ECC: A Performance Harness for Claude Code Agents

TL;DR: affaan-m/ecc is an MIT-licensed agent harness performance system that bundles skills, instincts, memory, security, and a research-first workflow for Claude Code, Codex, and Opencode. JavaScript, at 260,331 stars.

What it is

affaan-m/ecc describes itself as “The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, and more.” The repo lives under the affaan-m GitHub account, the language is JavaScript, the license is MIT, and the repository has 260,331 stars and 38,964 forks. It is not archived.

Topics include ai-agents, anthropic, claude, claude-code, developer-tools, llm, mcp, and productivity. The homepage is https://ecc.tools.

What it does well

The star count alone tells you this is one of the more popular repositories in the Claude Code ecosystem. What it actually does is more interesting than the count.

• Skills. A curated skills layer that the host agent can pick up automatically, which is the same packaging pattern Claude Code itself uses.
• Instincts. Heuristic patterns the agent falls back on, rather than freshly reasoning from first principles every time.
• Memory. Persistent context the agent carries across sessions, which is the part most single-session coding-agent demos skip.
• Security. Guardrails around what the agent is allowed to read, edit, and execute.
• Research-first workflow. A development style where the agent does a research pass before writing code, which reduces the “vibes-only” failure mode that pure agent loops sometimes hit.
• Multi-host. The repo names Claude Code, Codex, and Opencode as supported hosts, which is broader coverage than most agent-harness projects.

How to try it

The install path follows the standard skills/sources pattern:

git clone https://github.com/affaan-m/ecc.git
# then load it from the supported agent host (Claude Code, Codex, Opencode)

The homepage at https://ecc.tools carries the latest install flags and host-by-host setup notes.

When it fits, and when it doesn’t

ECC fits developers who already use a coding agent as their main workflow and want a higher-quality skills, instincts, and memory layer on top. It is less attractive if you only fire up a coding agent occasionally — the value of the persistent memory and instincts compounds with daily use — or if you are barred from installing third-party skill bundles into your host agent.

260k stars and a homepage separate from the repo signal both an active user base and a productised surface.

Source and Accuracy Notes

• Project page: https://github.com/affaan-m/ecc
• Source repository: affaan-m/ecc (verified via GitHub API html_url)
• License: MIT (verified via GitHub API license.spdx_id)
• Stars at time of writing: 260,331
• Language: JavaScript
• Homepage: https://ecc.tools
• Topics: ai-agents, anthropic, claude, claude-code, developer-tools, llm, mcp, productivity