Conduct deep market and positioning research for any product or idea. Works in two modes — give it a URL to analyze an existing product's website, or give it a short memo describing your idea. Uses WebSearch and WebFetch for most research (including Reddit via site:reddit.com + old.reddit.com, and G2/Capterra reviews). Browser tools (Chrome or playwright-cli) used only as fallback. Asks clarifying questions before diving in. Produces an Obsidian wiki — interconnected markdown pages with wikilinks, YAML frontmatter, and a Map of Content index. Use this whenever the user mentions market research, positioning, competitive analysis, or wants to understand how a product fits in its market.
Conduct deep market and positioning research for any product or idea. Works in two modes — give it a URL to analyze an existing product's website, or give it a short memo describing your idea. Uses WebSearch and WebFetch for most research (including Reddit via site:reddit.com + old.reddit.com, and G2/Capterra reviews). Browser tools (Chrome or playwright-cli) used only as fallback. Asks clarifying questions before diving in. Produces an Obsidian wiki — interconnected markdown pages with wikilinks, YAML frontmatter, and a Map of Content index. Use this whenever the user mentions market research, positioning, competitive analysis, or wants to understand how a product fits in its market.
argument-hint
[URL or short product memo, e.g. 'https://linear.app' or 'A tool that helps sales teams automate follow-up emails based on CRM activity']
Positioning & Market Research
You are conducting positioning and market research. The user provided: $ARGUMENTS
Research Philosophy
This skill exists to help someone make a high-stakes decision — whether to build, invest, pivot, or kill an idea. The worst outcome is false confidence. Every phase of research is oriented around invalidation: finding reasons the thesis is wrong, then seeing what survives.
Tier 1 (strongest): Churned user language ("I left X because..."), negative reviews, user complaints, support threads, "I switched from" posts
Tier 2: Job postings in the category (real demand signal), hiring velocity of competitors, actual usage data (GitHub stars trends, npm downloads, BuiltWith adoption)
Tier 3: Funding data, revenue estimates, analyst reports from credible firms (Gartner, Forrester — but note these firms sell reports, so size estimates skew high)
When Tier 4 sources contradict Tier 1-2 sources, Tier 1-2 wins. Always. A competitor's homepage says "loved by 10,000 teams" but G2 reviews say "we churned after 3 months" — lead with the G2 data.
Step 0 — Create the Research Wiki
Before doing anything else, create the wiki folder structure inside research/.
All playwright-cli sessions must use session names prefixed with ${RUN_ID}. All files go into $RESEARCH_DIR/. Nothing in the project root.
Wiki Output Structure
Every research run produces this wiki:
$RESEARCH_DIR/
├── index.md # Map of Content — the entry point
├── verdict.md # The Case Against + go/no-go decision
├── market-reality.md # User pain points, switching signals, quotes
├── competitive-landscape.md # Overview table, pricing landscape, positioning map
├── competitors/
│ ├── {competitor-slug}.md # One page per major competitor (top 3-5)
│ └── ...
├── market-sizing.md # Bottom-up + top-down sizing, demand signals
├── trends-and-shifts.md # Structural shifts, timing, funding signals
├── positioning-strategy.md # Positioning + pricing recommendation
├── next-steps.md # Validation experiments, action items
└── sources.md # Centralized numbered reference list
Page Format
Every wiki page uses this format:
---
date: YYYY-MM-DD
type: research
product: {product-name}
tags:
- positioning
- {relevant-tags}
---# Page Title
Content here. Link to other pages with [[page-name]] and to specific
sections with [[page-name#section-heading]]. Link to competitors
with [[competitors/{slug}]].
Cite sources with numbered references like [1] that link to [[sources]].
Source citations: Use numbered [1] inline, all resolved in [[sources]]
Backlink awareness: When page A discusses something covered in depth on page B, link to it. This makes Obsidian's backlinks panel and graph view useful.
The credibility annotation after the URL helps readers weight evidence at a glance. Every factual claim in any wiki page needs a reference number that resolves here.
Step 1 — Determine Input Mode
Look at $ARGUMENTS:
URL mode — input contains a URL. Scrape the website to understand the product.
Memo mode — input is text. Work from the description.
Carry any focus directives ("focus on the developer tooling angle") through the entire research.
Step 2 — Understand the Product
URL Mode: Scrape the Website
Use WebFetch to read the main page. If it fails (empty, 403, JS-only), fall back to browser.
Discover subpages from actual navigation links — don't guess paths like /pricing. Follow the 2-5 most useful ones.
Memo Mode: Work from the Description
Extract: what it does, who it's for, how it works, what makes it different.
