Five agent repos gaining momentum. Every one measured.

Warming up: ranked by stars per day since creation.

Short candidate pool today: 5 of 5 picks were filled with relaxed rules (5 by star gains since the last snapshot).

Rank 1

morluto/rea

REA is an MCP server and CLI (npm package rea-agents) that connects AI agents to reverse engineering tools for native binaries, JavaScript/Electron apps, .NET assemblies, and websites. Analysis runs locally and results include the evidence and limitations behind each conclusion; native analysis can use Hopper, Ghidra, or IDA.

Developers building agents that need to inspect how an existing app or binary works, without its source code, can register REA with agents like Claude Code, Codex, Cursor, or Gemini CLI via `npx rea-agents setup`.

★ 293,747
MIT
README ≈ 8,041 tokens to load into context

AI-written summary

Score95.6

Rank 2

affaan-m/ECC

ECC is an open-source (MIT) harness layer for AI coding agents, described as bundling skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor, and others. It installs as the `ecc@ecc` plugin for Claude Code and is also distributed via npm packages (`ecc-universal`, `ecc-agentshield`), with a separate paid hosted GitHub App for private repos.

Developers who run Claude Code, Codex, Cursor, or similar agents and want a shared set of skills, memory, and security tooling for them should look at it, though the provided README excerpt does not detail how these components work.

★ 281,652
MIT

AI-written summary

Score67.3

Rank 3

stablyai/orca

Orca is an MIT-licensed desktop app (macOS, Windows, Linux) with a mobile companion that runs multiple coding agents such as Codex, Claude Code, OpenCode or Pi side by side, each in its own git worktree, tracked in one place. It also provides terminal splits, an embedded browser, GitHub and Linear integration, diff annotation, and SSH remote worktrees, and you use your own agent subscriptions.

Developers who run several coding agents in parallel and want isolated worktrees, one place to compare their results, and the option to steer them from a phone or remote machine should look at it.

★ 94,623
MIT
README ≈ 8,087 tokens to load into context

AI-written summary

Score63.2

Rank 4

DietrichGebert/ponytail

Ponytail is a skill/prompt package for AI coding agents (the README says it works with 20 agents, and it ships on npm) that pushes the agent to write only the code a task needs, reusing existing repo components instead of building new ones. It also includes a /ponytail-review command, and the README reports benchmark results from Claude Code runs on 39 tasks.

Developers who build or use coding agents and want smaller diffs and fewer tokens without dropping validation, error handling, security, or tests should look at its benchmark method and rules.

★ 165,319
MIT
README ≈ 5,855 tokens to load into context

AI-written summary

Score61.3

Rank 5

Leonxlnx/taste-skill

Taste Skill is a set of Agent Skills for coding agents (Codex, Cursor, Claude) that has the agent pick an art direction, source real photographs from open archives with credits, and check its own output in a real browser before finishing. It installs via the `npx skills` CLI, and its scripts need Node 22+ and a local Chrome, Edge or Chromium.

Developers whose coding agents generate frontends that all look like the same generic landing page may want to try it, since it adds design direction, real imagery and a browser-based self-review step.

★ 98,094
MIT
README ≈ 7,939 tokens to load into context

AI-written summary

Score56.4