Market Intelligence Report

This skill conducts YouTube + Twitter/X competitive intelligence — analyzing channels, discovering content gaps, generating video ideas, and scanning for daily opportunities — and then packages the findings into a BenAI-branded instant-ui HTML dashboard the user can open, screenshot, or share.

Every run has two halves:

1. Gather + synthesize the intelligence (Standard Workflows below).
2. Render the dashboard from the bundled template in the instant-ui design language (see "Dashboard Output").

Connectors

youtube connector (Primary for all YouTube data)

The @kirbah/mcp-youtube connector is the primary data source for YouTube search, video details, channel stats, and transcripts. It uses the YouTube Data API v3 (free tier: 10,000 units/day).

Tool	What it does	API Cost (units)
searchVideos	Search videos/channels by keyword with recency filter (pastHour, pastDay, pastWeek, pastMonth, pastQuarter, pastYear)	100 per call
getVideoDetails	Video metadata, stats, duration, category	~1 per video
getChannelStatistics	Subscriber count, view count, video count	~1 per channel
getChannelTopVideos	Top/recent videos for a channel	~3 per call
getVideoComments	Video comments	~1 per call
getVideoCategories	List video categories	~1 per call
getTrendingVideos	Trending videos by region/category	~1 per call
getTranscripts	Video transcripts (captions)	Free (0 cost)

Quota budget: 10,000 units/day. searchVideos is the most expensive at 100 units — use it intentionally (~50-80 searches/day max). All other tools cost ~1 unit each. Transcripts are always free.

apify connector (Fallback for YouTube search + primary for Twitter/X)

Use Apify as a fallback if the YouTube connector hits quota limits, or as the primary source for Twitter/X data.

YouTube fallback via automation-lab/youtube-scraper

Uses YouTube's InnerTube API (direct HTTP, no browser) — no API key needed, no quota limits.

Capability	How	Cost
Search videos by keyword	searchQueries array	$0.005/run + $0.003/video
Get video details	Pass video URLs via startUrls	$0.005/run + $0.003/video
Get channel data	Pass channel URLs via startUrls	$0.005/run + $0.003/channel

Can combine search queries + video URLs + channel URLs in a single run.

Input examples:

// Search by keyword
{ "searchQueries": ["Claude AI tutorial"], "maxResults": 20 }

// Scrape specific channel
{ "startUrls": [{"url": "https://www.youtube.com/@channelname"}], "maxResults": 50 }

// Combined: search + channel + video in one run
{ "searchQueries": ["Claude AI tutorial"], "startUrls": [{"url": "https://www.youtube.com/@channelname"}], "maxResults": 20 }

Twitter/X scraping via apidojo/twitter-scraper-lite (Primary)

apidojo/twitter-scraper-lite (Twitter Scraper Unlimited) is the primary Twitter actor. Use this for all Twitter/X calls.

Parameter	Type	What it does
searchTerms	string[]	Keyword search. Supports Twitter advanced syntax ("from:handle", "to:handle")
twitterHandles	string[]	Direct handle scraping — preferred for watchlist batch scanning
author	string	Single author filter
start / end	string	Date range filter (ISO 8601). Use for 48-hour lookback windows
minimumFavorites	int	Engagement floor — only return tweets with N+ likes
minimumRetweets	int	Only return tweets with N+ retweets
sort	enum	"Top", "Latest", or "Latest + Top"
maxItems	int	Max tweets to return per call
tweetLanguage	string	ISO 639-1 language code (e.g., "en")

Watchlist scan example:

{ "twitterHandles": ["AnthropicAI", "claudeai", "trq212"], "maxItems": 50, "start": "2026-04-13T06:00:00Z", "sort": "Latest" }

Keyword scan example:

{ "searchTerms": ["Claude Cowork", "Anthropic launch"], "maxItems": 30, "sort": "Top", "minimumFavorites": 100 }

Twitter/X fallback chain (IMPORTANT)

When a Twitter call fails or returns empty (noResults: true), follow this exact order:

1. Switch actor first — retry the same request with apidojo/tweet-scraper using identical input params. Do NOT relax filters on the primary actor. The most likely cause of empty results is the actor being down, not the filters being too strict.
2. Only if both actors fail — then relax filters (remove date range, lower minimumFavorites, broaden searchTerms).
3. Only if both actors fail with relaxed filters — report the failure to the user and continue with other data sources.