Both Modes: Internal Summary
What the product does (1 paragraph)
Target audience
Core value propositions (3-5 bullets)
Claimed differentiators — flag these as claimed, not validated
Pricing model (if known)
Category terms to research
Step 3 — Ask Clarifying Questions
Pause and ask 2-5 questions. Good ones:
Target audience: "Who's the primary buyer — practitioners, team leads, or company-wide?"
Competitive awareness: "Competitors you already know about?"
Differentiation: "What's hardest for a competitor to replicate?"
Stage: "Existing product you're repositioning, or new idea you're validating?"
Category: "What would someone Google to find this?"
Priorities: "What matters most — competitive intel, user language, positioning, or something else?"
Wait for answers before proceeding.
Prerequisites
Chrome browser with claude-in-chrome extension (preferred)
playwright-cli (fallback) — must be in PATH
Browser Tool Selection
Try mcp__claude-in-chrome__tabs_context_mcp. If it works, use Chrome.
Otherwise, fall back to playwright-cli.
Research Tool Selection
Use the lightest tool that works:
Task
Tool
Why
Search queries
WebSearch
Fast, no browser
Read pages, blogs, docs
WebFetch
Direct content grab
Reddit discovery
WebSearch (site:reddit.com)
Google index finds threads
Reddit reading
WebFetch (old.reddit.com/r/.../comments/ID.json)
.json endpoint returns structured data; fall back to old.reddit.com HTML
Reddit fallback
Chrome/playwright-cli
Escalation only
G2 reviews
WebSearch + WebFetch
G2 acquired Capterra (Jan 2026) — review ecosystem is now consolidated. Search both but expect overlap
Google Trends
Chrome first, then proxy sources
Browser often fails; use explodingtopics.com or trend articles
Discord servers
WebSearch (site:discord.com or "discord.gg")
Find public servers; can't read private channels
Job postings
WebSearch (site:linkedin.com/jobs or site:greenhouse.io)
Demand signal — companies hiring = real need
Glassdoor/Blind
WebSearch + WebFetch
Employee sentiment = competitor health signal
JS-only/403 sites
Chrome/playwright-cli
Last resort
Web Fetching Protocol — Haiku Summarization Layer
Every WebFetch in this skill goes through a summarization layer to control token cost. No agent should WebFetch a page directly into its own context.
Instead, for each URL that needs reading, spawn a Haiku subagent (model: "haiku") with this pattern:
Fetch this URL and extract ONLY the following:
- URL: [the URL]
- Extract: [what you need — e.g., "user complaints about [product]", "pricing tiers and price points", "main arguments for/against [category]"]
Rules:
- Use WebFetch to read the page
- Extract only what was asked for — ignore navigation, ads, sidebars, unrelated content
- If the page is a Reddit thread, extract: top-level post summary, top 5-10 comments by relevance, any specific quotes about [topic]
- If the page is a review site, extract: rating, pros, cons, reviewer context, key quotes
- If the page is a pricing page, extract: tier names, prices, what's included, billing model
- Keep your response under 500 words
- Include the source URL at the top
- If the page fails to load or is paywalled, say so in one line
This means a Reddit thread that's 80K tokens raw becomes a 500-token extract. The Sonnet agent never sees the full page — only the Haiku-compressed summary.
Batch fetches when possible. If an agent needs to read 8 URLs, spawn 8 Haiku subagents in parallel, collect all summaries, then proceed with analysis.
Cap: each Phase 2 agent should fetch at most 15 URLs total. Prioritize ruthlessly — not every search result needs fetching. Read the WebSearch snippets first and only fetch pages where the snippet suggests high-value content.
Phase 2 — Parallel Research
Launch 4 parallel agents using model: "sonnet". All agents use the Haiku summarization layer for WebFetch (see above). They use WebSearch directly.
Each agent writes directly to wiki pages — these are first-class output, not intermediate files. Agents must follow the page format (YAML frontmatter, wikilinks, numbered citations). Each agent also appends its sources to a temporary sources-partial-{N}.md file; these get merged into sources.md during synthesis.
This agent builds the competitive picture from external evidence, not competitor marketing copy. The goal is to understand what competitors actually deliver, not what they claim.
Step 1 — Discover competitors via WebSearch:
"[category]" alternative
best [category] tools [current year]
"[category]" vs (comparison articles from third parties)
"[category]" review site:g2.com
Step 2 — For each competitor, gather third-party evidence:
G2/Capterra reviews (especially 2-3 star reviews — these are most informative)
HN threads mentioning the competitor
Reddit threads discussing the competitor
Blog posts comparing competitors (from users, not the competitors themselves)
Step 3 — Only then, check pricing pages:
Read competitor pricing pages via WebFetch — but treat pricing as factual data, not their marketing narrative. Capture tiers, price points, model type.
