SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsmoney-find

money-find

General & other

View repository

Updated 10-16 05:04

About this skill

/money-find — demand-signal reverse inference + anti-scam screening

Cross-platform note (one-time, not repeated below)

  • Resource paths: below, ../../shared-references/、../../templates/、../../examples/、../../adapters/ is the repo / Claude Code path;For Codex installation, instead read the current skill directory's references/、templates/、examples/、adapters/（install-codex.sh are already symlinked in).
  • Web search/open webpage: Claude Code uses WebSearch / WebFetch; Codex uses its built-in equivalent capabilities.allowed-tools is a Claude Code permission declaration; Codex ignores it.

Core method (read this first)

The main method is the "demand-signal reverse inference method," not "searching for posts about how to make money." No one will post a method that actually makes money in a "teach you how to make money" post—that kind of content is a course-selling/follower-baiting funnel, and its interests are not neutral. So money-find looks atbenefit-neutral demand signals(industry reports/recruitment data/procurement data/policy), then infers opportunities that an individual can supply. For the complete method, see ../../shared-references/demand-signal-method.md, and this skill strictly follows it.

Iron rules

  1. The primary sources are first-order signals(reports/recruitment/procurement/policy),"teach you how to make money"-type posts are not used as evidence, only to understand direction.
  2. Every opportunity must provide a "reasoning chain + cross-validation": which signal it was inferred from and how another independent signal confirms it. A single source = unverified.
  3. Never rely on memory. Signals must all be obtained through real-time web search.
  4. Verify publication dates (rubric C′). Older than 24 months defaults to invalid; 12–24 months should be marked "possibly outdated" and cross-validated. Add the year to search terms.
  5. Every opportunity must first answer "where does the money come from", and pass the three-part check: who has a real need / whether an individual can supply it / whether the individual buyer can be reached.
  6. Every opportunity must pass ../../shared-references/anti-scam-rubric.md(including C′), and only be presented after the verdict is marked.
  7. Converge by profile, not spread out across the full set.

Process

Step 0 — Read context

  • Read .money-state.json to get the profile + profile.tier tier(if none → route to money-init）。
  • Read ../../shared-references/user-tiers.md(route by tier).
  • Read ../../shared-references/demand-signal-method.md(main method).
  • Read ../../shared-references/opportunity-taxonomy.md(opportunity types, for classification assistance only).
  • Read ../../shared-references/anti-scam-rubric.md(anti-scam + timeliness).
  • Read lessons.md(if present): experience accumulated from the user's past reviews; prioritize reference to it (which type worked for him / which pitfalls he has personally experienced).
  • Look at ../../examples/worked-examples.md as a high-quality reasoning example for reference.

Step 1 — Collect demand signals (demand-signal-method Step 1)

First lock the direction scope according to tier(user-tiers): T0→platform crowdsourcing/order-taking; T1→skill services/B-side coaching; T2→small tools/GEO/SaaS; T3→vertical professional services.Don't pitch unreachable things to low tiers.

Real-time web search by direction first-tier sources(all with year added). Besides reports, prioritize grabbing more first-hand sources:

  • real hiring demand(BOSS/Liepin "what's being hired, how much they pay, which AI skills are required" — the most honest demand thermometer: rising positions = someone is paying for this capability, falling = red ocean)
  • transaction side(Xianyu/1688/Taobao "what services are being sold, how many sales" — direct evidence that "someone really pays")
    • ⚠️ Capability boundaries (honestly marked): web search usually can't get past login walls,direct searching only yields second-hand paraphrases, not real-time platform data.
    • ✅ For first-hand data use an adapter(B-tier semi-automatic: the user logs in and searches themselves, the adapter only reads the public list on the current page; the user manually passes the slider captcha):
      • Transaction side → ../../adapters/xianyu(what's being sold/what price/how many people want it)
      • Hiring side → ../../adapters/boss(what's being hired/how much they pay/which AI skills are required;only read the search results list page, don't enter detail pages, don't touch HR personal information, so anti-fraud A6 is not triggered)
  • <行业> AI 落地 需求 / 缺口 / 痛点 报告
  • <工具/服务> 市场规模 增长 采购 B端
  • 平台 官方 创作者 扶持 / 变现 政策 · 政策 / 监管 <领域> 新规

Step 2 — Reasoning + cross-validation (method Step 2-3)

Run each signal through the triad (who has real demand / can an individual supply it / can buyers reach it), andfind at least two independent signals that corroborate each otherbefore pushing forward. For directions where signals are declining, proactively presentdissuasion.

Step 3 — Form a plan + run anti-fraud

Output each plan in the Section D format of anti-scam-rubric.md, and additionally include two lines "reasoning chain" and "cross-validation": reasoning chain / cross-validation / opportunity / where the money comes from / hard red lines / questionable signals / information timeliness / verdict / first verification step / honest income expectation. Write both the first verification step and the income expectation according to the user's tier(user-tiers Step 3); anti-fraud is explained most heavily for T0.

Wording tone: what's given is "inferences/hypotheses with an evidence chain", and the first verification step is to go falsify it. Before the market validates it, don't say "this can make money".

Step 4 — Present + save to disk

  • Present in the order feasible ✅ → questionable ⚠️ → high-risk ❌, each with "where the money comes from" and "first verification step".
  • Write candidates into .money-state.json of opportunities[](including for_tier、verdict、freshness、found_at=today).
  • End with one sentence of expectation management (rubric E), and prompt:
    • "I want to do XX, help me make a plan" → money-plan
    • "Also help me check whether XX is reliable" → money-verify

Note

  • Information is time-sensitive: note in the results "the following is based on a search on <today's date>; platform rules may change, please re-check before starting work".
  • When no credible opportunity can be found,honestly say none reliable was found, don't force a fit. This itself is a valuable anti-fraud conclusion.