affaan-m/ECC was #1 on GitHub’s daily trending repository list when checked on September 6, 2026. GitHub showed 1,486 stars gained that day. The public repository API reported 250,753 stars, 37,717 forks, and JavaScript as the primary language at the same check. Those numbers are a snapshot of attention, not a guarantee that the project fits every team.

ECC is an operating layer for AI coding work. It does not replace the model or turn Claude Code, Codex, or another host into a new foundation model. Instead, it gives the host a coordinated collection of instructions, workflows, hooks, rules, memory tools, and specialist agents so that planning, implementation, review, security checks, and handoff do not have to be improvised in every session.

1. What ECC puts around an AI coding agent

The repository describes a loop that looks like this:

plan -> test -> implement -> review -> verify -> remember -> improve

That loop is the useful mental model. A normal coding agent can generate a patch from a prompt. ECC adds reusable procedures for asking better questions before implementation, running test-driven workflows, reviewing a diff from a fresh context, repairing builds, checking agent configuration, and recording useful knowledge for later sessions.

The package is deliberately broader than a prompt library. The README lists skills for engineering, research, security, documentation, frontend work, data, machine learning, operations, and agent workflows. It also ships installation adapters because the same instructions are not automatically discovered in every coding host.

The current README reports:

Included	Count	Purpose
Agents	68	Planning, review, build repair, security, architecture, and domain work
Skills	286	Reusable task workflows across technical and operational areas
Command shims	94	Convenient entry points while the project moves toward a skills-first surface
Hooks and memory	Runtime surfaces	Enforcement, context persistence, continuous learning, and context controls

The count is useful for understanding the size of the distribution, but installing everything is not automatically the best choice. More instructions can mean more context, more hooks to inspect, and more opportunities for two workflows to overlap.

2. Choose one installation path

ECC’s current recommended entry point for Claude Code is the universal package:

npx ecc-universal setup

The README says this path requires Node.js 18 or newer, Git, and Claude Code 2.1 or newer on PATH. It handles setup, updates, scope changes, and hook-profile choices for the ecc@ecc plugin.

For a reviewed installation across more than one harness, the repository documents:

npx ecc-universal install --guided

That wizard lets you select Claude Code, Codex, and Kimi Code, shows the target and destination for each choice, preflights the selections, and asks for one final confirmation. Provider-owned model and authentication settings remain with the provider; ECC is not a replacement for those settings.

There are also native plugin paths. Claude Code can add the marketplace and install the plugin with:

/plugin marketplace add https://github.com/affaan-m/ECC
/plugin install ecc@ecc

Current Codex releases can use the repository marketplace path:

codex plugin marketplace add affaan-m/ECC
codex plugin add ecc@ecc
codex plugin list --json

For other hosts, the README lists target-specific adapters. A minimal Hermes Agent installation is documented as:

./install.sh --profile minimal --target hermes

The important boundary is not the command syntax. Pick one method per harness. Do not install the native Claude plugin, then run a full manual Claude installation over it. Do not stack the Codex marketplace plugin on top of the older sync flow. Duplicate skills, hooks, and configuration make troubleshooting harder and can cause the same action to run twice. If you want an example of a local agent that executes real work and scheduled automation, see our Hermes Agent guide.

3. Start with workflows, not the full catalogue

ECC’s README suggests starting with a specific job. The documented surfaces include:

• /ecc:plan "describe the feature" before building a feature;
• tdd-workflow for a failing-test-first implementation loop;
• /code-review for a fresh-context review;
• /build-fix when a build is broken;
• /refactor-clean for a focused cleanup;
• /context-budget when the conversation is becoming too large; and
• /security-scan or npx -y ecc-agentshield scan --path . for agent-configuration checks.

A low-risk trial is one small, reversible task in a disposable or non-sensitive repository. Use one workflow, inspect the files it reads and writes, and compare the resulting diff with your normal process. Keep a record of defects, review time, test failures, and unnecessary changes. That gives you evidence about whether the workflow helps your codebase instead of treating the catalogue size as proof of value.

