claude/for
Search
Subscribe
Claude CodeUpdated

Repo radar: ECC: community agent harness for Claude Code

ECC is a community, non-Anthropic agent harness for Claude Code with skills, hooks, memory, and a research-first workflow.

ECC (affaan-m/ECC) is a community, non-Anthropic agent harness aimed first at Claude Code. Per its README, it packages skills, instincts, memory, security scanning, and a research-first loop so the model follows plan → test → implement → review → verify → remember → improve instead of one-off prompts. Homepage: ecc.tools. License: MIT. README release: 2.2.2 (2026-08-31). Checked 2026-09-20.

TL;DR. Use ECC if you already run Claude Code 2.1+ and want a reusable plan, TDD, review, and memory catalog installed once. Pick one install path. Use official channels only. Stars show popularity, not product quality or security. ECC is not an Anthropic product or endorsement.

When to use / when not

Use when you already work in Claude Code, accept a large third-party catalog with selective profiles, and want an engineering loop installed once rather than rebuilt in every chat.

Skip when day-one Claude Code is enough. Start with the official marketplace and local skills: Add Anthropic's skills marketplace and Sync Claude.ai skills and plugins into Claude Code. ECC does not replace Anthropic docs, the Claude Code changelog, or anthropics/skills.

Install: one path

The README requires Node.js 18+ for the universal package. Do not stack a plugin install on top of a full guided install.

npx ecc-universal@2.2.2 setup

The alternate README path is for use instead of the guided path, not with it: /plugin marketplace add https://github.com/affaan-m/ECC, then /plugin install ecc@ecc. That path needs Git and Claude Code 2.1+.

The README says to install only from official channels: this GitHub repository, the named npm packages (ecc-universal, ecc-agentshield), the GitHub App, plugin id ecc@ecc, and ecc.tools. Treat third-party mirrors as a malware risk.

After installation, sanity-check skills and plugins using the usual Claude Code workflow: Skill doctor and Plugin eval. Nearby release context: Claude Code 2.1.273 to 2.1.278.

What's inside: README claims

The counts below are README claims only, not an independent recount of the repository tree [[GAP]]:

SurfaceREADME claim
Agents68
Skills292
Command shims94
Plushooks, memory, selective rules, AgentShield scanning

The README positions ECC as best with Claude Code. It also describes a Codex native plugin and capability-limited adapters for other harnesses. Read its support matrix before assuming parity: Cursor and OpenCode are beta; Copilot is instruction-only; other integrations are experimental. Windows native limitations cited in README issues were not re-tested here [[GAP]].

The README describes OSS as MIT and ECC Pro as a hosted GitHub App for private repositories. Its pricing language, including a claimed $19 per seat per month, is README marketing only and was not confirmed outside that block [[GAP]].

Limits

  • Not Anthropic. Listing this repository is not an endorsement.
  • Stars are not quality. GitHub showed 263086 stars on 2026-09-20. Treat that as popularity and discourse heat, not a quality or security signal [[GAP]].
  • Context cost. A full catalog can bloat context. The README recommends selective or minimal profiles and watching MCP and tool count.
  • Parity. "Works everywhere" does not mean feature parity across Cursor, OpenCode, Copilot, and other harnesses.
  • Unverified this run: npm digests and AgentShield audit depth [[GAP]].

Sources (checked 2026-09-20)

The claims above are dated field notes. Re-check the README, package versions, and provider support before installing into a production repository.

One Claude move in your inbox, every Sunday

Four minutes, tested on a real job, then back to your weekend. Free.