State of AI Agents

2026-Q2 Report

We evaluated 16 public AI agent config repositories selected from high-star GitHub search results across Claude Code, Cursor, and Copilot. Here's what we found.

Repos evaluated

16

Platforms covered

3

Generated

2026-04-21

By Platform

PlatformReposAvg ScoreTiers
claude-code 7 0.604 7 usable
cursor 4 0.569 3 usable, 1 experimental
copilot 5 0.699 2 strong, 3 usable

All Repos

#RepositoryPlatformStarsTierScoreCommit
01 forrestchang/andrej-karpathy-skills claude-code 68,336 Usable 0.613 2c60614
02 PatrickJS/awesome-cursorrules cursor 39,180 Usable 0.589 fc2ce04
03 wasp-lang/open-saas claude-code 14,183 Usable 0.572 746cea4
04 Piebald-AI/claude-code-system-prompts claude-code 9,188 Usable 0.578 32f09bd
05 grapeot/devin.cursorrules cursor 5,960 Usable 0.585 284b743
06 drona23/claude-token-efficient claude-code 3,644 Usable 0.621 23a4dab
07 Piebald-AI/tweakcc claude-code 1,817 Usable 0.573 6062474
08 kinopeee/cursorrules cursor 1,118 Usable 0.590 46ec520
09 BayramAnnakov/claude-reflect claude-code 921 Usable 0.648 8dc9db4
10 iamfakeguru/claude-md claude-code 913 Usable 0.623 4036d80
11 s-smits/agentic-cursorrules cursor 376 Experimental 0.511 f8df721
12 Robotti-io/copilot-security-instructions copilot 39 Strong 0.757 0875363
13 he-yufeng/RuleForge copilot 31 Usable 0.589 bd64e60
14 andreaswasita/copilot-agents-dojo copilot 29 Usable 0.677 90a30bb
15 CsabaKovacs/rulesmith copilot 12 Usable 0.655 971db3c
16 jonhill90/vibes copilot 8 Strong 0.817 8ab8fd3

Top failures

  • Prompt-injection and jailbreak guidance is missing
  • Model is not specified
  • Trigger description is vague

Top passes

  • (no common passes found)

Sample selected from the highest-starred public GitHub search results for CLAUDE.md, .cursorrules, copilot-instructions.md, and AGENTS.md. Only repos with supported agent config files discovered by subagent-evals are included. Results reflect pinned commits and may not generalize.