Not configs, a system, and not a tool you run

ECC describes itself as the harness-native operator system for agentic work, and both halves of that phrase matter. It is not a single application you install and launch. It is a curated layer of skills, hooks, rules, MCP configurations, memory optimization, security scanning, and research-first workflows that sits on top of the coding agents you already use, distilled from more than ten months of intensive daily product building. It works across Codex, Claude Code, Cursor, OpenCode, Gemini, Zed, and GitHub Copilot, so the practices are portable across harnesses rather than tied to one.

The honest framing, which the README states plainly, is that the repository is the raw code only and the guides explain everything. That is the single most important thing to understand before starring it: ECC is as much a body of knowledge as it is a set of files, and the files without the reasoning behind them are dotfiles you did not write.

What the system actually covers

The guides map to the recurring problems of running agents seriously:

  • Token optimization: model selection, slimming system prompts, and pushing work to background processes.
  • Memory persistence: hooks that save and load context across sessions automatically, so an agent does not start cold each time.
  • Continuous learning: auto-extracting patterns from sessions into reusable skills, so the setup compounds.
  • Verification loops: checkpoint versus continuous evals, grader types, and pass@k metrics, the testing discipline most agent setups skip.
  • Parallelization: git worktrees and a cascade method for when to scale up multiple agent instances.
  • Security: attack vectors, sandboxing, sanitization, and CVEs, an area most agent dotfiles ignore entirely.

That security and verification emphasis is what separates ECC from a casual collection of prompts. It treats running agents as an engineering practice with failure modes, not a vibe.

How it is distributed

ECC is MIT-licensed and the code is free. The surrounding product has tiers: a free GitHub App with PR audits, ECC Pro at $19 per seat per month for private repos, and an option to fund the open-source work from $5 a month. The thing to register is that the deepest operational guidance lives in the guides and the paid surface rather than scattered through the code. You can absolutely read the repo and adopt pieces for free, but the curated path is a product.

The project hit v2.0.0 in June 2026, which adds a public Hermes operator story on top of the reusable layer, so it now documents how to drive the Hermes Agent with the same practices.

Where it fits, and where it does not

Reach for ECC if you run coding agents daily and you want a vetted starting point for the unglamorous parts, memory, evals, sandboxing, parallel runs, rather than reinventing each one. As a reference for how a heavy user actually configures these harnesses, it is unusually complete, and the cross-harness portability means the lessons survive your switching tools.

Do not reach for it expecting an app. There is nothing to launch; you are adopting configuration and practice. If you run an agent occasionally for small tasks, the system is more than you need, and the value is in the guides as much as the files. With a comparatively small 57 open issues as of 2026-06 against a very large star count, the repo itself is stable; the activity is in the surrounding guides and product.

A note on positioning

ECC sits in a young category of opinionated agent operator setups, alongside community skill collections and harness dotfiles. Its distinguishing move is breadth and rigor in one place: it does not just share prompts, it covers the full operational loop including the security and verification work that most collections leave out. If you only want a grab-bag of skills, lighter collections exist; if you want a coherent system with the reasoning attached, that is ECC’s pitch.

ECC’s v2.0 leans into driving the Hermes Agent, and it configures the same family of tools that cc-switch helps you manage. For what else is climbing, see LLM tooling, the daily digest, and the weekly report.

FAQ

Is ECC an app I install? No. It is a layer of skills, hooks, rules, and MCP configs you adopt on top of your existing coding agents, with guides that explain the practices.

Is it free? The code is MIT-licensed and free. There is a free GitHub App, a Pro tier at $19 per seat per month for private repos, and a sponsorship option from $5 a month.

Which agents does it work with? Codex, Claude Code, Cursor, OpenCode, Gemini, Zed, GitHub Copilot, and others, plus a Hermes operator story added in v2.0.0.

Do I need the guides? Effectively yes. The README says the repo is raw code only and the guides explain everything, so the reasoning behind the files is where the value concentrates.