Never relax filters or remove date ranges as a first response to empty results. Switch actor first.

vidiq connector (YouTube SEO, keyword research, outlier/breakout discovery)

The vidiq connector exposes vidIQ's proprietary YouTube intelligence layer — data the public YouTube Data API does not provide. Use it whenever you need search demand, competition scores, viral outliers, breakout channels, or title/thumbnail scoring.

Tool	What it does	When to use
vidiq_keyword_research	Search volume, competition score, related keywords for a YouTube query	Idea validation, gap analysis, title brainstorming
vidiq_outliers	Videos significantly outperforming a channel's baseline (viral hits)	Identify what's working in the niche right now
vidiq_breakout_channels	Channels growing unusually fast in a niche	Discover emerging competitors before they're obvious
vidiq_trending_videos	Trending videos by category/region	Trend scouting and daily scans
vidiq_trend_categories	Trending category breakdown	Understand which topic clusters are hot
vidiq_channel_analytics	Deep channel performance data (engagement, growth, view velocity)	Competitive research on a specific channel
vidiq_channel_performance_trends	Channel growth/performance trends over time	Long-arc competitor tracking
vidiq_channel_stats	Channel statistics snapshot	Quick channel overview
vidiq_channel_videos	Channel's video list with vidIQ metrics	Pull recent videos with engagement scoring
vidiq_similar_channels	Channels similar to a given channel	Competitor discovery beyond the known set
vidiq_score_title	Score a draft video title for SEO/CTR	Ideation refinement, before pushing to outline
vidiq_score_thumbnail	Score a thumbnail for CTR	Pre-publish thumbnail validation
vidiq_video_stats	vidIQ's video stats (views, engagement, score)	Cross-check against YouTube getVideoDetails
vidiq_video_comments	Video comments with vidIQ enrichment	Audience signal mining
vidiq_video_transcript	Video transcript via vidIQ	Backup transcript source if YouTube getTranscripts fails
vidiq_get_videos_by_ids / vidiq_get_channels_by_ids	Bulk lookup by ID	Batch enrichment
vidiq_balance	Remaining vidIQ API balance	Quota check

When vidIQ wins over YouTube/Apify:

• Keyword research: vidiq_keyword_research returns search volume + competition score, which YouTube Data API does not expose. Use this for every idea generation pass.
• Outliers: vidiq_outliers identifies viral breakouts in a channel's recent uploads automatically — no need to compute "3x channel average" manually.
• Breakout channels: vidiq_breakout_channels surfaces channels gaining momentum that wouldn't show up in keyword search yet.
• Title/thumbnail scoring: vidiq_score_title and vidiq_score_thumbnail give SEO/CTR scores before publishing.

Reference Documents

File	Contains
references/youtube-strategy.md	Strategic positioning, content strategy, competitive landscape
references/content-tier-guide.md	3 video formats: Big Feature Launch, Bundled Small Updates, How-To for Business
references/niche-analysis-framework.md	Channel analysis methodology
references/idea-generation-framework.md	Ideation methods: gap analysis, trend riding, format innovation, audience needs
references/validation-methodology.md	Idea scoring framework
references/export-templates.md	Output schemas for niche-analysis.json and niche-report.md
references/icp-ideal-customer-profile.md	Target audience: non-developer professionals using AI tools
references/youtube-scraping-guide.md	How the scraper collects data
references/twitter-watchlist.md	330+ scored/tiered Twitter handles for ecosystem monitoring
references/anthropic-official-sources.md	Official Anthropic/Claude update sources: GitHub releases, platform changelog, blog, SDKs
references/dashboard-template.html	Instant-ui dashboard scaffold rendered at the end of every run
references/instant-ui/design-tokens.md	BenAI instant-ui :root token block and typography rules
references/instant-ui/page-shell.md	Instant-ui page wrapper, BenAI smiley SVG, responsive breakpoints
references/instant-ui/components.md	Instant-ui component library (pills, cards, stat strips, tables, CTA)
references/instant-ui/build-rules.md	Instant-ui voice rules, hard build constraints, after-build checklist

Mandatory: Use ALL Connectors

Every time this skill is invoked, you MUST use every available connector. Do not skip any connector. Do not limit the number of queries or calls. Be thorough — cast a wide net.