The security surface deserves explicit attention. ECC contains instructions, hooks, scripts, MCP configuration, and installer logic. Those files influence what an agent can do. Read them before enabling hooks in a repository with credentials, review any shell commands and network integrations, and run the documented AgentShield scan where it matches your setup. A permissive open-source license does not make every future script safe for every environment.

The project itself warns users to install from verified channels: the ECC GitHub repository, its documented npm packages, the GitHub App, the ecc@ecc plugin, and ecc.tools. Avoid third-party re-uploads and unofficial mirrors.

4. What the latest release adds

The latest listed release is ECC v2.2.0, published on August 28, 2026. Its headline change is one guided installer for Claude Code, Codex, and Kimi Code. It also adds native Antigravity 2.0 support and an experimental, opt-in Nasiko CLI lifecycle bridge.

The release notes are careful about that bridge: ECC can inspect, install, and remove one pinned Nasiko build, with content-addressed downloads and SHA-256 verification, but the bridge does not connect agents, start a server, enable telemetry, or operate Nasiko’s dashboard. That distinction is a good example of how to read an agent-tool release: separate what the installer manages from what the external product does.

The same release reports 68 agents, 286 skills, and 94 command shims, alongside improved uninstall ownership tracking, safer migrations, and more workflows. The release notes also describe a stricter package pipeline that checks the exact npm archive across Linux, macOS, and Windows before the latest tag moves.

The current default branch is not identical to the latest release tag. The public commit endpoint I retrieved listed September 3 commits including a patched @humanfs/node dependency and a Dependabot update for softprops/action-gh-release. Pin a release you have tested for a production workflow, and review the default branch separately when you want newer changes.

5. A practical first run

1. Choose the host. Start with Claude Code or Codex if you want the repository’s most explicit installation paths. Use another adapter only after checking its support notes.
2. Choose one installation method. Prefer the guided installer when you need a deliberate scope, target, or hook decision.
3. Use a test repository. Avoid giving a newly installed hook or script access to production credentials on the first run.
4. Pick one workflow. Try planning, TDD, or a security scan on a change with a clear acceptance test.
5. Review the complete result. Check the diff, tests, generated files, permissions, and any network or hook behaviour before adopting the workflow elsewhere.

ECC is a good fit when your agent work is becoming repetitive but inconsistent: the same planning questions, review checks, build repairs, and handoff steps appear across repositories. It is less useful as a giant bundle installed without a boundary. If you want a smaller collection of composable coding workflows, compare it with Matt Pocock’s skills repository. For a stricter “earn every new line” approach, see our guide to Ponytail.

Bottom line

ECC’s appeal is not simply its large catalogue. It packages engineering discipline around an AI coding agent: plan before editing, test the change, review from a fresh perspective, verify the result, and preserve what was learned. Its cross-harness adapters make that discipline portable, but the README is clear that portability does not mean identical feature support everywhere.

Install one path, start with one workflow, and keep normal code review and security review in the loop. That is the practical way to judge whether ECC improves your work.

Sources

• GitHub daily trending repositories
• affaan-m/ECC repository
• ECC README
• ECC v2.2.0 release notes
• Latest listed ECC commit
• ECC MIT license

Hero image: JOE developers / User:Joy, GPL, via Wikimedia Commons.

Frequently Asked Questions

What is affaan-m/ECC?

ECC is an open-source agent harness toolkit. Its repository packages skills, agents, commands, rules, hooks, memory, security scanning, and installation adapters for Claude Code, Codex, and other AI coding hosts.

How do I install ECC?

The repository documents `npx ecc-universal setup` for the guided Claude Code path and `npx ecc-universal install --guided` for a reviewed multi-harness setup. Claude Code and Codex also have native plugin commands. Choose one installation method per harness.

Does ECC work with Codex and Hermes Agent?

Yes, but feature parity depends on the host. ECC documents a native Codex marketplace/plugin path and a project-local minimal installer target for Hermes Agent. The README says Claude Code is its strongest-supported surface and other adapters can be capability-limited.

Is ECC free for commercial use?

The ECC repository is MIT licensed, which permits broad use, modification, distribution, and commercial use subject to the license notice and disclaimer. Review dependencies, scripts, hooks, and any third-party components before shipping a particular installation.