AI Ecosystem Intelligence

You measure what the dev community is BUILDING AROUND each competitor (external signal). Distinct from competlab-agent-adoption which measures FIRST-PARTY agent infrastructure.

Workflow

Step 1: Identify competitors

list_projects + list_competitors.

Step 2: For each competitor, probe public APIs (all unauth-OK)

GitHub org meta:

curl -sS https://api.github.com/orgs/{slug} | jq '{public_repos, followers, created_at}'

Try slug variants: {brand}, {brand}-inc, {brand}io, {brand}HQ (BetterStackHQ style), {brand}-ai.

Top-N starred repos:

curl -sS "https://api.github.com/search/repositories?q=org:{slug}&sort=stars" | jq '.items[0:5] | .[] | {name, stargazers_count, description}'

MCP/agent search:

curl -sS "https://api.github.com/search/repositories?q={brand}+mcp+in:name" | jq '{total_count, items: .items[0:3]}'

And same for: {brand}+agent-toolkit, {brand}+claude-skill, {brand}+claude-connector.

npm downloads:

curl -sS "https://api.npmjs.org/downloads/point/last-month/{pkg}" | jq '.downloads'

Try variants: {brand}, @{brand}/sdk, @{brand}/{brand}, {brand}-sdk, {brand}-client, {brand}-api.

PyPI downloads: check via https://pypistats.org/api/packages/{pkg}/recent (or pepy.tech for richer data).

Step 3: For brands with no obvious public surface (e.g., bootstrapped operator with no GitHub org)

Use Perplexity to confirm "X has no GitHub presence; their developer surface is API-only / closed" — that's itself an Ecosystem signal in a category where peers DO have GitHub presence.

Step 4: URL-verify any Perplexity-returned GitHub URLs

Per validation in the CRM category: Perplexity hallucinates GitHub URLs with citation-grade star counts. ALL claimed GitHub URLs MUST be URL-verified via mcp__competlab__fetch_url OR direct api.github.com/repos/{owner}/{repo} check before surfacing in output.

Step 5: Marketplace presence (lighter pass)

HubSpot App Marketplace, Salesforce AppExchange, VS Code Marketplace, Anthropic MCP catalog, Smithery.ai (where applicable to category). One quick search each.

Step 6: Synthesize per-vendor + cross-vendor

Per vendor:

• GitHub org: repo count, follower count, created year, top-N repos by stars
• npm monthly downloads (with package names verified-real)
• Community-built MCP servers found (with star counts, verified-real)
• Marketplace listings (where applicable)
• Verdict: "rich developer ecosystem" / "moderate" / "early-stage" / "no public surface"

Cross-vendor:

• Coverage: X of N vendors have public GitHub org
• Volume comparison (e.g., "category-leading vendor publishes multi-million npm/month while niche peer ships single-digit downloads — orders-of-magnitude gap is itself the developer-ecosystem read")
• Pattern detection (e.g., "community-built MCP demand outpaces official builds for X — indicates strong dev-pull")
• Strategic gaps (e.g., "vendor X has Series B funding but zero public dev surface — atypical")

Output Structure

# AI Ecosystem — [Category / Project]
> Generated [date] | X of N citations URL-verified via GitHub API / fetch_url

## Summary
[Coverage breakdown, volume range, strategic gap signals]

## Per-competitor
### [Competitor]
- GitHub: org [repos/followers/created] / top-N repos / NONE (with signal interpretation)
- npm: [pkg → downloads/month] for top SDKs / NO SDK FOUND
- Community MCP servers: [list of verified-real repos with stars] / NONE FOUND
- Marketplace: [Hub/Salesforce/VS Code if relevant]
- Verdict: [rich / moderate / early / none]

## Cross-competitor patterns
[Coverage gap, volume orders-of-magnitude difference, dev-pull signals]

Decision Questions

1. "Do we want to build a public GitHub org if we don't have one yet?" — material if we're in a category where peers have one
2. "Should we encourage community-built MCP servers (link to community-built ones from our docs like Smartlead does) or only have first-party?"

Error Handling

• GitHub API 403 rate-limit (60/hr unauth): complete the vendors already probed; record remaining as "GitHub data unavailable this run." Suggest authenticated mode (GH_TOKEN) in output if user has it. Don't synthesize from cached or stale data.
• npm / PyPI registry 5xx: retry once with 10s delay; if persistent, mark "package registry data unavailable" rather than dropping vendor silently.
• Perplexity citation 404 / repo URL hallucinated: strip silently, log to _internal/url-verification-log.md. Real-world validation found ~5 of 7 Perplexity-cited GitHub repo URLs were fabricated (with citation-grade confidence). The skill MUST verify every github.com/X/Y URL via api.github.com/repos/X/Y (HTTP 200 + matching owner) before surfacing.
• Vendor has zero GitHub org / zero npm presence: record as "no developer-ecosystem signal detected." Common for B2B SaaS without a developer surface; surface as classification finding rather than implying failure.

What NOT To Do

• Don't list GitHub URLs without verifying they exist (Perplexity hallucination risk observed in real-world validation).
• Don't conflate first-party with community-built — they're different signals. First-party = agent-adoption skill. Community-built + npm/marketplace = ai-ecosystem skill.
• Don't over-weight a single repo — a 9k-star OSS competitor under another org (like bluewave-labs/Checkmate for UptimeRobot) means a competitor is filling YOUR brand's GitHub surface.