market-scan
Skill2 starsChanged 4 months agoMITView on GitHub
Scan the public market for evidence on a product hypothesis — competitors and their pricing/coverage (supply lens) and what real users say about the pain in forums, app reviews, FB groups, YouTube/Reddit (demand lens). Spawns a fresh research subagent from a fixed brief template + source playbook, codes findings into ranked failure modes, and returns an artifact that drops into /evidence-ledger as signal rows. Use when you need to test "does a solution already exist", "is this an acute would-pay pain", "what do users of competitor X complain about", or to gather demand/supply signal before building. The discovery-research companion to /evidence-ledger. NOT a substitute for interviews or a willingness-to-pay test. Russian triggers: "просканируй рынок", "что говорят пользователи", "найди конкурентов и отзывы", "есть ли уже решение", "собери сигнал по гипотезе", "голос пользователя", "voice of customer".
5 other projects use a version of this.
Adopted by
Most recently taken up by
- TrueScrape/truescrapetheir own version, first seen Nov 8, 2026
- Latifox/find-me-saastheir own version, first seen Sep 6, 2026
- tjdwls101010/Harness-of-Minervinitheir own version, first seen Aug 18, 2026
- AIDevGTM/gtm-cofoundertheir own version, first seen Aug 17, 2026
- InfiniteRoomLabs/agent-opstheir own version, first seen Jun 23, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- AIDevGTM/gtm-cofounder★ 340same name
- Latifox/find-me-saas★ 10same name
- InfiniteRoomLabs/agent-ops★ 1same name
- tjdwls101010/Harness-of-Minervini★ 1same name
- TrueScrape/truescrape★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository11 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- designSkill
- diagramSkill
- evidence-ledgerSkill
- motionSkill
- prototypeSkill
- reframeSkill
- retroSkill
- review-motionSkill
- speechSkill
- sprite-animSkill
The file
We split this skill into the rules it gives an agent
1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 93
Run it as a background subagent (these take minutes).
“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/market-scan — hypothesis → coded market signal
Takes a product hypothesis and sweeps public sources for evidence, then returns coded, ranked, source-tagged signal. It does not run the experiment or make the decision — it gathers found text. The orchestrator fills the brief template below and delegates to a fresh research subagent (the "researcher" — it gets web tools and no drafting context).
The engine, repeated at every stage: found public text is hypothesis-SHAPING, never validating. It cannot conclude willingness-to-pay; only a dry-wallet / fake-door test or real interviews can. Absence of signal is itself a finding. A vendor's marketing, a firm's voice, and a homeowner's complaint are not your target user's voice.
Pairs with /evidence-ledger: market-scan produces signal → the ledger records it as a
signal row and decides (kill/pivot/double-down) only with a behavioural test.
Different from /deep-research (fan-out + adversarial fact-check → cited report):
market-scan adds the two-lens frame (§2), error-analysis coding of voices into ranked
failure modes (§4), and discovery honesty (found-text ≠ WTP, absence-as-finding,
selection-bias) — disciplines /deep-research has no reason to carry.
Argument (optional): the hypothesis in any convenient form (a sentence, bullets). If absent, use the bet under discussion.
1. Intake — normalize the loose ask
Pull these from the input; fill gaps with stated defaults, don't block on them:
- Hypothesis / premise — what we believe and want to confirm or kill.
- Product & segment — and precisely who the target user is (e.g. "SOLO/micro plumber", not "trades").
- Geo + language — search language(s); name them (the scan reads in that language).
- Lens —
supply,demand, orboth(§2). - Feeds which
/evidence-ledgerassumption — name the row(s) it informs. - Seeds — any known competitors / sources to start from.
2. The two lenses (one playbook)
- Supply — what solutions/competitors exist, their coverage of the job(s), pricing, and geo/language fit. Answers: is the "no convenient solution" premise true? Beware: a tool existing ≠ the target adopting it conveniently — say which you've shown.
- Demand — what real target users say about the pain in their own words. Answers: is this an acute, would-pay pain? Is there integration pain across current tools?
3. Source playbook (what to pull from each)
- Google / search aggregators — entry points, "best X for Y", comparison posts.
- Competitor sites + pricing pages — coverage, price tiers, who they target (solo vs firm).
- App reviews — Google Play, App Store, Capterra, G2 — why users adopted / churned / what's missing. Highest-signal demand source after interviews.
- Vertical forums — peer-to-peer pain talk in the segment's own words.
- Facebook groups — often the richest peer vein, but usually login-walled → flag as inaccessible, don't fake it.
- Reddit / YouTube vlog comments — "day in the life" narration of real pain.
