researcher
Skill2 starsChanged 50 days agoMITView on GitHub
Desk research on the web, filed as cited Evidence (never Signals) to fill gaps in a persona's data. Use on /researcher or when asked to find external reports/studies/data on a topic.
42 other projects use a version of this.
Adopted by
Most recently taken up by
- sphnx-io/Claude-Skillstheir own version, first seen Nov 16, 2026
- jasoncg8/claude-skillstheir own version, first seen Nov 16, 2026
- Mazalucas/El-DT-Lightweight-Armytheir own version, first seen Nov 13, 2026
- AhmedAbdelfattah0/AI-Skillstheir own version, first seen Nov 12, 2026
- Jakeschincariol/closer-skilltheir own version, first seen Nov 8, 2026
- natgalla/context-historiantheir own version, first seen Nov 6, 2026
Used with
Projects with it also use build-premium-website, albert-dm and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
yawningwhaleAISkillbuild-premium-website
Build a premium, animated marketing website (React + Vite + Tailwind + GSAP) for any industry. Use when the user asks to "build a website", "make a landing page for [business]", "build a marketing site", or wants a high-end animated company website. Adapts color, copy, services, and signature animation to the industry.
Also in 9 projects that use this
yawningwhaleAISkillalbert-dm
Reply to DMs and messages the way Albert would. Use when the user asks "how would Albert reply to this", pastes a conversation and wants an Albert-style response, or asks you to draft/write a DM in Albert's voice. Albert sells AI voice agents / automations to local service businesses (Go High Level, lead follow-up, bookings).
Also in 7 projects that use this
huyvpSkillcost-reducer
Reduce cloud, infrastructure, and operational costs while maintaining performance. Use when writing database queries, configuring cloud resources, optimizing bundles, setting up caching, choosing between services, sizing instances, configuring CDN, managing storage, or reviewing code for cost inefficiencies. Covers AWS/GCP/Vercel pricing, database optimization, serverless tuning, image pipelines, observability costs, and FinOps practices.
Also in 13 projects that use this
RoxenRobloxSkillyoutube-broll-maker
Generate 10-second B-roll footage of yourself doing an action using the Higgsfield MCP (Seedance 2.0). Upload one photo of yourself, describe the action ("typing on laptop", "walking at night"), pick from 5 cinematography style presets, get an MP4. Use when the user wants B-roll, cutaway footage, "me doing X" video, YouTube B-roll, or any image-to-video of themselves performing an action.
Also in 4 projects that use this
RoxenRobloxSkillyoutube-popup-graphic
Generate stunning Adobe-Premiere-Pro-style UI popup motion graphics for YouTube videos using the Higgsfield MCP. The user uploads a style reference image (any UI mockup, dashboard, glass card, design they love), names one skill/feature/item, and the skill produces an MP4 of a single card popping in styled to match the reference. Use when the user wants a "popup graphic", "popup animation", "card popup", "skill popup", "AE/Premiere-style animation", "stunning UI animation for YouTube", or says "make me a popup for <skill/feature/item> styled like <reference>".
Also in 4 projects that use this
xfstudioSkilltrigger-dev
Trigger.dev expert for background jobs, AI workflows, and reliable async execution with excellent developer experience and TypeScript-first design. Use when: trigger.dev, trigger dev, background task, ai background job, long running task.
Also in 12 projects that use this
- +4 more from yawningwhaleAI
Other versions42 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 42
- keli-wen/agy-staff★ 873same name
- bitwize-music-studio/claude-ai-music-skills★ 668same name
- understudy-ai/understudy★ 461same name
- krzysztofdudek/ResearcherSkill★ 269same name
- antonbabenko/deliberation★ 183same name
- svy04/ballast★ 73same name
- timscheuerai/content-vault★ 38same name
- Jakeschincariol/closer-skill★ 33same name
- AlekseiUL/heisenberg-team★ 17same name
- maher-naija-pro/claude-researcher★ 13same name
32 more not listed
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 repository31 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- affinitySkill
- ai-personaSkill
- analytics-syncSkill
- backlogSkill
- catalogSkill
- cold-startSkill
- contradictionsSkill
- demo-dataSkill
- dovetail-syncSkill
- exportSkill
- extract-findingsSkill
- feature-panelSkill
- forget-participantSkill
- graph-lintSkill
- interview-guideSkill
- journey-mapSkill
- opportunity-treeSkill
- persona-avatarSkill
- persona-querySkill
- persona-talkSkill
- persona-voiceSkill
- persona-workshopSkill
- prdSkill
- prioritizeSkill
- retentionSkill
- source-syncSkill
- subject-accessSkill
- welcomeSkill
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 4 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 38 edited versions we read keep these.
4
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 37
If genuinely ambiguous, ask one clarifying question — don't guess at scope silently.
- Very specific: line 40
Use WebSearch broadly first, then WebFetch on the most promising, most authoritative pages.
- Distinctive: line 52
Never invent a statistic or quote to fill a thin result: If the search comes up empty or weak, say so plainly instead of padding the file.
- Very specific: line 55
Before creating a new file, scan Evidence/ for an existing entry on the same topic.
- Distinctive: line 61
Create the confirmed file(s).
- Very specific: line 62
If the research targeted a persona, offer to add a line under that persona's ## Evidences section, linking the new file and stating what it …
- Distinctive: line 75
Prefer the primary source (the original study/report) over a secondary write-up when both are findable.
