Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Manages the founder's market understanding — running initial market research, updating findings, and answering questions about market size, customer segments, buying behavior, pricing benchmarks, and industry trends. Use when the conversation touches market size (TAM/SAM/SOM), who the buyers are and how they make decisions, what people typically pay, industry tailwinds or headwinds, or when the founder wants to understand the broader landscape their idea sits in.
.claude/skills/davepoon-market-research/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | -8% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 48% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -12% | 0% |
Help the founder understand the market they're entering — whether it's real, who's in it, how buyers behave, and what they expect to pay. This context strengthens interviews, sharpens hypotheses, and gives the founder a confident answer to "what's the market opportunity?"
Read startup/core.md to load project context (name, seed description, type of business — B2B or B2C — and all fields under ## Core).
Check if startup/market-research.md exists.
Load it for context. Infer intent from the conversation — don't ask "what do you want to do?" If the founder is:
last_updated in frontmatterstatus: needs-refresh in frontmatter and offer to re-run the full research or a targeted pass on specific sectionsweb-researcher agent (fast model) with a focused prompt about that market dimension; incorporate findings into the relevant section; save the full output to startup/research/{YYYY-MM-DD}-{topic-slug}-research.md with frontmatter date, topic, and source_skill: market-researchWhen updating the file, follow the format conventions:
version (number), last_updated (ISO date), type (b2b or b2c), and status (draft, complete, or needs-refresh)Market Research — {Project Name}Read before writing, propose before saving, get confirmation.
Load the reference file for the guided first-time workflow:
.claude/skills/market-research/references/initial-market-research.mdThe reference file's instructions take over from this point.
Not every web question needs a subagent:
WebSearch / WebFetch (you, the main agent): single-fact lookups ("what does Pendo charge?", "is ACME still active?"), quick verification of a claim the founder made, one data point asked about in flow. Stays in conversation, no persistence needed.web-researcher: multi-source passes that benefit from an isolated context (scanning pricing across a whole category, surveying buyer-behavior signals across communities, structured landscape passes). Output is structured and gets saved to startup/research/{YYYY-MM-DD}-{topic-slug}-research.md for later reference.Rough rule: one fact in flow → inline. Multi-source or results-should-persist → dispatch.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 2,647 | 5,068 | +91% | 1 | 1 | 0% | 257 | 1,175 | +357% | 0 | 0 | — |
case-02 | fail→fail | 28,267 | 5,177 | -82% | 1 | 1 | 0% | 5,399 | 1,023 | -81% | 0 | 0 | — |
case-03 | fail→fail | 7,016 | 4,571 | -35% | 1 | 1 | 0% | 1,086 | 1,031 | -5% | 0 | 0 | — |
case-04 | fail→fail | 13,949 | 17,838 | +28% | 1 | 1 | 0% | 2,372 | 3,468 | +46% | 0 | 0 | — |
case-05 | fail→fail | 18,926 | 3,530 | -81% | 1 | 1 | 0% | 3,881 | 963 | -75% | 0 | 0 | — |
case-06 | fail→fail | 27,651 | 4,439 | -84% | 1 | 1 | 0% | 6,165 | 932 | -85% | 0 | 0 | — |
case-07 | fail→fail | 10,956 | 2,137 | -80% | 1 | 1 | 0% | 1,783 | 1,043 | -42% | 0 | 0 | — |
case-08 | fail→pass | 7,328 | 1,957 | -73% | 1 | 1 | 0% | 1,144 | 1,051 | -8% | 0 | 0 | — |
case-09 | fail→fail | 5,617 | 4,917 | -12% | 1 | 1 | 0% | 947 | 1,107 | +17% | 0 | 0 | — |
case-10 | fail→fail | 11,171 | 5,531 | -50% | 1 | 1 | 0% | 1,297 | 1,113 | -14% | 0 | 0 | — |
case-11 | pass→pass | 6,664 | 7,925 | +19% | 1 | 1 | 0% | 1,131 | 1,501 | +33% | 0 | 0 | — |
case-12 | fail→pass | 6,161 | 2,572 | -58% | 1 | 1 | 0% | 1,124 | 1,223 | +9% | 0 | 0 | — |
case-13 | fail→pass | 4,158 | 1,781 | -57% | 1 | 1 | 0% | 750 | 1,111 | +48% | 0 | 0 | — |
case-14 | fail→pass | 10,602 | 3,523 | -67% | 1 | 1 | 0% | 1,964 | 1,398 | -29% | 0 | 0 | — |
case-15 | fail→fail | 1,769 | 5,568 | +215% | 1 | 1 | 0% | 265 | 1,045 | +294% | 0 | 0 | — |
case-16 | fail→pass | 19,403 | 3,126 | -84% | 1 | 1 | 0% | 1,480 | 1,308 | -12% | 0 | 0 | — |
case-17 | fail→pass | 5,836 | 2,008 | -66% | 1 | 1 | 0% | 1,096 | 1,027 | -6% | 0 | 0 | — |
case-18 | fail→pass | 9,737 | 3,980 | -59% | 1 | 1 | 0% | 1,562 | 1,453 | -7% | 0 | 0 | — |
case-19 | fail→pass | 11,150 | 1,459 | -87% | 1 | 1 | 0% | 1,995 | 988 | -50% | 0 | 0 | — |
case-20 | fail→pass | 8,987 | 1,643 | -82% | 1 | 1 | 0% | 1,493 | 947 | -37% | 0 | 0 | — |
case-21 | pass→pass | 8,032 | 3,727 | -54% | 1 | 1 | 0% | 1,267 | 1,407 | +11% | 0 | 0 | — |
case-22 | pass→pass | 15,237 | 3,601 | -76% | 1 | 1 | 0% | 2,498 | 1,309 | -48% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 14 counted toward the lift figure. The other 8 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +41 percentage points is the difference between those two pass rates over the 14 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.