JSX 文本、JSX 属性里的「接着跑」报出来', strings.filter((s) => s.text === '接着跑').length === 2 && flagged('接着跑', '跑'))
|
||
check('普通字符串里的「后端」、模板字符串里的「tok」报出来', flagged('后端没有响应', '后端') && flagged('已用 {x} tok', 'tok'))
|
||
check('读屏名（aria-label）按界面文案查', flagged('不能再跑', '跑'))
|
||
check('Copilot、句尾语气词、裸键名、半角标点都报出来', flagged('让 Copilot 改', 'Copilot') && flagged('好的吧', '语气词')
|
||
&& flagged('按 report_from 取', '键名') && flagged('提示:请稍后', '半角标点'))
|
||
check('注释、正则、includes / === 的原文、对象键、import 路径、case 标签、className / data-*、console 都不算',
|
||
!seen('注释') && !seen('块注释') && !seen('跑完') && !seen('收尾轮') && !seen('在跑') && !seen('可续跑')
|
||
&& !strings.some((s) => s.text === '跑') && !seen('./跑.ts') && !strings.some((s) => s.text === '后端'))
|
||
check('合规的文案不误报', !lint('保存并继续运行', true).length && !lint('56.0k token', false).length
|
||
&& !lint('已达上限（3 句），这句未裁判', true).length)
|
||
check('「快照链接」报出来，「分享链接」「查询快照」不误报', lint('复制快照链接', true).some((h) => h.rule === '快照链接')
|
||
&& !lint('复制分享链接', true).length && !lint('打开查询快照', true).length)
|
||
|
||
const py = [
|
||
'from fastapi import HTTPException',
|
||
'NODE_REFERENCE = "给模型：接着跑"',
|
||
'def f(x):',
|
||
' """docstring：后端"""',
|
||
' logger.info("后端日志")',
|
||
' if "接着跑" in x: pass',
|
||
' if x.startswith("跑"): pass',
|
||
' raise HTTPException(409, detail="后端还在跑")',
|
||
' return {"message": f"已用 {x} 令牌", "for_model": "照样重试"}',
|
||
].join('\n')
|
||
const tmp = join(process.env.TMPDIR ?? '/tmp', `check-copy-selftest-${process.pid}.py`)
|
||
const { writeFileSync, rmSync } = await import('node:fs')
|
||
writeFileSync(tmp, py)
|
||
try {
|
||
const got = backStrings(dirname(tmp), [tmp]).filter(backShown)
|
||
const texts = got.map((s) => s.text)
|
||
check('后端：HTTPException 的 detail、返回给界面的 message 报出来',
|
||
texts.includes('后端还在跑') && texts.includes('已用 {x} 令牌')
|
||
&& lint('后端还在跑', true).some((h) => h.rule === '跑') && lint('已用 {x} 令牌', true).some((h) => h.rule === '令牌'))
|
||
check('后端：docstring、日志、in / startswith 的原文、写给模型的常量和 for_model 都不算',
|
||
!texts.some((t) => /docstring|日志|给模型|照样/.test(t)) && !texts.includes('接着跑') && !texts.includes('跑'))
|
||
} finally {
|
||
rmSync(tmp, { force: true })
|
||
}
|
||
}
|
||
|
||
console.log(failed ? `\n✗ ${failed} 项未通过` : '\n✓ 文案全部通过')
|
||
process.exit(failed ? 1 : 0)
|