• YouTube connector — Primary for ALL YouTube data: searchVideos (with recency filter for date-scoped searches), getVideoDetails, getChannelStatistics, getChannelTopVideos, and getTranscripts (free). Be mindful of the 10K units/day quota — searchVideos costs 100 units per call; all other tools cost ~1 unit.
• vidIQ connector — Mandatory for keyword research, outlier detection, breakout channel discovery, and title/thumbnail scoring. YouTube Data API does not expose search volume or competition scores — vidIQ does. Use vidiq_keyword_research on every ideation pass, vidiq_outliers and vidiq_breakout_channels on competitive research, vidiq_trending_videos on trend scouts, and vidiq_score_title to validate draft titles.
• Apify connector — Primary for Twitter/X via apidojo/twitter-scraper-lite. If it fails or returns empty, fall back to apidojo/tweet-scraper (same input params). Fallback for YouTube search via automation-lab/youtube-scraper if YouTube connector hits quota limits.
• Official Anthropic sources — Claude Code GitHub releases, Platform Changelog, and Anthropic News Blog must be checked in every daily scan and trend scout (see references/anthropic-official-sources.md). Use WebFetch to pull the latest entries.

If a connector fails or is unavailable, try the fallback first. Only tell the user if both primary and fallback fail. Never silently skip a connector.

IMPORTANT: Prefer spawning sub-agents to run MCP calls in parallel. Delegating MCP calls (YouTube, vidIQ, Apify, WebFetch) to sub-agents is encouraged — the sub-agents can access the MCP tools too. Fan the data collection out across sub-agents so many connectors are hit at once, then synthesize their results. You may still make MCP calls directly when a single quick call is all that's needed.

Transcript Expansion (Agent-Driven)

The agent autonomously decides when to pull transcripts. Do NOT wait for the user to ask. After reviewing search results, expand into transcripts when:

• A video is an outlier (3x+ above channel average views)
• A video covers a direct competitor topic or uses a similar angle to what we'd produce
• A video's title/description suggests a novel approach worth understanding
• The task is idea generation or competitive research and the video's content matters more than its stats

Use getTranscripts with key_segments format first (intro hook + outro CTA). Only pull full_text if the key segments reveal the video is highly relevant and you need the full argument/structure.

Standard Workflows

Daily Scan ("check today's vids", "morning scan", "what's new", "scan for updates")

Compulsory checklist: every item below MUST run. Never skip, defer, or partially execute any of these. If a connector fails, report the failure explicitly and continue with the rest.

Prefer spawning sub-agents to run MCP calls in parallel. Delegating MCP tool calls to sub-agents is encouraged — the sub-agents can access the MCP tools too. Fan the collection work out across sub-agents to hit as many connectors at once as possible, then synthesize their results. Direct MCP calls are still fine for quick one-offs.

File saving: Only save intelligence data files if the user's prompt contains an explicit line like Vault: /some/path. If the prompt does not contain that line, do NOT save data files anywhere. Do NOT search the filesystem for a vault. Do NOT look for .obsidian directories. Just run the scan and present results in chat. (The dashboard HTML is always written — see "Dashboard Output".)

Phase 1: Collect data (fire all MCP calls in parallel)

Launch as many of these as possible in a single turn. Do not wait for one to finish before starting the next. Do not analyze or summarize results as they return. Collect everything first, then synthesize once.

Official Anthropic sources — Use WebFetch to check these three sources for anything shipped in the last 48 hours (see references/anthropic-official-sources.md):

• Claude Code GitHub releases — Fetch https://github.com/anthropics/claude-code/releases.atom for new CLI releases
• Platform Changelog — Fetch https://platform.claude.com/docs/en/release-notes/overview for API changes, model launches, feature additions
• Anthropic News Blog — Fetch https://raw.githubusercontent.com/taobojlen/anthropic-rss-feed/main/anthropic_news_rss.xml for major announcements

YouTube searches — Use searchVideos with recency: "pastDay" covering ALL key topic groups. Use a 48-hour lookback to avoid missing videos near the boundary. Every topic group is compulsory:

• Core: "Claude Cowork", "Claude Desktop", "Claude AI"
• Tools: "Anthropic AI", "Claude MCP", "Claude skills"
• Niche: "AI tools for business", "AI automation tutorial"
• Competitors: "AI tutorial non-technical", "no code AI"
• Trending: any current hot topics in the AI tools space

vidIQ scans — Run all of these in parallel with the YouTube searches:

• vidiq_trending_videos — Pull current trending videos in the AI/tech category to surface what's breaking out beyond keyword search.
• vidiq_outliers for each known competitor channel (AI Jason, Corbin Brown, Jack Roberts) — surfaces any of their recent videos that have outperformed their baseline (instant viral signal).
• vidiq_breakout_channels in the AI tools/automation niche — catches new competitors gaining momentum that aren't on the watchlist yet.
• vidiq_keyword_research on the day's emerging Twitter keywords (e.g., new feature names from Anthropic releases) — quantifies search demand before deciding whether a topic is worth a video.

Twitter/X watchlist — Use apidojo/twitter-scraper-lite (fall back to apidojo/tweet-scraper on failure). See references/twitter-watchlist.md for handle arrays. Every tier is compulsory. Do not skip any tier or handles within a tier.

• P1 handles — All ~30 P1 handles via twitterHandles, maxItems: 50, 48-hour start window, sort: "Latest". No engagement filter.
• P2 handles — All ~69 P2 handles via twitterHandles, maxItems: 50, 48-hour start window, minimumFavorites: 10.
• P3 handles — All ~120 P3 handles via twitterHandles (split into 2-3 batches), maxItems: 30 per batch, 48-hour start window, minimumFavorites: 50.
• P4 handles — All ~93 P4 handles via twitterHandles, maxItems: 30, 48-hour start window, minimumFavorites: 100.
• Keyword scan — searchTerms: ["Claude Cowork", "Claude Desktop", "Anthropic", "Claude MCP", "Claude skills", "Claude agents"], maxItems: 30, 48-hour start window, sort: "Top".

Twitter/X discovery (beyond the watchlist) — catches signals the watchlist might miss. All with 48-hour start window:

• Broad ecosystem — searchTerms: ["Claude AI", "Anthropic", "Claude Code", "Claude Cowork"], maxItems: 50, sort: "Top", minimumFavorites: 50.
• Adjacent AI tools — searchTerms: ["AI automation for business", "AI tools non-technical", "AI workflow no code", "AI assistant for work"], maxItems: 30, sort: "Top", minimumFavorites: 100.
• Competitor mentions — searchTerms: ["ChatGPT vs Claude", "Cursor vs Claude", "Copilot vs Claude", "best AI tool 2026"], maxItems: 30, sort: "Top", minimumFavorites: 50.
• Viral AI content — searchTerms: ["AI changed my workflow", "AI saved me hours", "built this with AI"], maxItems: 30, sort: "Top", minimumFavorites: 200.

Phase 2: Expand (after Phase 1 data is back)

• Video details — Use getVideoDetails on promising YouTube results to get full stats (views, likes, duration)
• Transcript expansion — Pull getTranscripts (key_segments) on any standout/outlier videos

Phase 3: Synthesize (after all data is collected)

With all collected data:

1. Deduplicate all tweets across watchlist and discovery results by tweet ID. Flag any discovery tweets that DON'T overlap with watchlist results — these are the net-new signals from outside the watchlist.

2. Cross-reference official sources with Twitter signals. If a release matches high-engagement tweets, that's a strong video signal.

3. Identify patterns across all three data sources: what topics are showing up in official sources AND Twitter AND YouTube simultaneously?

4. Present ONE unified summary in chat, organized by recency time slots. Group all signals (official sources, YouTube, Twitter) into these buckets based on when they were posted/published:

   Time slots (use these exact headings):

   • 🔴 Last 3 hours — breaking right now, act-on-it-today signals
   • 🟠 3-6 hours ago — fresh, still developing
   • 🟡 6-12 hours ago — happened today, worth watching
   • 🔵 12-24 hours ago — yesterday's signals still circulating
   • ⚪ 24-48 hours ago — older tail, include only if high engagement

   Within each time slot, list items across all sources together (not separated by platform). For each item include:

   • Source tag: [Twitter], [YouTube], [GitHub Release], [Changelog], [Blog]
   • The signal: what happened, who posted it, key numbers
   • Engagement: likes/views/retweets as applicable

   After the time slots, add:

   📊 Cross-platform patterns — topics appearing across multiple sources and time slots simultaneously. These are the strongest signals.

