ECC is MIT-licensed and free for public repositories on the OSS installer. A Pro tier at $19 per active developer seat adds private-repo coverage and the GitHub App workflow; Enterprise adds governance and policy packs.

Which coding agents can ECC drive?

ECC 2.1.0 supports Claude Code, Codex, Cursor, OpenCode, Gemini, Zed, GitHub Copilot, and now Hermes, Kimi Code, and OpenClaw as install targets. One skills library plus the ecc.session.v1 adapter lets the same 281 skills run across every harness.

What does the Kimi harness target add?

Passing --target kimi installs ECC directly into Kimi Code, so Moonshot's coding harness picks up the 67-agent, 281-skill catalog and the AgentShield security scanner without a manual copy step.

How is Itô self-hosted compute different from the OpenAI or Anthropic APIs?

Itô is a marketplace of GPU capacity that ECC discovers with ecc ito find. Runs use open-source models against your own compute pool with a read-only, guarded handoff, so prompts and diffs stay off shared inference endpoints.

ECC · 2026-07-27 · major

ECC 2.1.0 — Plan Canvas turns terminal agent plans into a clickable browser review

ECC 2.1.0, the 234k-star cross-harness agent system for Claude Code, Cursor, and Codex, adds Plan Canvas (annotate agent plans in the browser), a Kimi Code harness target, and self-hosted GPUs via Itô.

affaan-m/ECC repository card — 234k-star agent harness for Claude Code, Codex, Cursor, and OpenCode

A 234k-star agent harness adds a browser plan-review canvas, a Kimi Code target, and self-hosted GPUs.

Key specs

GitHub stars	234k+
Agents	67
Skills	281

Quick facts

Maker	affaan-m (open source)
Version	v2.1.0
Released	2026-07-27
License	MIT
Install	npm install -g ecc-universal
Catalog	67 agents · 281 skills · 94 command shims
Harnesses	Claude Code, Codex, Cursor, OpenCode, Hermes, Kimi Code, OpenClaw

What is it?

ECC 2.1.0 ships Plan Canvas, a browser view of the plan an agent just wrote. You click any element or highlight text to attach numbered annotations, then chat with the agent from a side rail while the run continues in the terminal. The release also adds a Kimi Code install target, Hermes and OpenClaw targets, and self-hosted compute through Itô GPUs.

How does it work?

Plan Canvas is a loopback-only local server that renders the plan file with Mermaid diagrams, tables, and task lists; edits to the plan file live-reload the page. The canvas speaks JSON, so any harness can drive it. The new ecc.session.v1 adapter layer lets ECC install into Kimi Code the same way it installs into Claude Code, and `ecc ito find` discovers open GPU capacity through Itô with read-only, guarded handoff.

Why does it matter?

Reviewing an agent plan in raw terminal markdown is the friction step for most ECC users. Pointing at the paragraph you disagree with, instead of re-typing it, cuts a round trip out of every plan-then-execute loop. The new install targets mean the same skills library now runs from Kimi Code and Hermes, and Itô compute lets teams keep sensitive agent runs off shared inference endpoints.

Who is it for?

Developers running Claude Code, Cursor, Codex, or OpenCode day-to-day.

Frequently asked questions

Is ECC free?: ECC is MIT-licensed and free for public repositories on the OSS installer. A Pro tier at $19 per active developer seat adds private-repo coverage and the GitHub App workflow; Enterprise adds governance and policy packs.
Which coding agents can ECC drive?: ECC 2.1.0 supports Claude Code, Codex, Cursor, OpenCode, Gemini, Zed, GitHub Copilot, and now Hermes, Kimi Code, and OpenClaw as install targets. One skills library plus the ecc.session.v1 adapter lets the same 281 skills run across every harness.
What does the Kimi harness target add?: Passing --target kimi installs ECC directly into Kimi Code, so Moonshot's coding harness picks up the 67-agent, 281-skill catalog and the AgentShield security scanner without a manual copy step.
How is Itô self-hosted compute different from the OpenAI or Anthropic APIs?: Itô is a marketplace of GPU capacity that ECC discovers with ecc ito find. Runs use open-source models against your own compute pool with a read-only, guarded handoff, so prompts and diffs stay off shared inference endpoints.

Try it

npm install -g ecc-universal