Step 4 — Pricing landscape analysis:
Map price ranges across the category
Identify pricing model patterns (per-seat, usage, flat, freemium)
Where are the gaps?
What does pricing signal about willingness-to-pay?
Step 5 — Job posting scan (demand signal):
site:linkedin.com/jobs "[category]" or site:greenhouse.io "[category term]"
Count open roles that would use tools in this category
Note which companies are hiring — this is a real demand signal, harder to fake than market reports
Output — write two things:
competitive-landscape.md — Overview page with: competitor table (name, what users say, pricing, G2 rating, URL), pricing landscape analysis, positioning map showing clusters and white space, job posting demand signals. Link to each competitor's detail page with [[competitors/{slug}]]. Max 1,500 words.
competitors/{slug}.md for each of the top 3-5 competitors — one page per competitor containing:
---
date: YYYY-MM-DD
type: research
product: {product-name}
tags:
- competitive
- competitor
---# {Competitor Name}**URL:** ...
**Pricing:** ...
**G2 Rating:** ...
**Funding:** ...
## What Users Actually Say
Key quotes from G2, Reddit, HN — with source numbers linking to [[sources]].
## Strengths (from user evidence)## Weaknesses (from user evidence)## Employee Sentiment
Glassdoor/Blind signals if available.
## Threat Assessment
How dangerous is this competitor and on what axis? Link to relevant
[[market-reality#switching-signals]] and [[trends-and-shifts]] as needed.
What NOT to include: competitor taglines, hero copy, or self-described positioning. If you reference what a competitor claims, explicitly frame it as "Competitor X claims..." and contrast it with user evidence.
Agent 2: User Pain Points → market-reality.md
This agent gathers authentic user language. It is the most important agent — user voice is the foundation of everything else.
Layer 1 — Reddit via WebSearch + WebFetch (primary):
WebSearch: site:reddit.com "[category]" pain OR frustration OR switched OR alternative
WebSearch: site:reddit.com "[competitor name]" review OR experience OR "switched from"
Read threads via WebFetch using old.reddit.com. For structured data, try the .json endpoint:
G2 acquired Capterra in January 2026 — the review ecosystem is consolidated. Focus on 2-3 star reviews (most informative) and "switched from" mentions.
WebSearch: site:g2.com "[competitor name]" reviews
WebSearch: "[competitor name]" "switched from" OR "moved to" OR "pros and cons"
Layer 3 — Discord (important for dev tools and community-led products):
Many authentic discussions now happen in Discord, not Reddit. Public servers are discoverable:
WebSearch: "[category]" discord server OR discord.gg
WebSearch: site:discord.com "[category]"
Note which Discord communities exist and their size. If threads are publicly indexed, read them.
Layer 4 — HN and community forums:
WebSearch: site:news.ycombinator.com "[category]" OR "[competitor]"
WebSearch: "[category]" forum discussion pain OR frustration
Layer 5 — Blind/Glassdoor (employee sentiment as competitive intel):
Competitor employee sentiment reveals things marketing never will — low morale, talent flight, leadership chaos, product direction disagreements.
---
date: YYYY-MM-DD
type: research
product: {product-name}
tags:
- user-voice
- positioning
---# Market Reality## What Users Actually Say
Pain points ranked by frequency and intensity, as a table.
User language vs. product language — do they match?
Key quotes with source numbers → [[sources]].
## Switching Signals
Why users leave current tools. Specific triggers.
Link to relevant [[competitors/{slug}]] pages.
## What Users Have Tried
Tools and workarounds mentioned, ranked by frequency.
What works, what doesn't, the gap.
## Employee Sentiment
Glassdoor/Blind signals on key competitors.
Talent flow direction. Links to [[competitors/{slug}#employee-sentiment]].
Max 2,000 words. Prioritize direct quotes over summaries — quotes are the raw material the synthesis needs.
Citation rule: Every quote MUST have a direct URL. No secondhand quotes. If you can't link to the original, mark as [unverified — secondhand source].
Google Trends — try browser first, fall back to proxies:
Using Chrome:
mcp__claude-in-chrome__tabs_create_mcp → get tabId
mcp__claude-in-chrome__navigate(tabId, "https://trends.google.com/trends/explore?q=TERM1,TERM2,TERM3&hl=en")
mcp__claude-in-chrome__read_page(tabId)
mcp__claude-in-chrome__javascript_tool(tabId, "window.close()")
If browser fails, use proxies:
WebSearch: "google trends" "[category term]" growth OR rising OR declining
WebSearch: site:explodingtopics.com "[category term]"
Funding signals:
WebSearch: "[category]" funding OR raised OR series [current year]
WebSearch: "[category]" acquisitions OR shutdown OR pivot [current year]
Funding tells you where money flows. But also search for shutdowns and pivots — a competitor leaving the category is a signal too, and not always a positive one.
Structural shifts — the most valuable part of this agent:
Look for forces reshaping the segment. Not "the market is growing" but "the rules are changing."
WebSearch: "[category]" disrupted OR commoditized OR "paradigm shift" OR "future of" [current year]
WebSearch: "[category]" AI impact OR automation OR "no longer need" [current year]
WebSearch: "[adjacent technology]" replacing "[category]"
WebSearch: "[category]" Gartner OR Forrester OR "magic quadrant" [current year]
If Gartner has a Magic Quadrant → established category. Market Guide → emerging. Nothing → either too niche or too new.
Output — write trends-and-shifts.md:
---
date: YYYY-MM-DD
type: research
product: {product-name}
tags:
- structural-shift
- demand-signal
---# Trends & Structural Shifts## Search Interest
Trend data with sources.
## Funding & Exits
Funding velocity, shutdowns, acquisitions. What the money signals say.
Link to [[competitors/{slug}]] where relevant.
## Structural Shifts
For each shift:
-**The shift:** What's changing and why
-**Evidence:** Concrete signals [N]
-**Winners and losers** — link to [[competitors/{slug}]] as appropriate
-**Timeline:** Now / 1-2 years / 3-5 years
-**Implication:** Does this create opportunity or threaten the premise?
## Analyst Coverage
Gartner/Forrester positioning. Confidence flags on estimates.
## Market Timing
Emerging / growing / crowded / consolidating?
**Decision point:** Is the window open, closing, or not yet open?
Flag confidence level honestly. Grand View Research, Mordor Intelligence, etc. sell reports — their TAM estimates are marketing for the reports themselves. Treat them as upper bounds, not expected values.
Step 3 — Platform risk scan:
WebSearch: "[big player]" "[category feature]" announced OR launched OR acquired
Check if any major platform has shipped or announced overlapping features.
Step 4 — Invalidation search (critical):
Actively search for reasons this market/product might not work:
WebSearch: "[category]" failed OR shutdown OR "didn't work" OR "waste of money"
WebSearch: "[category]" "solution looking for a problem" OR "nobody needs"
WebSearch: site:news.ycombinator.com "[category]" dead OR overhyped OR "who uses"
WebSearch: "[similar product that failed]" post-mortem OR "lessons learned"
Look for:
Startups that tried this and failed — why?
HN threads where practitioners are skeptical
Categories that sound big but have low actual adoption
"Graveyard" signals — lots of funded companies, few survivors
This is the most important step. If you can't find reasons the thesis might be wrong, you haven't looked hard enough.
Output — write two pages:
market-sizing.md:
---
date: YYYY-MM-DD
type: research
product: {product-name}
tags:
- market-sizing
- demand-signal
---# Market Sizing## Bottom-Up (Primary)
Number of potential customers, with math shown and assumptions explicit.
Sources linked to [[sources]].
## Top-Down (Supplement)
Analyst estimates with confidence flags and publisher context.
## Demand Signals
Job postings, funding velocity, shutdowns.
Link to [[trends-and-shifts#funding--exits]] for detail.
verdict.md:
---
date: YYYY-MM-DD
type: research
product: {product-name}
tags:
- risk
- positioning
---# The Case Against**Verdict:** [One sentence: go / no-go / conditional go / needs validation]
## Failed Companies
Startups that tried this and failed — why? With citations.
## Strongest Skeptic Arguments
From practitioner communities. Numbered, with evidence.
## Platform Risk
Which big players could enter and ship this as a feature?
## What This Research Couldn't Answer
Data gaps, unknowns, unverifiable claims.
## Verdict on Survivability
After considering all the above, does the opportunity still hold?
What specifically survives scrutiny?
Links to supporting evidence across [[market-reality]],
[[competitive-landscape]], [[trends-and-shifts]].
Max 2,000 words across both pages.
Phase 3 — Deep Dives (Opt-In)
Deep dives are not automatic. After Phase 2 completes, review the wiki pages and ask the user:
"Phase 2 is done. Here's what I found: [2-3 sentence summary of key findings and gaps]. I'd recommend a deep dive into [specific question]. Want me to go deeper, or should I proceed to synthesis?"
If the user says proceed, skip to Phase 4. If they want deep dives, launch 1-2 agents (not 3) using model: "sonnet". Each deep dive uses the Haiku summarization layer for fetches. Choose based on what Phase 2 surfaced:
Deep-dive into the #1 threat — actual user sentiment, employee morale, growth trajectory
Deep-dive into the invalidation case — if Phase 2 found red flags, investigate further
Deep-dive into user pain — more Reddit/G2/Discord threads, specific workflows
Deep-dive into adjacent market — tools in neighboring categories
Deep Dive Quality Criteria
Each deep dive must:
Answer a specific question (not just "research competitor X" — but "can competitor X's enterprise moat be broken?")
Use Tier 1-2 evidence primarily
Include at least 10 numbered citations with URLs
End with a clear "so what" — what this means for the product
Be honest about what it couldn't find or verify
No word minimums. Answer the question thoroughly, then stop.
Save each deep dive as its own wiki page in $RESEARCH_DIR/ with a descriptive filename and appropriate frontmatter/tags. Link to and from other wiki pages as appropriate.
Phase 4 — Synthesize the Wiki
Phase 4 is where the wiki becomes a coherent whole. The agents wrote the research pages; now you connect them and write the synthesis pages.
Step 1 — Merge Sources
Combine all sources-partial-*.md files into a single sources.md. Deduplicate, renumber sequentially, and update all references across all wiki pages to match the final numbering.
Step 2 — Write positioning-strategy.md
---
date: YYYY-MM-DD
type: research
product: {product-name}
tags:
- positioning
- pricing
---# Positioning Strategy## Core Tension
The fundamental tension this product must navigate.
## Recommended Positioning
Grounded in user language from [[market-reality]], not aspirational marketing.
## One-Liner## Differentiation
Which differentiators are real (evidence from [[market-reality]], [[competitive-landscape]])
vs. claimed. Ranked by market need, defensibility, evidence strength.
## Pricing Recommendation- Recommended price point(s) with rationale tied to gaps in [[competitive-landscape#pricing-landscape]]
- Pricing model with justification
- Tier structure if applicable
- Evidence basis: which data points support this price
-**What's NOT validated:** willingness-to-pay is inferred, not tested. Specific validation steps.
Step 3 — Write next-steps.md
---
date: YYYY-MM-DD
type: research
product: {product-name}
tags:
- positioning
---# Next Steps## Priority Actions
Ranked list. Each action links to the evidence that motivates it.
## What to Validate
Specific experiments: landing page tests, interview targets, pre-sale approaches.
## What to Kill or Deprioritize
Based on [[verdict]] findings.
Step 4 — Write index.md (Map of Content)
This is the entry point. Write it last so it reflects everything that was actually produced.
---
date: YYYY-MM-DD
type: research
product: {product-name}
tags:
- positioning
- MOC
---# {Product Name} — Positioning Research**Date:** {today}
**Input:** {URL / Memo summary}
**Verdict:** {One sentence — links to [[verdict]] for full analysis}
---
## Research Map- [[verdict]] — The case against, risks, platform threats, go/no-go
- [[market-reality]] — User pain points, switching signals, authentic quotes
- [[competitive-landscape]] — Competitor overview, pricing, positioning gaps
- [[competitors/{slug-1}]] | [[competitors/{slug-2}]] | [[competitors/{slug-3}]] | ...
- [[market-sizing]] — Bottom-up & top-down sizing, demand signals
- [[trends-and-shifts]] — Structural shifts, timing, funding signals
- [[positioning-strategy]] — Recommended positioning, pricing, differentiation
- [[next-steps]] — Priority actions and validation experiments
- [[sources]] — All references with credibility annotations
{If deep dives were produced, list them here with links}
## Key Findings
3-5 bullet summary of the most important findings, each linking to
the relevant wiki page for detail.
Cross-Linking Pass
After writing all synthesis pages, do a quick pass across every page:
Ensure competitor names link to their [[competitors/{slug}]] page
Ensure claims link to their evidence page
Ensure [[sources]] references are consistent
Writing Rules
Audience: Decision makers who need the truth, not reassurance.
If the product is poorly positioned, say so. If a differentiator is weak, call it out.
Surface hard truths in [[verdict]], not buried in [[next-steps]].
Every page answers "so what?" — not just describes the landscape.
No filler, no hedging, no "it could be argued that."
Challenge assumptions. In URL mode, compare website claims against user evidence.
Rank everything. Don't present options without a recommendation. Take a position even when evidence is mixed — and state the risk of being wrong.
Distinguish claimed vs. evidenced. When a finding comes from competitor marketing, say so. When it comes from user reviews, say so. Let the reader see the evidence quality.
