LLM Coding Assistant Census 2025
TL;DR - Key Take-aways
Only 1.3% of top 100,000 repos have checked-in LLM coding assistant instructions, but usage has grown 8× from 166 to 1,316 repos in 2025. Claude Code took the lead from Cursor, with other TUI LLM tools also having strong growth in the last 3 months.
What I measured (and what I could not)
I queried top 100,000 public starred repos on GitHub and matched their files against signatures of known tools. For example for Claude Code I looked for presence of CLAUDE.md.
Since use of a code assistant is a personal choice and strongly tied to the coder rather than the code base the expectation is that not many repos will have optional instructions for particular tool or agent. This makes it a weak signal for overall code assistant adoption, but a good one for relative adoption of the tools.
OpenAI Codex has embraced the use of the non-tool specific AGENTS.md as its only instruction file. Hopefully more tools will adopt in the future, making coder’s lives easier when working on other people’s repositories. For now, I gave all the uses of AGENTS.md to OpenAI Codex, which is likely the most common tool reading it right now; this will overstate its standing in the charts. In the future however, adoption of a standard AGENTS.md will make this kind of analysis much less useful.
Leaderboard
Claude Code leads the number of repos with 521 out of 100,000, with Cursor in second place at 411 repos. GitHub Copilot, OpenAI Codex and Gemini CLI round out the current top 5. All of top 5 tool are either closed source or open source but closely tied to an LLM provider. Pure OSS tools from non-llm providers like cline and aider are at the bottom of the chart or like continue and crush failed to pass the low 5 repo bar for making it onto the chart. This is surprising since public repos on GitHub are generally OSS, and OSS developers prefer OSS tools. Overall only 1,316 out of 100,000 repositories had instructions for any kind of assistant.
You have one assistant yes, but what about a second assistant?
Big popular repositories are worked on by multiple developers so we would expect them to have instructions for multiple llm assistants. Multiple, potentially conflicting, llm instruction files issue could be solved by adopting a standard file or mitigated by symlinking tool-specific files to single one, as discussed for the Linux kernel. With low adoption of checking-in llm instructions into repos, the proliferation of files has not yet become a problem. Out of 1,316 repos with instructions, 245 had instructions for more than one. The two most popular assistants, Claude Code and Cursor, having the most popular combination at 105 repos is unsurprising.
Motivation
In my own work, I use both Claude Code and Cursor and with their instructions symlinked to AGENTS.md. I got curious about how other people people manage their llm coding assistants, and since I did a census of task runner usage, the tooling was already there. The motivation for the task runner census itself came from figuring out which runners I should support in dela, my OSS universal task runner.
Appendix
Repository data was collected between July 28th 2025 and July 31th 2025. Timeline data was collected on August 5th 2025.
LLM assistants with repo and star count
| llm_assistant | repo_count | repo_star_sum |
|---|---|---|
| claude-code | 521 | 3,577,716 |
| cursor | 411 | 3,470,711 |
| github-copilot | 274 | 2,466,648 |
| openai-codex-cli | 201 | 1,818,098 |
| gemini-cli | 118 | 1,231,781 |
| sweep-ai | 44 | 185,628 |
| windsurf | 32 | 326,805 |
| cline | 28 | 198,736 |
| aider | 5 | 7,804 |
| sourcegraph-cody | 3 | 20,697 |
| crush | 3 | 7,418 |
| continue | 1 | 27,948 |
Most popular repository by LLM assistant
| llm_assistant | repository | stars |
|---|---|---|
| aider | storj/storj | 3,141 |
| continue | continuedev/continue | 27,948 |
| cursor | kamranahmedse/developer-roadmap | 332,466 |
| windsurf | f/awesome-chatgpt-prompts | 131,503 |
| github-copilot | microsoft/vscode | 175,075 |
| claude-code | Significant-Gravitas/AutoGPT | 177,288 |
| sourcegraph-cody | sourcegraph/sourcegraph-public-snapshot | 10,185 |
| sweep-ai | AntonOsika/gpt-engineer | 54,608 |
| gemini-cli | flutter/flutter | 171,565 |
| openai-codex-cli | Significant-Gravitas/AutoGPT | 177,288 |
| crush | Creators-of-Create/Create | 3,597 |
| cline | zed-industries/zed | 63,049 |
LLM instructions adoption by repos timeline
Note that this timeline data came from a different set of queries run on August 5th, and thus its number of repos August 1st doesn't tie out with the number of repos at the end of July used throughout this blogpost.| month | cumulative_repositories |
|---|---|
| 2024-01-01 | 67 |
| 2024-02-01 | 70 |
| 2024-03-01 | 75 |
| 2024-04-01 | 80 |
| 2024-05-01 | 82 |
| 2024-06-01 | 92 |
| 2024-07-01 | 97 |
| 2024-08-01 | 101 |
| 2024-09-01 | 107 |
| 2024-10-01 | 117 |
| 2024-11-01 | 130 |
| 2024-12-01 | 148 |
| 2025-01-01 | 166 |
| 2025-02-01 | 199 |
| 2025-03-01 | 275 |
| 2025-04-01 | 389 |
| 2025-05-01 | 510 |
| 2025-06-01 | 703 |
| 2025-07-01 | 1,010 |
| 2025-08-01 | 1,344 |