Install
openclaw skills install @cod1doc/audit-nanjing-overseas-agencies南京留学、移民、海外身份及因私出入境机构的中英双语证据审计;用于全量列举南京机构、核验资质或合同主体、横向比较与评分、审计指定机构、识别初创或渠道型中介、调查客诉与营销话术,以及起草防坑条款。Bilingual evidence-led audit of Nanjing study-abroad, immigration, overseas-identity, and private-entry/exit agencies; use for full rosters, entity and qualification checks, comparisons, scoring, startup/
openclaw skills install @cod1doc/audit-nanjing-overseas-agencies执行消费者立场、来源可追溯的审计。保持直接,但不得把怀疑、规模小、成立时间短或网络曝光少写成事实指控。
Produce a consumer-side, source-traceable audit. Be direct, but never turn suspicion, small size, recent registration, or low web exposure into an allegation.
中文 / English in each cell.不得直接接受“现行双持证机构”的说法。基线日期下,中国已取消自费留学中介资格和因私出入境中介许可证。每次审计都必须从最新官方来源重新核验,因为法律会变化。
Do not accept “currently dual-licensed agency” at face value. At the baseline date, China had cancelled both former special-licence regimes. Reverify current official law during every audit because regulation changes.
严格区分 / Separate:
不得因规模小就称其为“无资质作坊”。两项特别许可取消后,真正有意义的风险类别是主体、合同、资金或交付链条异常。
Never call a small operator an “unlicensed workshop” merely for its size. After cancellation of the two special permits, the meaningful category is an irregular entity, contract, payment, or delivery chain.
记录用户提供的 / Capture when supplied:
若背景缺失,继续盘点或机构审计,但明确个人匹配度和申请概率不可测;不得编造成功率。
If the profile is missing, continue the roster or institution audit, but state that personal fit and application probability are not measurable. Never invent a success percentage.
写明审计日期及“全部”的确切范围。使用工商登记查询与历史名单,不使用搜索热度替代全量。无法取得官方批量名录时,说明当前市场召回不完整,并记录检索过程。
State the audit date and the exact meaning of “all.” Use business-registration queries plus historical lists, not search popularity. If no official bulk roster is accessible, disclose incomplete recall and document the searches.
至少检索 / Search at minimum:
目录、地图、社媒和广告仅用于发现线索;每个品牌都要解析到法人主体。
Use directories, maps, social media, and advertisements only for discovery. Resolve every brand to a legal entity.
逐个独立法人记录 / Record for every independent legal entity:
按统一社会信用代码去重,不按品牌去重;共享品牌的不同法人必须各占一行。
Deduplicate by unified social credit code, not brand. Keep different legal entities on separate rows even when they share a brand.
使用法律基线中的证据等级和措辞规则;可变信息必须检索最新官方来源,并附访问日期。
Use the evidence grades and wording rules in the legal baseline. Browse current official sources for mutable claims and attach access dates.
成立不满三年的主体进入加强审计,但要分别核验:参保与用工、留学是否只是身份业务配套、文书/案管归属、渠道依赖、可比案例、境外持牌顾问、大额第三方付款、佣金退款链、持续经营保障及政策不利面披露。
Flag entities established within three years for enhanced review, then separately verify staffing, whether study abroad is accessory to identity work, writer/case-manager ownership, channel dependence, comparable cases, overseas regulated advisers, third-party payments, commissions/refunds, continuity, and downside disclosure.
未知事项只能标“待核 / unverified”并转化为合同条件,不得写成既成短板。
Convert unknowns into “待核 / unverified” and contract conditions, never factual shortcomings.
仅在取得证据后分类 / Classify only after evidence collection:
依据 评分规则 / scoring-rubric.md,从10分起只扣有来源支持的项目;逐项展示事实、来源、等级、日期和分值。
Use the scoring rubric. Start at 10 and subtract only source-supported items; show fact, source, grade, date, and points for each deduction.
同时计算证据覆盖率。低于60%时标记“低证据覆盖的暂定分,不可排名 / provisional low-coverage score; not rankable”。未知不等于违规,不得用不均衡数据制造完整排名。
Also calculate evidence coverage. Below 60%, label the result as provisional and not rankable. Unknown evidence is not misconduct; never manufacture a ranking from uneven data.
结构化审计可保存为JSON并运行 / For structured audits, save JSON and run:
python scripts/validate_audit.py path\to\audit.json
发布前修复全部错误。/ Fix every validation error before publishing.
完整审计必须依次输出 / A full audit must follow:
使用 报告模板 / report-template.md。每个法人独占一行和一个深度透视块,不得合并小型、初创、移民导向、低曝光或渠道型主体。
Use the report template. Give every entity its own row and deep-dive block; never merge small, new, migration-led, low-exposure, or channel-based entities.