   🎬 Video opportunities:

   • Big Feature Launch opportunities: any major Cowork feature or update dropped?
   • Bundled Small Updates opportunities: minor updates accumulating?
   • How-To for Business opportunities: business use cases getting traction?

   For each opportunity, note which time slot the signal is in so Ben knows how urgent it is.

Phase 4: Build the dashboard (ALWAYS — see "Dashboard Output")

After presenting the chat summary, render the instant-ui dashboard from the synthesized data.

Competitive Research ("research channels", "analyze competitors", "niche analysis")

1. getChannelStatistics to pull subscriber counts, total views, and video counts for competitor channels (AI Jason, Corbin Brown, Jack Roberts, and any others discovered)
2. getChannelTopVideos to get recent/top videos from each competitor channel
3. searchVideos to find new creators in the professional AI tools niche
4. getVideoDetails on standout videos for full metadata
5. vidIQ deep dive on each competitor:
   • vidiq_channel_analytics — engagement rate, view velocity, growth signals
   • vidiq_channel_performance_trends — long-arc trajectory (rising/declining/flat)
   • vidiq_outliers — auto-detect their viral wins so we know what's working for them
   • vidiq_similar_channels — discover new competitors related to each known one
6. vidiq_breakout_channels in the niche to surface fast-rising channels that aren't yet on our radar
7. Twitter/X watchlist scan — All tiers (P1 through P4) scanned with same methodology as Daily Scan Phase 1, plus targeted searchTerms for competitor names and sentiment keywords. Cross-reference Twitter signals with YouTube competitor performance.
8. Agent decides which outlier videos need transcript analysis → getTranscripts (free)
9. Cross-reference YouTube, vidIQ, and Twitter findings
10. Build the dashboard (Phase 4 — see "Dashboard Output")

Idea Generation ("video ideas", "brainstorm", "content ideas", "ideation")

1. searchVideos to find content gaps, trending formats, and high-performing videos in the niche
2. getVideoDetails on top results for engagement analysis
3. vidIQ keyword research — Run vidiq_keyword_research on every candidate topic (e.g., "Claude Cowork tutorial", "AI agent for business", "no code AI workflow"). Capture search volume + competition score for each. Reject topics with low search volume; flag low-competition + high-volume topics as priority opportunities.
4. vidIQ outlier mining — vidiq_outliers on each known competitor channel to identify their recent breakouts. These are validated angles worth riffing on.
5. vidIQ trending discovery — vidiq_trending_videos and vidiq_breakout_channels to surface emerging formats and creators.
6. Twitter/X watchlist scan — All tiers (P1 through P4) scanned with same methodology as Daily Scan Phase 1, plus keyword scan for pain points and emerging topics. Use minimumFavorites: 100 on keyword scan to surface only high-signal tweets. Focus on tweets showing demos, workflows, and setups getting outsized engagement.
7. Agent identifies high-performing videos worth analyzing → getTranscripts for angle inspiration
8. Apply strategy test from Domain Knowledge section
9. Categorize every idea into one of the 3 video formats: Big Feature Launch, Bundled Small Updates, or How-To for Business
10. Score draft titles with vidiq_score_title before presenting — surface the SEO/CTR score alongside each idea
11. Present ideas with supporting data from all platforms (YouTube + vidIQ + Twitter)
12. Build the dashboard (Phase 4 — see "Dashboard Output")

AI Trend Scout ("trend scout", "what's trending on twitter", "twitter scan", "X scan", "viral topics")

Structured Twitter/X deep analysis for YouTube content opportunities. Scans the watchlist and keyword landscape to identify what's gaining viral momentum RIGHT NOW so you can film before the trend peaks.

Priority tiers (scan in this order):

1. Tier 1 — Anthropic/Claude ecosystem: Claude Code, Claude Cowork, Claude Desktop, Claude API changes, Anthropic product announcements, viral Claude demos, Claude workflow setups people are sharing
2. Tier 2 — Other AI tools trending: OpenAI, Google, open-source models, new tools, integrations, or workflows that are trending
3. Tier 3 — Meta-trends: Recurring themes or debates dominating AI Twitter (e.g., "agents replacing SaaS", "vibe coding")

Execution — run ALL in parallel: