This is the most important capability of the whole toolkit:Determining whether an opportunity is actually usable. Better to reject wrongly than to let something slip through.
Cross-platform notes (stated once, not repeated below)
- Resource path: what is written below as
../../shared-references/is the repo / Claude Code path;Under a Codex installation, instead read the current skill directory'sreferences/（install-codex.shis already symlinked in). - Web search/open webpage: Claude Code uses WebSearch / WebFetch; Codex uses its built-in equivalent capabilities.
allowed-toolsis a Claude Code permission declaration; Codex ignores it.
The only standard source
Scoring must strictly follow ../../shared-references/anti-scam-rubric.md。Do not go by intuition, and check the red lines one by one.
Process
Step 1 — Restore the essence of the opportunity
Have the user provide the information they can get (name / recruitment copy / link / how the other party promised). First answer that decisive sentence:"Who pays? Why do they pay? How much?" — if this cannot be answered clearly, it already starts at ⚠️.
Step 2 — Hard red lines (A1–A6)
Check against Section A of the rubric one by one. If any one is hit:
- Verdict High-risk ❌, clearly state which item was hit and why.
- If A4 (funds/accounts handled) or A6 (illegal/gray industry) is hit → additionally warnmay involve illegal or criminal activity, and advise against it.
Step 3 — Pass questionable signals (B1–B6)
Check against Section B of the rubric and list the items hit.
Step 4 — Real-time verification (rubric C)
You must actually search online; you cannot draw conclusions from memory:
- Web search:
<名称> 骗局/<名称> 靠谱吗/<名称> 被骗 投诉(add the year to the search terms). - Check the entity's authenticity: search Qichacha/Tianyancha to see whether the company actually exists and whether there are abnormalities; whether the platform has filed records and whether it uses escrow transactions
- Check reputation: search Black Cat Complaints for real complaints about the platform/company
- Find third-party genuine payment/scam feedback (distinguish advertorials from real people)
- Mandatory drill-down on key numbers: any number to be written into the verdict (income/market/commission) must be checked by opening the primary source text; if it comes only from a search snippet, mark it "unverified" in the verdict.
Step 4.5 — Timeliness check (rubric C′, mandatory)
- Check the publication date of each key source (channel/price/gameplay).
- Over 24 months defaults to invalid and is not used as a basis for "how to do it now"; 12–24 months is marked "possibly outdated" and near-12-month sources are sought for cross-verification.
- In the output, truthfully mark the information year and the risk of being outdated.If it is outdated but not a scam, also remind that "the gameplay may have changed and needs rechecking".
Step 5 — Output (rubric D format)
机会：<名称>
钱从哪来：<谁付钱/为什么/多少>
硬红线：<命中的 A 编号，或"无">
存疑信号：<命中的 B 编号，或"无">
判定：高危 ❌ / 存疑 ⚠️ / 可行 ✅
验证第一步：<今天就能做、低成本、能确认钱真能到账的动作>
诚实收入预期：<前期多久/多少；说明是时间换小钱还是有快钱>
Decision rules: any A hit → high risk; B hit and not clarified → questionable; no A, B clarified, money flow clear → feasible.
Step 6 — Persist
If there is already .money-state.json, write/update this conclusion to opportunities[](including verdict, red_flags, suspect_flags, verify_first_step).
Wrap-up
- Feasible ✅ → suggest using
money-planto make an action plan. - Questionable ⚠️ → give a "first verification step" so the user can try at low cost first, and not rush to invest.
- High risk ❌ → explain clearly where the pitfall is, with the sobering reminder from rubric E.
- In all cases, add one sentence: this conclusion is based on a search as of <today's date>; please verify it yourself in the end, and you bear the risk.