platform-opportunity-scout
Use when the user names a platform or medium (Roblox, YouTube, Instagram, TikTok, Twitch, Etsy, Substack and the like) and wants business ideas built on or around it, or asks what to make, sell or bui
SKILL.mdplugins/platform-opportunity-scout/skills/platform-opportunity-scout/SKILL.md
GitHubPlatform Opportunity Scout
Overview
Finds business ideas on or around one platform, then tries to kill each one before recommending it. Ideas are cheap; the value is in the ones that survive. Everything you report must trace to a source you actually opened this session. Memory is a source of hypotheses, never of facts. A search result, including its summary, is a lead and not a source: only a page you opened counts as one.
Setup
Take the platform from the user. Do not interrogate. State these defaults in one line and proceed, unless they already told you otherwise: solo founder, under $1k to start, a few hours a week, goal is real revenue within 6 months, no particular skills assumed. If they gave a skill, budget or goal, use it as a filter in step 4.
Process
- Map the money. Search for how money moves on the platform right now: platform payouts and revenue share, what creators or developers pay for (tools, freelancers, assets, ads), what audiences pay for, what brands pay for. Note the date of each source. Platform economics change often, so anything older than about 18 months is a hypothesis to re-check, not a fact.
- Generate candidates with the lenses in
references/lenses.md. Aim for 10 to 15, written out one line each before you cut, and printed in the report under Considered with the count; fewer is fine only if the lenses are used up. Cover all three kinds: native (a game, channel, shop or account on the platform), picks-and-shovels (a tool or service sold to people who earn on the platform), and arbitrage (moving something between platforms or audiences). Picks-and-shovels is usually where a solo founder wins, so do not let it get crowded out by "make a popular thing". - Gather evidence for each candidate: who pays, how much, how many of them, what they use today, and what they complain about in their own words (forums, subreddits, app and tool reviews, creator Discords, job boards, freelance listings).
- Cut to 3 to 5 by fit with the user's constraints and by evidence strength. Drop the rest without ceremony.
- Verify. REQUIRED: follow
references/verification.mdfor every survivor, including its independent-verifier step. A task instruction not to delegate does not cancel this step; if you truly cannot run a separate verifier, say in the report that verification was not independent. This is not optional and not a formality. An idea that fails is reported as dead with the reason, which is useful output. - Report in the format below. Verify once; if fewer than two ideas survive, run one second round: rerun the lenses for up to 5 fresh candidates and verify them the same way, with an independent verifier. Do not rerun a candidate that already died.
Report format
Lead with the ranked survivors, best first. For each:
- The idea in one sentence, and which kind it is.
- Who pays and why, with a number if you have one. Any number you could not source may appear only labeled
unverified. - Evidence ledger: each claim, its source link, its date, and a status:
confirmed(two independent sources, or one primary source such as the platform's own docs or a public payout),single-source, orunverified. - Strongest reason it could fail, from the verification pass.
- Competitor check: the searches you ran and what they found. If a source was blocked, say the check is incomplete; never report "no competitors" from an incomplete check.
- Cheapest test the user can run this week for under $100, with the result that would make them stop.
Then a short Killed list: ideas that died in verification, one line each with why. Then say plainly what you could not check, including any source that was blocked (Reddit and Discord often are); name the substitute you used or say there was none. Do not present unverified claims as findings.
Common Mistakes
- Citing a stat from memory ("Roblox pays out $X"). Search it, open the source, check the date.
- Treating platform popularity as demand for the business. Many people watching something is not many people paying for it.
- Reporting an idea as open because your first search found no competitor. Search the obvious names and phrasings at least three ways before saying nothing exists.
- Ranking by excitement. Rank by evidence of payment.
- Recommending a business that depends on a platform feature or policy you did not check. Read the current terms for the part the idea relies on.
- Handing over ten ideas with no verification. Five checked ideas beat ten guessed ones.