“Common” is measured: kept by at least 6 in 10 of the 38 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
researcher — Desk Research Assistant (UX Researcher + Analyst)
On request, search the web for existing desk research — reports, studies, official statistics, community threads, journalism — that fills a gap in a persona's data. Cite properly, then turn findings into Evidence/ files. This is the tool for "go find out if this is actually true" — it doesn't run interviews, it finds what's already published.
Language: work in the user's language; keep any quoted source text verbatim (don't translate a quote from a source — translate your own commentary around it).
Trigger
/researcher <topic or question>
/researcher Personas/Emma.md ← find desk research for gaps in her data
/researcher Personas/Emma.md "student music discovery habits" ← persona + a specific angleAlso invoked mid-conversation from /persona-talk, when a persona hits a Known unknown and the user explicitly agrees to a web-fallback offer — see persona-talk/SKILL.md's "Warm-up and unknowns" section. Same skill, same rules, just triggered from inside a session instead of run standalone.
Real user research always outranks this. Whatever this skill finds is desk research (Evidence), not an interview (Signal) — say so plainly whenever it's offered as a stand-in for a data gap: a real interview/test with an actual user is the stronger, preferred source, and finding something on the web doesn't close that gap, it only softens it temporarily.
Competitor quick-research (app-added files)
A Competitors/*.md with needs_research: true was created by name in the app. When the user confirms (the SessionStart hook prompts you to ask), honor the lookback window they chose — 1, 2 or 3 years back from today — and skip sources published earlier. Fill the file's own template sections (Market position, Features vs the product, User voices, Sources) with linked public data; recent-first. Then set retrieved: <today> and delete the needs_research: line. The competitor file is opt-in context, not Evidence — file separate Evidence/ entries only for findings that ground persona claims.
Why this always produces Evidence, never Signal
This project's grounding model draws a hard line: Signal = an observation from OUR OWN interview/test (has an "Interview date" and a Transcript link back to a session we ran). Evidence = desk research — data someone else already published, that we found and cited. A community thread, a market report, an academic paper — no matter how compelling the quote inside it — is desk research, not our interview. Filing it as a Signal would silently inflate the Level scale (Signal-only = L3, "from experience") for something that was never actually witnessed firsthand. Everything this skill produces is Evidence.
Workflow
- Scope the research.
- Given a persona file: read its
Pains,Jobs to be Done, andCorrelations. Identify claims that are thin — L2 (Evidence-only, no interview) or L3 (Signal-only, no external confirmation) — or gaps with no data at all. - Given free text: treat it as the research question directly.
- If genuinely ambiguous, ask one clarifying question — don't guess at scope silently.
- Given a persona file: read its
- Search.
- Use
WebSearchbroadly first, thenWebFetchon the most promising, most authoritative pages. - Priority order: official statistics/company reports > academic papers > established journalism > community/forum threads (fully valid — this repo already cites Spotify Community threads throughout) > generic blogs (only if nothing better exists, and say so).
- Cross-check a surprising number against a second source when one exists.
- Use
- Draft candidate Evidence entries.
- One
Evidence/<Title>.mdper distinct finding, usingEvidence/_template.mdexactly,retrieved:= today. Fill what the AI needs first:claim:— the finding in one plain sentence;source_kind:,published:(the source's own date),population:(who, where, how many) — empty when the source does not say, never guessed;primary_checked: trueonly when you read the figure in the original report or dataset (rule 7); press coverage alone isfalse, and say so;- Key figures: one row per number with its population and source — no figure without who it is about;
- Does not settle: at least one honest line on what the source cannot tell you (rule 8).
- Content and Takeaways stay readable prose for people. Never add
==highlights==— they are the team's emphasis. - Never invent a statistic or quote to fill a thin result. If the search comes up empty or weak, say so plainly instead of padding the file.
- One
- Check for duplicates.
- Before creating a new file, scan
Evidence/for an existing entry on the same topic. If one exists, prefer updating itsretrieved:date and content over creating a near-duplicate.
- Before creating a new file, scan
- Present a plan, then confirm before writing.
- List candidates: title — one-line summary — source domain(s). Ask which to keep. Don't silently write a batch of files the user hasn't seen (same pattern as
dovetail-sync).
- List candidates: title — one-line summary — source domain(s). Ask which to keep. Don't silently write a batch of files the user hasn't seen (same pattern as
- Write, then wire.
- Create the confirmed file(s).
- If the research targeted a persona, offer to add a line under that persona's
## Evidencessection, linking the new file and stating what it confirms. - If a finding contradicts an existing Pain/Correlation, flag it explicitly rather than quietly overwriting — this repo's standing rule is that contradictions are data, not noise.
- Freshness is automatic at creation (new Evidence starts at L2+ eligible with today's
retrieved:date) — but this skill is also the right tool to periodically refresh Evidence the 3-month rule has flagged as stale.
Target-population estimates (for the app's sample-confidence math)
When asked to find the size of the user's target population (total active accounts, subscribers in a segment, market size in people): use public sources no older than 9 months — investor reports, regulator filings, established market research; cite the specific page and date. File it as Evidence/ like everything else, then tell the user the number and offer to enter it in the app's Settings → Target population (stored in the gitignored .claude/preferences.local.md). If only older data exists, say so plainly and mark it as needing refresh — never silently use a stale figure for confidence math.
Rules
- Real, verifiable, linkable sources only — the same hard rule as the rest of this project. No fabricated statistics, no invented quotes, ever.
- Cite the specific page, not a bare domain.
- Prefer the primary source (the original study/report) over a secondary write-up when both are findable.
- A weak source is still worth including if it's the only one — just label it plainly ("only source found — treat as thin") rather than omitting the gap silently.
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.

