Skip to content

research_census_2026.md

Latest commit

 

History

History
72 lines (53 loc) · 3.34 KB

research_census_2026.md

File metadata and controls

72 lines (53 loc) · 3.34 KB
Raw

📊 2026 State of AI Agent Skills & MCP Ecosystem Censuses

This document details the methodology, metrics, and key takeaways from the 5 Empirical Censuses conducted by the GitTrends AI Research Desk.

1. Executive Summary

Census Report Key Metric Finding & Significance
The Agent Economy Census Gini Index: 0.96 Extreme power-law install distribution; top 1% of skills capture 78.4% of all verified agent sessions.
The Agent Security Census 87.8% Unaudited High ambient execution permissions; 4.2% contain dangerous credential read patterns.
The Agent Clone Census 15.8% Duplicate High rate of copycat clones across Claude and Cursor marketplaces with minimal attribution.
The Agent Use-Case Census 44.2% UI/Frontend Frontend design, browser automation, and TDD testing dominate developer adoption.
The Maintenance Census Median Freshness: 6 days Thriving ecosystem vitality with rapid patch cadence matching core model updates.

2. Deep Dive: The 5 Censuses

Census 1: The Agent Economy Census

  • Dataset Size: 79,848 catalogued packages and repositories across GitHub, npm, and community registries.
  • Install Concentration: 164.8M total verified tool invocations.
  • Economic Model: 99.93% of agent skills are distributed under permissive open-source licenses (MIT, Apache 2.0). Monetization exists primarily via cloud API backends rather than upfront licensing.

Census 2: The Agent Skill Security Census

  • Vulnerability Breakdown:
    • Unscoped shell execution (exec, system): 31.2%
    • Unchecked file write permissions: 18.7%
    • Potential environment credential leakage: 4.2%
  • Recommendation: Enterprise teams should implement permission gating and sandbox execution for third-party MCP servers.

Census 3: The Agent Skill Clone Census

  • Distribution: 12,610 duplicate or near-identical prompt variants discovered across 79,848 listings.
  • Primary Cloned Packages: Generic prompt collections ("Senior React Engineer", "Python Expert") show 40+ verbatim forks.

Census 4: The Agent Use-Case Census

  • Top 5 Domain Adoption Share:
    1. Frontend, Glassmorphism & UI Taste: 44.2%
    2. Browser Automation & Testing: 23.5%
    3. Database & SQL Orchestration: 14.1%
    4. Security & Penetration Testing: 10.3%
    5. Documentation & Spec Generation: 7.9%

Census 5: The Agent Skill Maintenance Census

  • Repository Health: 81.4% of indexed packages had at least one commit in the last 30 days.
  • Decay Velocity: Only 6.2% of skills suffered from unmaintained abandonware status (>180 days stale).

3. Raw Data Access

Researchers can download the complete raw census JSON dataset:

4. Citation & Academic Reference

If citing these censuses in research papers, articles, or blog posts, please use:

@misc{gittrends_ai_census_2026,
  author = {GitTrends Research Desk and jastfan},
  title = {2026 Empirical Census of the AI Agent Skills and MCP Ecosystem},
  year = {2026},
  publisher = {GitHub},
  howpublished = {\url{https://jastfan.github.io/github-trending/}}
}