- Marketplaces / lead-gen sites — distinguish lead-gen language ("free slots!") from pain — they are not the same.
4. Method — error-analysis coding (the rigor)
- Collect real quotes (original language and translation).
- Open-code each fragment into a short failure-mode label.
- Axial-code — group labels into named failure modes.
- Rank — demand: by # distinct target voices per mode; supply: by coverage of the target job at the target price point. Tag each mode/assumption with an evidence-strength rung: strong / moderate / weak / absent.
- Stop at saturation — when new sources stop surfacing new modes; state the
source/search count at which it saturated.
A lens or job that surfaced zero target-voice fragments is recorded as an explicit
absentrow, never omitted. Throughout: attribute every voice — only the target segment counts as demand signal. A non-target voice (firm/homeowner/vendor) is context — except when it runs counter to the premise (e.g. an adjacent group documenting against your target user): that direction is a finding, not noise — report it.
5. Delegation brief template (fill and send to a fresh subagent)
You are a [supply scanner | qualitative VoC researcher] doing a public-source scan of the
[GEO] market in [LANGUAGE]. Use WebSearch/WebFetch (load via ToolSearch:
"select:WebSearch,WebFetch"); search in [LANGUAGE].
WHO (exact target): [precise solo/segment definition — NOT the broad category].
HYPOTHESIS / JOB(S): [the premise + the specific job(s) to probe].
WHY: this feeds assumption [A-n] in a product evidence-ledger. Confirm or KILL it. Be
adversarial toward the founder's premise. [State what a kill looks like.]
WHERE: [the §3 playbook, narrowed to this hypothesis — name concrete competitors/forums to seed].
METHOD: [§4 coding — quotes (orig+translation) → open-code → failure modes → rank + strength rung → stop at saturation and **state the saturation count**].
DELIVERABLE: [the §6 output contract — incl. a per-assumption directional read with evidence strength + a confidence %].
HONESTY (mandatory): found text is hypothesis-shaping, not validating; cannot conclude WTP;
absence of signal is a finding (report it as an explicit `absent` read); state the
**selection-bias direction** for THIS segment (which voices found text over- vs
under-represents); a non-target voice running counter to the premise is a finding;
distinguish target voice from firm/homeowner/vendor; REPORT any source you could not read
(login walls, timeouts) — absence ≠ true absence; tag every claim with a source URL; flag
inference vs citation.Run it as a background subagent (these take minutes); drop results into the ledger when it returns.
6. Output contract (the artifact the subagent returns)
- Ranked findings — failure-mode table (demand) or tool-inventory table (supply): item · job · # voices / coverage · representative translated quote(s) · source URL. State the saturation count. For the supply lens, follow the table with a one-paragraph coverage verdict — which job(s) are solved, at what price, for the target segment, and which job is the weak link.
- Per-assumption read — directional answer for each ledger row it feeds, with explicit evidence strength (strong / moderate / weak / absent) and a confidence %. Every assumption the scan found no signal on must appear with strength absent and a one-line note on whether that's true absence or a measurement gap. Flag any off-target voice whose direction contradicts the hypothesis.
- Selection-bias caveat — found text over-represents complainers/vendors and under-represents the silent-satisfied; say plainly what it can and cannot conclude.
- Inaccessible-sources report — what was login-walled / timed out, and which reads rest on weaker channels as a result.
- Sources — deduplicated URLs.
The artifact lands in the target product's docs dir (e.g.
docs/0N_<NAME>.md), not in Bennu.
7. Honesty discipline (non-negotiable)
- Hypothesis-shaping, never validating — the ledger still needs a behavioural test.
- Cannot conclude WTP. Stated interest, clicks, and forum enthusiasm are not money.
- Absence of signal is a finding — report it; don't pad to look thorough.
- No synthetic users as a substitute for real found/voiced signal.
- Voice attribution — the target segment's words only; everything else is context.
- Report what you couldn't read — a walled FB group is a measurement gap, not proof of absence.
8. Handoff to /evidence-ledger
Scan results become signal rows with [Observed: <artifact>, <strength>, <date>] — never a
decision. AI-surfaced themes enter as new assumption rows to test, not as findings.
Worked examples (this skill's calibration)
Illustrative pair for a bet like a tutor parent-summary tool — the skill's two lenses in action:
- Supply lens →
docs/market-scan.md(competitor/coverage scan; if rival tools already cover the job at the target price, it falsifies the literal "no solution" premise and feeds the corresponding ledger assumption). - Demand lens →
docs/voice-of-customer.md(failure-mode coding of real user voices; ranks which pains are acute and which expected pain is absent, feeding the demand assumptions). Each shows the output shape and honesty reporting this skill expects.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

