Rising AI repos
Top 10 rising AI repos, week of Oct 26 (ET)
ET · 10 repos
Data: ET · As of · Published
Total stars are a snapshot at that time. Stars gained come from GitHub’s daily star data, whose day boundaries may not line up with ET midnight.
The short version
morluto/rea, an MCP server that lets agents inspect binaries and apps, gained +30,956 stars, about 2.9 times the runner-up. Impeccable and DeepTutor followed with about +10.8k and +10.5k.
Five of the ten are agent skills (design, security or process). Two more are multi-agent workspaces (T3 Code and Orca), and DeepSeek’s harness stayed in the top ten.
-
morluto/rea
+30,956 stars gained. One MCP server that lets your agent inspect native binaries, apps and websites, then explain how a feature works down to the binary.
Why it matters for agent builders: It puts heavy local reverse-engineering tools behind a single MCP, a pattern worth copying when you wire desktop tools into an agent.
Source: morluto/rea on GitHub
-
pbakaus/impeccable
+10,807 stars gained. A design vocabulary for coding agents. Steer with polish and bolder, review with critique and audit, and catch AI-looking defaults before they ship.
Why it matters for agent builders: Agents keep reinventing the same cream page and purple gradient. A shared vocabulary is a low-effort way to push them off that default.
Source: pbakaus/impeccable on GitHub
-
HKUDS/DeepTutor
+10,470 stars gained. An agent-native learning workspace that ties tutoring, research, reading and practice into one system, with partners that can call Claude Code or Codex.
Why it matters for agent builders: It treats learning as something another agent can operate (JSON event streams, headless runs), not only a chat UI for people.
Source: HKUDS/DeepTutor on GitHub
-
Nutlope/hallmark
+9,887 stars gained. A design skill for Claude Code, Codex, Cursor and OpenCode that refuses machine-looking pages, with fifty themes and browser lint gates.
Why it matters for agent builders: Floor, Reflex and Finish gates give the agent a hard bar on contrast, layout tells and empty frames instead of another soft style prompt.
Source: Nutlope/hallmark on GitHub
-
cloudflare/security-audit-skill
+8,451 stars gained. A coding-agent skill that runs a multi-phase security audit with isolated hunter and verifier agents and writes findings against a JSON schema.
Why it matters for agent builders: A worked example of find-then-try-to-disprove, which may cut false positives in any agent that reports problems.
Source: cloudflare/security-audit-skill on GitHub
-
deepseek-ai/deepseek-harness
+6,378 stars gained. DeepSeek’s open-source agent harness (dsh), built so that everything is a plugin. It is a developer preview with breaking changes expected.
Why it matters for agent builders: A model lab’s own harness shows which plugin points it thinks an agent needs. Expect the APIs to move while it is in preview.
Source: deepseek-ai/deepseek-harness on GitHub
-
pingdotgg/t3code
+6,222 stars gained. An open agent-harness control surface with mobile, web and desktop apps that drive Claude Code, Codex, Cursor, Grok Build, OpenCode and others already on your machine.
Why it matters for agent builders: Remote-ready control of agents you already pay for is a crowded niche. An MIT app you can fork is a useful baseline.
Source: pingdotgg/t3code on GitHub
-
stablyai/orca
+6,012 stars gained. A desktop and mobile workspace that runs coding agents side by side, each in its own git worktree, tracked in one place.
Why it matters for agent builders: Fanning one prompt across several agents and keeping the best result is becoming a common workflow, and this shows one way to manage it.
Source: stablyai/orca on GitHub
-
mattpocock/skills
+5,573 stars gained. A set of small, composable agent skills for everyday engineering work, installable as a plugin for Claude Code, Codex and GitHub Copilot.
Why it matters for agent builders: Short, editable skills you own are an alternative to process frameworks that take over the whole workflow.
Source: mattpocock/skills on GitHub
-
DietrichGebert/ponytail
+5,558 stars gained. A skill that tells your coding agent to write only the code a task needs, reuse what the repo already has, and keep validation, error handling and tests.
Why it matters for agent builders: Its own Claude Code benchmark claims less code and fewer tokens with tests kept. Worth checking against your own tasks before you rely on it.
Source: DietrichGebert/ponytail on GitHub
What to watch next
- rea is still #1 after last week’s +60k spike, but this period’s gain halved. Next chart shows whether the climb cools further.
- Impeccable and hallmark both surged on anti-slop messaging among design skills. See which keeps climbing once the novelty fades.
- pingdotgg/t3code is new to this chart at +6,222. Watch whether remote control of existing agent subscriptions holds interest.