Agent skill repos with the most new GitHub stars, 12–17 October 2026: one repo took 74%

Between our 12 and 17 October snapshots, the 74 agent-skill repositories that Skill Rank tracks daily gained 261,307 GitHub stars, and one of them took 192,624 of those: [morluto/rea](https://github.com/morluto/rea), a reverse-engineering toolkit for coding agents. That is 74% of all new stars. The other 73 repositories shared 68,683; the median among them gained 288.

The numbers come from Skill Rank, our agent-skills leaderboard. Once a day it records stars, forks, open issues, open pull requests and pull requests merged in the last 90 days for every repository behind a skill in the catalogue, straight from the GitHub API. The window here runs from the snapshot taken between 20:21 and 21:22 UTC on 11 October (stored as 12 October) to the one taken between 02:39 and 03:48 UTC on 17 October, about five and a quarter days. Two repositories that GitHub lists under both an old and a new name are counted once, under the new name.

The biggest star gains, 12–17 October

The live version of this table, with 7-day and 30-day windows, is on the Skill Rank trending page.

rea: from 99,616 to 292,240 stars in five snapshots

rea calls itself "one MCP for reverse engineering across binaries, applications, and runtime behavior". Its README describes a setup command, npx rea-agents setup, that registers an MCP server and workflow instructions with agents such as Claude Code, Codex, Cursor and Gemini CLI, backing up existing configuration first. Native analysis can drive an existing Hopper, Ghidra or IDA install, and the README says analysis runs locally. The repository is MIT-licensed and was created on 14 April 2026. We catalogue its bundled skill as reverse-engineer-anything.

Here is what our snapshots recorded:

The pace has fallen every day, but its last daily gain was still 16 times that of the next-fastest repository (mattpocock/skills, +1,188 in the same snapshot). rea started climbing before our tracking did. A GitHub-trending account on X put it at 7,743 stars on 6 October, and a post by Matthew Berman on 8 October had about 3.8 million views and 45,000 bookmarks when we checked.

Two other numbers set rea apart from the rest of the list:

• Forks. rea gained 27 forks for every 100 new stars. Among the other repositories that gained more than 500 stars, the median was 9 forks per 100.
• Pull requests. rea merged 1,513 pull requests in the last 90 days and had 474 open. Only five tracked repositories merged more: openclaw/openclaw (38,796), NousResearch/hermes-agent (10,416), sst/opencode (4,040), code-yeongyu/oh-my-openagent (1,965) and career-ops-hq/career-ops (1,792). Most of those are full agent apps or harnesses, not a single skill.

Star and fork counts alone can't tell you why a repository grows. A viral post, a crowd of contributors forking to send pull requests, or automated activity can all push the same numbers up, and the GitHub API doesn't separate them. Treat rea's count as a sign of attention, not as a security or quality review. Its own README warns that "a similar repository name or copied README does not establish that a download is official" and points to github.com/morluto/rea, rea.tools and the rea-agents npm package as the official sources. Check a link against those three before you run its setup.

At 17 October rea was 8,476 stars behind obra/superpowers and still adding more than that per day. If the pace holds, it will pass superpowers in the next snapshot and become the second most-starred repository we track, behind openclaw/openclaw (392,208).

Without rea, it was an ordinary week

Take rea out and the tracked repositories gained 68,683 stars, about 13,000 a day. A few things stand out:

• mattpocock/skills is closing on superpowers. The gap between the two shrank from 11,578 stars to 6,553 over the window, about 960 a day. superpowers itself passed 300,000 stars between the 16 and 17 October snapshots.
• diagram-design grew fastest among large repositories. cathrynlavery/diagram-design added 12.6% in five days (50,154 to 56,478) and merged 164 pull requests in 90 days. Among repositories with more than 10,000 stars, the next-fastest were cursor/plugins (+10.4%), latent-spaces/brag (+7.6%) and K-Dense-AI/scientific-agent-skills (+6.6%).
• Small official repositories moved too. expo/skills (+5.7%), remotion-dev/skills (+4.5%) and supabase/agent-skills (+3.9%) all grew faster than most large community collections.

Stars keep coming after the commits stop

Some of the gains went to repositories with little visible maintenance:

andrej-karpathy-skills is the sixth most-starred repository we track (220,383), yet nobody has merged a change since April. Roo-Code was the only tracked repository to lose stars in the window. If you're choosing between two similar skills, the pull-request counts on Skill Rank's author leaderboard and the activity score on the trending page say more about upkeep than the star count. andrej-karpathy-skills scored 59.7 there on 17 October, below every repository in the trending top ten.

How to use these numbers

Stars measure attention. They don't show whether a skill works, what it changes on your machine or whether it is safe. Use the trending list to find what people are looking at, the skill galaxy to see which authors publish what, and then read the SKILL.md and any setup script yourself before installing. Snapshot data on this page will be wrong for any repository GitHub reports incorrectly, and we only track repositories that already appear in the catalogue, so a fast-growing skill we haven't added yet won't show up here.