Choose useful AI skills for your workflow

Context & limitations

Every headline number is exactly right, which almost never happens. I counted them in the repository rather than taking anyone's word: 292 files named SKILL.md under skills/, each one its own folder, and 68 markdown files under agents/. The README states the same two numbers, plus 94 commands, and there are 122 rule files on top. MIT licensed, genuinely. Worth stating plainly because the write-ups about this project are badly stale — pieces still circulating describe it as "13 agents and 40+ skills", which was true at some point and is now off by a factor of seven.

The scale of the thing is the part the caption undersells. The repo was created on 18 January 2026 and as I write this it is on 260,851 stars and 39,039 forks, with commits landing today. Zero to a quarter of a million stars in eight months, maintained by one person shipping across seven harnesses. That is also why the malware warning in its own README exists: a project this visible attracts mirrors, and the official channels are the GitHub repo, the ecc-universal and ecc-agentshield npm packages, the GitHub App, the plugin slug ecc@ecc, and ecc.tools. Nothing else.

"One plans, one reviews, one fixes broken builds" is a fair description of the design and it hides what the 68 actually is. Sorted by name: about 23 reviewers, 12 build-error resolvers and 7 planners or architects — three roles, repeated per language. There's a python-reviewer, a rust-reviewer, a swift-reviewer, a vue-reviewer; a react-build-resolver, a pytorch-build-resolver, a kotlin-build-resolver. That's a good design, not a padded one, and it's the single most useful fact for deciding what to install: if you write TypeScript, most of the 68 are for languages you don't use, and the number that applies to you is small. Same for the rules, which the README explicitly says to take one stack pack of rather than wholesale.

The cross-tool claim needs the project's own matrix, because it doesn't rank those three as equals. ECC publishes a support table: Claude Code is the "stable primary", Codex is a "supported native plugin", Cursor is a "beta project adapter" with agent discovery that varies by build, and Gemini appears in the last row alongside Zed, Qwen and Kimi under "experimental/minimal adapters", where the README's own words are that "full Claude feature parity is not claimed". GitHub Copilot is instruction-only — no hooks, no runtime agents, no delegation. So it is genuinely portable, and portable is not the same as equivalent, and the project is the one saying so.

Then "free", which is true and slightly smaller than it sounds. The repository is MIT and the README commits to that permanently — "OSS stays free". Around it sits a business: ECC Pro, a hosted GitHub App for private repos from $19 per seat per month, GitHub Sponsors, and sponsor and partner links in the README itself. None of that is hidden and none of it makes the repo less free; it's the ordinary difference between a free repo and a free project, and it's what pays for one maintainer shipping weekly. The cost that will actually reach you is a different one. The README notes that the plugin advertises the installed catalogue to the model and recommends a selective profile "when context footprint matters" — which is the maintainer telling you, in the install instructions, that 292 skills announced at the top of every session is a bill you pay per session, forever. That's the whole argument for the prompt above, and it's why the useful version of this download is a short list.